AMD has finally released its Radeon HD 7950 graphics card to the world, the second tier of the manufacturer's Southern Islands/AMD HD 7000 series cards.

Was it worth the wait? Well you will be pleased to know that yes it was…and then some.

This new graphics card is based on the same 28nm Graphics Core Next GPU as the AMD Radeon HD 7970, code-named Tahiti.

This Tahiti Pro is ever so slightly cut down compared to the HD 7970's Tahiti XT, but the disparity in performance is mostly down to the lower core clockspeed.

The AMD Radeon HD 7950 is clocked at a fairly conservative 800MHz compared to the HD 7970's 925MHz.

We say conservative as the Tahiti Pro GPU, from all the cards we've seen, is more than capable of topping the 1GHz mark. So it's a bit of a shame that we haven't seen a HD 7000 series card hitting the 1GHz core clock speed out of the box.

Though there are rumours a certain Taiwanese graphics card manufacturer may be releasing an overclocked Radeon HD 7950 with that very rating.

We've been playing with a couple of the cards for a while now and of everything it's the overclocking prowess of the Radeon HD 7950 that makes it such an outstanding card.

The fact that it also bests the Nvidia GeForce GTX 580 in pretty much all the benchmarks, and ably keeps pace with the Radeon HD 7970 only adds to that.

And does it all for less cash than either.

You can check out our AMD Radeon HD 7950 review and the factory-overclocked Sapphire Radeon HD 7950 OverClock edition review on the TechRadar Components Channel right now.

Suffice to say that if we were spending serious money on a graphics card right now it would be on an AMD Radeon HD 7950.

Nvidia is going to have to work hard to produce a card for the same price that out does it on performance.

AMD HD3D technology: what's it all about?

Stereoscopic 3D might not be to everyone's tastes, but it's certainly a lucrative business.

For the past few years Nvidia has enjoyed a virtual stereo 3D monopoly on the PC with its 3D Vision technology, but AMD has belatedly entered the ring with its own technology called HD3D.

For those who have heavily invested in Nvidia's technology it might be too little too late, but AMD's new offering introduces a number of innovations that make it worth considering – even if you think stereo 3D is just an expensive way to get a headache.

• Read our Nvidia 3D Vision review

The biggest attraction of HD3D is AMD's commitment to making it an open platform that supports many different standards. Unlike Nvidia's approach, where you're tied to Nvidia 3D Vision-certified hardware, HD3D's open ecosystem should allow for a more diverse choice.

It's pretty encouraging that AMD seems to have identified one of consumer stereoscopic 3D's biggest problems, and the reason why so many of us have so far held back from embracing stereo 3D: the vast and confusing array of incompatible 3D standards. If AMD's mission with HD3D is to allow us to buy 3D hardware from various different manufacturers, without us having to worry whether it will all work together, that can only be a good thing for us consumers – and might lead to more of us setting up our rigs for 3D.

How it works

When it comes to three-dimensional gaming, the bulk of the stereo horsepower comes from the HD3D driver. The driver uses a quad-buffer to produce stereo 3D. Usually with non-stereo 3D graphics (monoscopic) the driver uses double-buffering.

This is where the GPU renders content to one specific place in memory, known as a buffer. At the same time, a second buffer is used to deliver the display output of the GPU to a monitor or other display. Quad-buffering essentially doubles this, producing two images – one for the left eye and one for the right – for every frame generated.

AMD's Direct3D engineering team produced an API that supports OpenGL and DirectX 9, 10 and 11. AMD's open approach allows middleware partners, such as DDD and iZ3D to convert games from monoscopic to stereoscopic. This takes some of the pressure off AMD to ensure games are compatible with HD3D and has led to an already impressively long list of compatible games.

Launching into HD3D

If you want to set up HD3D then the first thing you need to do is make sure that you've got a supported Radeon graphics card. The ATI Radeon HD 5000 Series using Catalyst 10.10 or later and AMD Radeon HD 6000 series or above are compatible.

You can also use HD3D if you have an HP Envy 17 3D with ATI Mobility Radeon HD 5850, an MSI Wind Top AE2420 All-in-One with ATI Mobility Radeon HD 5730, or a Lenovo IdeaPad y560d with ATI Mobility Radeon HD 5730.

You'll also need a 3D-capable display device, like a 3D monitor, TV or projector. There are plenty of supported devices – click here for a full list. If the display device is a 3D TV or a 3D projector then it needs to support HDMI 1.4a, and you'll need an HDMI 1.4a cable.

Unless you can afford a glasses-free 3D display, you'll need the chunky eyewear too. You'll also want DDD's TriDef 3D for the AMD HD3D driver, or the iZ3D 3D driver for 3D gaming.

Thanks to AMD's partnership with the middleware driver makers, you can get 50 per cent off either driver if you have HD3D compatible hardware. Make sure you have the latest AMD Catalyst 10.10 software or later from AMD's website. Install either the DDD TriDef 3D or iZ3D drivers and select either 'AMD – AMD HD3D Technology (HDMI 1.4a)' or '120Hz 3D Devices' respectively.

You'll then need to set your desktop resolution to 1,920 x 1,080, 24Hz or 1,280 x 720, 60Hz. Using either TriDef 3D or iZ3D you need to select the game profile and point the software to the game's executable file, and then launch the game itself from inside the 3D software.

AMD has outsourced a lot of the work involved in creating compatible games to the third party middleware companies iZ3D and DDD. Headline games include Call of Duty: Modern Warfare 2, Mass Effect 2 and World of Warcraft, and are included in a growing list of compatible games.

You can be pretty confident that the biggest new releases will be compatible, but for a full list of compatible games go to iz3d.com/games and the TriDef website.

It's not all down to third parties though. AMD is working directly with games developers on native support for HD3D. This year AMD worked closely with Eidos to support native stereo 3D in Deus Ex: Human Revolution. This meant that Eidos rendered images for both the left eye and the right eye in game, and was able to utilise HD3D's quad-buffer without the need for middleware.

Hopefully as the HD3D standard gains popularity, more games will include native support as they are released. So how does AMD's plucky little David compare to Nvidia's Goliath?

In terms of performance and raw power, Nvidia's head start in stereo 3D certainly gives 3D Vision the edge. HD3D's reliance on the HDMI 1.4a specification leaves it with a maximum TMDS throughput of 10.2Gb/s, allowing for 1080p gaming at 24 frames per second for each eye, or 720p at 60 frames per second.

This is a lot lower than 3D Vision Surround's dual-link DVI connection capable of resolutions up to 5,760 x 1,080 over dual monitors. AMD hopes that as more new monitors begin to support DisplayPort 1.2, HD3D will be able to get around HDMI 1.4a's limitations with a bandwidth of 17.28Gb/s – enough for 1080p at 60 frames per second per eye.

HD3D vs 3D Vision

AMD's embrace of open standards really sets the HD3D apart from Nvidia's 3D Vision. Not being tied to specific hardware and standard restrictions gives us far greater scope to build a 3D-capable rig that meets our needs.

It can also be more cost effective – not only can we shop around for the best components at the best prices, we can also avoid some of the hidden costs associated with Nvidia 3D Vision. You don't need a propriety 3DTV Play driver to play 3D Blu-rays, and you don't have to rely on expensive USB emitters.

You do need to pay for third party middleware drivers, though. Still, there is enough potential in HD3D to get us excited about its future, and to cause Nvidia some concern over 3D Vision's future.

AMD has just pulled the tarp off its new AMD Radeon HD 7970 Southern Islands graphics card and we can finally unveil our thoughts on its brand new card and architecture.

It's the fastest single-GPU graphics card and also a graphics card of firsts.

The HD 7970 is the first GPU manufactured with 28nm transistors, is the first to support DirectX 11.1 (when that finally emerges) and the first to support PCI Express 3.0.

The first of these firsts is the biggy though and has allowed AMD to jam 4.3 billion of those 28nm transistors into the Tahiti XT graphics processor at its heart.

It also represents a new direction in graphics architecture for AMD, more in line with what Nvidia has been doing for years.

AMD is adamant it is no copycat though.

"From a purely technical point of view, the architecture that we've moved to is similar," says Mike Mantor, AMD Senior Fellow Architect. "On the GPU side though we've been investigating this scalar architecture for some time. We had good ideas how we'd go about it, what our gains would be and the cost of actually implementing it."

As well as being the fastest at stock clocks it's also one hell of a serious overclockers graphics card. We managed to push ours from 925MHz up to 1,125MHz.

The AMD Radeon HD 7970 is also rather expensive.

We don't have final UK prices yet but our best guess is around £450, with future overclocked versions coming in considerably more.

That's good news for Nvidia as it makes the impressive GTX 580 a far more reasonable proposition at just £365.

And talking of Nvidia, it's own new architecture, the Kepler-based graphics cards, will be available early Spring 2012. According to Nvidia insiders it's feeling quietly confident about its ability to compete with the Radeon HD 7970.

We've got a full, in-depth review of the AMD Radeon HD 7970 here.

Nvidia is expecting strong growth in the next year, pointing to its new Tegra 3 platform as it insists that its strategy is coming into its own.

The chip giant believes that our increased focus on not just mobile but also on what we expect from graphics is behind a modest rise in revenue in its latest financial results, with brighter times ahead.

"Nvidia's strategy is coming into its own, as the world becomes increasingly visual and mobile," said Jen-Hsun Huang, president and chief executive of Nvidia.

Accelerated

"Our GPU business accelerated in the third quarter, driven by strong demand from gamers and the professional market. And our mobile business benefited from new devices coming onto the market," he added.

"With Tegra 3 phone wins well ahead of Tegra 2's pace, we're expecting strong growth in the year ahead."

The company recorded a 4.9 per cent increase in revenue to $1.07 billion (c£672m) over the last quarter and a 26.3 per cent rise year on year.

Nvidia took the opportunity to reflect on some of its recent announcements as it looks ahead – including the launch of Tegra 3 in the Asus Eee Pad Transformer Prime, the addition of three more Tegra phones, 13 more Tegra-toting tablets (which takes the total to 23) and the news that the Titan supercomputer will use 18,000 Tesla processors.

AMD has confirmed that its upcoming 7970 and 7950 graphics cards will come in dual-chip varieties.

The doubled-up variant, which could end up branded as the Radeon HD 7990, will be codenamed New Zealand in keeping with the AMD's island-based nomenclature.

Based on a 28nm architecture, the 7000 series is said to be AMD's fastest chip so far, beating its current top-of-the-range 6990.

Singularity

The singular cards are scheduled to be released in Q1 2012, with the dual-chipped card following shortly afterwards.

Nvidia is also working on a 28nm dual Kepler card, which we can expect to see around the same time.

There's no word on pricing yet, but it'll probably join the 6990 in the megabucks £500+ ($796+) bracket.

Although we were impressed by the 6990's performance, the extreme pricing was a big minus point. AMD could make the 7000 series more affordable - but we doubt it.

The future of PC graphics

What's next for graphics? Why, Graphics Core Next, of course. Thanks, AMD, for that nicely pallindromic way to start off a feature.

And also for talking about the successor to the current generation of Radeon graphics cards, which is due sometime next year.

The unveiling of GCN took place at June's Fusion Developer Summit. It's the first complete architectural overhaul of GPU technology it's risked since the launch of Vista.

That also, incidentally, makes it the first totally new graphics card design for AMD that isn't based on work started by ATI before it was purchased.

Vista, and specifically DirectX 10, called for graphics cards to support a fully programmable shader pipeline.

That meant doing away with traditional bits of circuitry that dealt with specific elements of graphics processing – like pixel shaders and vertex shaders – and replacing them with something more flexible that could do it all: the unified shader (see "Why are shaders unifi ed?", next page).

Schism

During the birth of DX10 class graphics, there was something of a schism between Nvidia and AMD.

To simplify: the former opted for an interpretation of unified shader theory in its G80 GeForce chips that was quite flexible. Place a few hundred very simple processors in a large array, and send them one calculation (or, in some circumstances, two) a piece to work on until all the work is done.

It's a method that creates a bit of a nightmare for the set-up engine, but it's very flexible and for well written code that takes advantage of the way processors are bunched together on the board, dynamite.

In designing the G80 and its successors, Nvidia had its eye on applications beyond graphics. Developers could create GPGPU applications for GeForce cards written in C and more recently C++.

AMD/ATI, meanwhile, focused on the traditional requirements for a graphics card. Its unified shaders worked by combining operations into 'Very long instruction words' (VLIW) and sending them off to be processed in batches.

The basic unit in an early Nvidia DX10 card was a single 'scalar' processor, arranged in batches of 16 for parallel processing.

Inside an AMD one, it was a four way 'vector' processor and with a fifth one for special functions. Hence one name for the Radeon architecture: VLIW5. While the set-up sounds horrendous, it was actually designed to be more efficient.

The important point being that a pixel colour is defined by mixing red, green, blue and alpha (transparency) channels. So the R600 processor – which was the basis of the HD2xxx and HD3xxx series of cards – was designed to be incredibly efficient at working out those four values over and over again.

Sadly, those early R600 cards weren't great, but with time and tweaking AMD made the design work, and work well.

The HD4xxx, HD5xxx and HD6xxx cards were superlative, putting out better performance and requiring less power than Nvidia peers. Often cheaper too. But despite refinements over the last four years, the current generation of GeForce and Radeon chips are still recognisable as part of the same families as those first G80 and R600.

There have been changes to the memory interface (goodbye power hungry Radeon ring bus) and vast increases to the number of execution cores (1,536 on a single Radeon HD6970 compared to 320 on an HD2900XT), but the major change over time has been separating out the special functions unit from the processor cores.

Graphics Core Next, however, is a completely new design. According to AMD, its existing architecture is no longer the most efficient for the tasks that graphics cards are called to do.

New approach

FUTURE GRAPHICS: VLIW5 has four vector processing units: one each for R, G, B and alpha

Proportionally, the number of routines for physics and geometry being run on the graphics card has increased dramatically in a typical piece of game code, calling for a more flexible processor design than one geared up primarily for colouring in pixels.

As a result, the VLIW design is being abandoned in favour of one that can be programmed in C and C++.

The basic unit of GCN is a 16 wide array of execution units arranged for SIMD (single instruction, multiple data) operations. If all that sounds familiar to G80 and on, it's because it is.

Cynically, this could be seen as a tacit acknowledgement that Nvidia had it right all along, and there's no doubt that AMD is looking at GPGPU applications for its next generation of chips. But there's more to it than that.

Inside GCN, these SIMD processors are batched together in groups of four to create a 'compute unit' or CU. They are, functionally, still fourway vector units (perfect for RGBA instructions) but are also coupled to a scalar processor for one off calculations that can't be completed efficiently on the SIMD units.

Each CU has all the circuitry it needs to be virtually autonomous, too, with an L1 cache, Instruction Fetch Arbitration controller, Branch & MSG unit and so on.

There's more than the CU to GCN, though. The new architecture also supports x86 virtual memory spaces, meaning large datasets – like the megatextures id Software is employing for Rage – can be addressed when they're partially resident outside of the on-board memory.

And while it's not – as other observers have pointed out – an out-of-order processor, it is capable of using its transistors very efficiently by working on multiple threads simultaneously and switching between them if one is paused and waiting for a set of values to be returned. In other words, it's an enormously versatile chip.

After an early preview of the design, some have noted certain similarities with Intel's defunct Larrabee concepts and also with the Atom and ARM-8 chips, except much more geared up for parallel processing.

INVENTIVE NAMING: GCN will still work with RGBA data, but boasts greater flexibility

"Graphics is still our primary focus," said AMD's Eric Demers during his keynote presentation on GCN, "But we are making significant optimisations for compute... What is compute and what is graphics is blurred."

The big question now is whether or not AMD can make this ambitious chip work. Its first VLIW5 chips were a disappointment, running hotter and slower than expected. So were Nvidia's first generation Fermi-based GPUs.

Will GCN nail it in one? We've got a while to wait to find out. The first chips based on GCN are codenamed Northern Islands and will probably be officially branded as Radeon HD7xxx. They were originally planned for this year, but aren't expected now until 2012.

Best graphics card under £150: 8 reviewed

Graphics cards are so important - theyproducing the biggest performance variable in a desktop PC.

We've capped the pricing at £150.If you only upgrade one component in your PC, make it your graphics card. It's going to have the most dramatic effect to your video playback, gaming frame rates and graphical quality.

Before we delve into the group test, there are a few things to mention. If you're sitting on a powerful last-gen GPU, you're faced with quandary: buy another of my existing card for a cheap SLI/CrossFire setup, or start again with a new single GPU.

In general, we'd recommend the newer single card, but it does depend on what you've got to work with in your existing rig.

Nvidia's GTX 280s have retained their value sickeningly well, yet can't hack DirectX 11 tech like tessellation, which makes them a horrible choice for SLI. More recent AMD Radeon HD 5 series cards might be a more viable option, though.

There's another reason to drop a little extra dollar on graphics: integrated graphics solutions are getting better. Intel's HD Graphics 3000 engine is pretty good, and AMD's Llano APUs are fantastic. If you buy a low-end card like an HD 4 series card, you're not really getting anything better than integrated graphics than may already be at your fingertips. So which card should you back? Read on…

Cards on test

Sapphire HD 5670 Ultimate - £75
Sapphire HD 6670 Ultimate - £80
Palit GTS 450 - £81
XFX HD 5770 - £81
Sapphire HD 6770 VaporX - £82
EVGA GTX 550 Ti - £98
XFX HD 6850 - £118
Gigabyte GTX 560 - £148

Best graphics card under £150: the test

Sapphire HD 5670 Ultimate - £75

When you're talking about budget gaming graphics cards then you really can't spend less than £75 if you want to get anywhere near playable frame rates. At the bottom of this particular pile then is AMD's Radeon HD 5670, here represented by Sapphire's passively-cooled Ultimate edition.

The Radeon HD 5xxx series of cards are getting a bit long in the tooth these days; this particular GPU was introduced way back in the small hours of January 2010. It's still almost a relevant card these days though, offering close to playable gaming speeds in a tiny, no nonsense package.

The passive cooler sitting atop this Ultimate edition of the HD 5670 makes it a perfect match for the living room machine where silence is more preferable to out and out performance.

This is a good thing as the straight line performance of the HD 5670 is looking a little out of date now, but we'll come to that in a second.

Read the Sapphire HD 5670 Ultimate review

Sapphire HD 6670 Ultimate - £80

As the naming might suggest this HD 6670 is the direct replacement for the HD 5670 we've just seen. The Sapphire Ultimate version again comes passively-cooled and is only a little more expensive than its older brother. So it's essentially more of the same then? Almost.

This updated GPU comes with an extra 80 shaders, a smattering of extra texture units and a small hike in core clock speeds. Other than that it really is pretty much business as usual between the HD 5670 and the HD 6670.

All that GPU tweakery means you do get some extra performance out of this newer card, though admittedly not much. At most you're looking at 3fps faster across our benchmarking suite, though that can sometimes make the difference between choppy and playable gaming.

That will ring especially true once you knock some of the advanced graphics settings down a notch or two, especially the resource hog that is anti-aliasing. Again you're looking at fairly playable frame rates in modern games at lower settings, and good performance in older titles like Far Cry 2 on Max settings.

Read the Sapphire HD 6670 Ultimate review

GeForce GTS 450 - £81

Now we're getting into the serious gamer's graphics card territory, though we're still not hitting the big numbers when it comes to pricing. Nvidia doesn't have a great line-up in the budget segment of cards and anything lower than this here GTS 450 is not really worth a look for those with any passing interest in frame rates.

This venerable card does have itself some gaming chops to offer, mind, and for the sum of £81 it's a tough card to argue against. The version we have benchmarked in this test is the vanilla reference version directly from Nvidia itself. That means it has no extras, no funky third-party cooling solution and definitely no factory-overclock.

We've taken the price from the cheapest GTS 450 we've found available and that's the version from Palit. It does come with a slightly different cooler, but the GPU itself is in reference trim.

Read the Nvidia GTS 450 review

XFX HD 5770 - £81

We love this little card. Here at PC Format we've been big fans of the HD 5770 since we first clapped eyes on it way back in the mists of time, circa November 2009. That opinion has only improved with age, like that of a fine rum-cask whisky.

Thanks to AMD's habit of monthly driver releases we've seen the out-right performance of the HD 5770 improve hugely and at the same time we've seen the price going in completely the opposite direction.

For example, at launch the card was £130 and ran Far Cry 2 at 33fps at 1,680 x 1,050. Now you can pick up this special version of the happy little GPU for £81 and it'll run Far Cry 2 at 53fps. That's a 61 per cent increase in performance in the same game, and at the same time the HD 5770 has seen a 38 per cent decrease in price. That's the sort of maths that we like here at PC Format.

Compared with the GTS 450, which has itself dropped 33 per cent in price but has gained little in subsequent driver releases, the HD 5770 comes out tops. In performance terms there's not a great deal in it, but the AMD card does edge it, and has a few other pleasing tricks up its sleeve too.

Read the full XFX HD 5770 review

Sapphire HD 6770 VaporX - £82

We opted to include the recent 6-series update to the HD 5770 simply to highlight a few things. As much as we've railed against Nvidia in recent times for confusingly rebranding graphics cards and GPUs with newer series numbers without changing silicon, AMD is not averse to the practice either.

The simple fact is there is almost no difference between the HD 6770 and the old school HD 5770. Nvidia has done this before, re-badging the excellent 8800GTX multiple times, each vaguely dulling that excellence as it emerged. First it became the 9800GTX and then later the GTS 250, with very little modification.

Almost the same has happened here with the HD 6770 coming out with the exact same silicon onboard, in fact AMD almost seemed a little sheepish about the launch of this card and didn't really seem to make any noise about it at all.

Despite the 6-series naming update for the card though vendors don't seem to be asking for any more than the standard new HD 5770s still in the marketplace.

Read the Sapphire HD 6770 VaporX review

EVGA GTX 550 Ti SC - £98

Now we're starting to hit the real price jump, the GTX 550 Ti is nearly £20 more than the last GPU. By rights in the graphics world that should actually mean quite a lot with cards packed together like yellowfin tuna in Japanese killing farms. Tip out your wallet right now and you can almost guarantee there's a graphics card available for that exact amount of cash. To the penny.

The step up from the AMD HD 5770/6770 though doesn't yield the sort of performance returns you might, quite reasonably, expect from the facts and figures on the specs sheet. This here is the generational refresh of the ageing GTS 450, as noted by the GF 116 GPU as opposed to the GF 106 chip.

Compared with its older brother the GTX 550 Ti does represent progress, with performance improving fairly significantly along with the improvements in silicon, especially in this EVGA Superclocked version with its much higher clock speed.

Read the EVGA GTX 550 Ti SC review

XFX HD 6850 - £118

To be honest we were rather unforgiving of the HD 6850, at launch it was pricing itself almost out of the market. It was going toe-to-toe with Nvidia's 1GB GTX 460 which, at the time, just about had it pipped in performance terms. It was also a little pricier than the GTX 460, coming in around the £160 mark.

Again though time has been kind to the HD 6850. The price has dropped a huge amount, indeed AMD recently announced a further price-drop bringing the card down to less than £120, which for a spec like this is a serious bargain. AMD's constant driver updates too have meant that performance has increased over time as well.

The Barts Pro GPU core at the heart of the HD 6850 is a reworking of the Cypress Pro that made the HD 5850 such an impressive card back in the day. It doesn't have the huge number of Radeon Cores the HD 5850 had, but still maintains the ROPs count of 32.

It's not, however, quite the performance powerhouse the older card still is, but then you won't find that GPU for the same price as the HD 6850.

Read the XFX HD 6850 review

Gigabyte GTX 560 OC Edition - £148

When the GTX 560 Ti first arrived Nvidia was quick to point out it wasn't a like-for-like replacement of the GTX 460, as the GTX 550 and GTX 570 had been for their respective brethren. The GTX 460 then still had life left in it.

Realistically though that life was snuffed out with the non-Ti version of the GTX 560. With the same core configuration as the out-going 1GB GTX 460: 336 CUDA Cores, 56 texture units and 32 ROPs - and the GTX 5-series' transistor-level enhancements over the GTX 4-series GPUs, this was always going to be putting the old classic out to pasture.

Indeed where once the 1GB version of the GTX 460 cold be picked up for around £120, now you're lucky to be able to find it for the £150-odd you can pick up this overclocked version of the GTX 560 for. And with a decent overclock this card can perform graphical wonders.

Sadly Gigabyte has only seen fit to provide 20MHz on top of the 810MHz core clock of the standard GTX 560, but with the impressive Windforce cooling array sitting atop it we managed to hit well over 900MHz without the card breaking a sweat.

Read the Gigabyte GTX 560 OC Edition review

Best graphics card under £150: Performance

TechRadar Labs

The XFX Radeon HD 6850 is not the absolute fastest card in our test, but it delivers the best all-round package of graphical grunt, value and efficiency.

For the money the sort of gaming performance you can squeeze out of the card is downright impressive. Any card that can offer double figure frame rates in Heaven 2.5 at 2,560 x 1,600 has to be taken seriously.

Below that the other XFX card, the Radeon HD 5770, is another excellent budget offering. At just over £80 for this single-slot version you can hit very playable figures in just about all the games in our benchmarking suite at the standard resolution of 1,680 x 1,050. Indeed the 1080p benchmark of Shogun 2 shows that it will even give that resolution a good go too.

On the Nvidia front though things don't look so good in the budget segment. With the GTX 460 starting to disappear, and not at a good price, the next generation cannot take up the slack.

Best graphics card under £150: The winner is...

When you're talking about budget gaming graphics cards you really can't be looking any lower than £75 for performance in-game that isn't going to make you wince every time you boot up a title and have to play it in retro 680 x 480 vision to get a decent frame rate. That may sound a trifle snobby, but if you just want a graphics card that will let you watch HD movies and play Half-Life then you're not talking about a gaming card.

Frame rate is king and these days the 20/22-inch native resolution of 1,680 x 1,050 is what you ought to be aiming to hit with a good budget graphics card. And you want to be able to get somewhere near the 30fps mark in your favourite titles to give you a decent gaming experience.

Bargain performer

At the low-end then you really cannot look past the increasingly brilliant XFX HD 5770. Every time we come to check out this venerable card it seems to just get better and better.

At launch the HD 5770 was a decent card, giving acceptable frame rates for the cash. Now, thanks to huge price cuts and performance driver updates it's a quite incredible value GPU. At just £81 (we have seen it for as low as £75, but it's currently out of stock at Scan) it's the same price as the HD 6670 and frankly blows it out of the water in performance terms.

It's not just the straight-line performance of the XFX HD 5770 that impresses us either. The fact it has been condensed down into a single-slot form factor is doubly impressive, despite the fact it does mean it gets a little hot under load. We haven't seen a decent single-slot card for a long time before and not since. There has been rumours of HD 6850s appearing in Hong Kong in this form factor but there's been nothing on these shores.

The lower-end Nvidia cards though haven't performed so well. Since owning the budget market with the GTX 460 for so long AMD has really clawed back ground with its line-up and price cuts. The old GTS 450 is a power hungry, under-performing card these days and the same can be said for its generational update, the GTX 550 Ti.

That card though has the added problem of being around £100, which just prices itself out of the market entirely.

Budget gamer

For just another £20 on top of that you can now pick up the suddenly bargainous HD 6850. The card we tested, again from XFX puts the slight overheating issues into the past.

The HD 6850 then is simply a quite sterling gaming card for less than £120. It's a card that wont disappoint at the standard 1,680 x 1,050 resolutions these budget cards are realistically going to be pointed at.

The HD 6850 is the winner then, but it was definitely a close run thing as it was up against the Gigabyte GTX 560 OC Edition. The £150 Gigabyte Nvidia card is a factory overclocked beast that will clock up further than its conservative out-of-the-box speed might suggest.

Already faster than the rest of the cards, hitting 900MHz will push this card even further ahead, and if you're sitting on a full HD, 1,920 x 1,080 monitor and only have a limited graphics budget, this is a steal at £150. But in the value stakes the XFX HD 6850 still holds sway at £120, if only by the hair on a gnats chin.

PC gaming champ Nvidia has unveiled its latest 3D Vision system, which features redesigned glasses as well as Nvidia's own LightBoost technology.

The 3D Vision 2 glasses feature 20 per cent larger lenses than the first generation 3D Vision goggles, which results in both a larger viewing area and increased external light blocking.

They're also made of soft composite materials, so putting on gaming headphones no longer makes it feel like you've got your head in a vice.

Head in a vice

The glasses are backwards-compatible with existing 3D Vision systems, such as laptops and monitors.

Nvidia seems to have solved the 3D darkness problem with its LightBoost technology, which it reckons delivers two times brighter 3D images and improved colour quality.

"NVIDIA's engineering team has made incredible enhancements in 3D on PCs, creating a breathtaking gaming experience that's better than the best Hollywood 3D movie," said Phil Eisler, general manager of 3D Vision at Nvidia.

Whack-a-mouse

There'll be no more fumbling to find your mouse and keyboard, too - Nvidia says that it will make "gaming keyboards and mice more visible".

To compliment the glasses, Asus has released the first LightBoost-certified desktop display, the VG278H.

It's a 27-inch full HD monitor featuring 3D HDMI 1.4 input, as well as Asus' own Trace-Free II technology for smooth visuals.

Toshiba's jumped on the good ship LightBoost as well, with its Satellite P770/P775, Dynabook Satellite T572, Dynabook T572, and Qosmio X770/X775 notebooks being compatible with the technology.

Asus' VG278H monitor will cost $699 (£444) and is expected to be available at the end of October.

Nvidia's 3D Vision 2 kit - with two glasses and a wireless USB IR emitter - will retail at $149 (£85) with additional specs costing $99 (£63) each.

Component company MSI has detailed its latest exclusive graphics card fan technology - it packs a number of new features, the most interesting of which is its dust removal technology.

On booting, the fan runs backwards for thirty seconds, sucking out any debris that may clog the heatsink.

MSI reckons dust may be to blame for many PC crashes, and that it could increase operating temperatures by as much as 15 degrees celsius.

But while the new technology may remove dust from the card itself, we can't help but wonder if it's simply sucked up by the card again when it starts running normally, or if it just roams the inside of your PC case looking for other heatsinks to inhabit.

Get into the groove

Other advances in MSI's new fans include the Propeller Blade, which uses arched edges, grooves and a gloss coating to provide a 20 percent increase in air flow.

In addition, the Smart Temp Sensor Tech makes the fans change colour from blue to white when the GPU is running at over 45 degrees celsius, which is great if you like looking at spinning objects more than you like watching Nazis' heads explode in-game.

The Smart Temp Sensor and Dust Removal Technology will be available on MSI's upcoming Nvidia N580GTX Lightning Xtreme Edition, and the Propellor Blade is available on MSI's latest Nvidia and AMD products.

How GPUs are made

These really are, as Paul Simon sang, the days of miracles and wonders. It seems almost impossible to believe that engineers have now managed to design and build a machine whose components measure as little as 40nm across. That's just one-thousandth the diameter of a human hair.

Yet we're making such devices right now. They're called general processing units, and they're starting to challenge the central processing unit for its long-held title of the most magical piece of engineering found in a computer.

What threat is the CPU facing? Consider a top-of-the-range desktop processor like the quad-core Sandy Bridge variant of Intel's Core i7. It boasts just short of a billion transistors. A leading edge general processing unit like AMD's Cayman, as used in the Radeon HD 6970 graphics card, clocks up a massive 2.64 billion transistors.

Given that it boasts no fewer than 1,536 shader processors, 24 SIMD (single instruction multiple data) engines and 32 ROPs (raster operator units), this perhaps isn't surprising. This is the story of how AMD GPUs are made - how an idea becomes silicon using some of the most advanced and intricate engineering modes and mechanisms. Read on as we delve into the real days of miracle and wonder.

1. The high level design

Designing a general processing unit doesn't start with any thought of transistors or copper tracks, but with something called the product requirement specification, or PRS - a prioritised definition of all the features the new chip must have. It might not sound wildly exciting, but the PRS acts as the checklist throughout the whole design process.

Given that design is a very costly exercise, in terms of time and money, it's vitally important that the PRS provides an adequate answer to the question: 'What exactly is it that we're trying to build here?'

Typically it will take six months to complete the PRS. Thousands of engineers, including architects, hardware designers, board designers, validation engineers, software engineers and firmware/BIOS engineers will be involved, as will representatives from product management, technology management and developer relations.

The document takes the form of a database and could include over 1,000 features, each of which could be anything from an odd sentence to a 100-page specification.

Another output from the high level design - one that most technically savvy PC users will be familiar with - is a block diagram. Although it bears no resemblance to how the elements of the GPU will be arranged on the chip, it includes each of the major functional blocks and shows how signals pass between them.

2. Floorplan and netlist

Teams of engineers now set to work on two distinct areas of the design. First, the floorplan must be defined. This is a physical representation that will take account of how large each block is expected to be and where it should be positioned relative to other blocks.

Here, account is taken of how many signals pass between the blocks with the aim of reducing the lengths of the pathways. Meanwhile, other engineers work on the component level design of each of the blocks.

However, this is nothing like electronic circuit design as envisaged by the layperson. Instead of a circuit diagram, the design is created in a hardware description language like VDHL. If you're interested, the compound acronym stands for very-high-speed integrated circuits (that's what the 'V' stands for) hardware description language.

Looking much like a programming language, this way of generating circuits provides many of the benefits on offer to the software engineer. Most importantly circuits can be defined hierarchically so, for example, having defined a logic OR gate from individual transistors, this can be used in the definition of a more complicated block like a one-bit adder.

In the same way, increasingly sophisticated building blocks are built up by reusing what's already been created. Often the designer won't even have to define the building blocks, because they'll be available from third-party libraries.

When the VDHL code is complete, it goes through a process called synthesis, which is the equivalent of compiling a programming language. Whereas compilation of a programming language checks the code for errors and, once it's error-free, generates a file containing individual processor instructions, the output of synthesis is called a netlist and it defines the connections between each and every component, including those 2.64 billion transistors.

3. Circuit verification and emulation

The netlist could go directly into the mask making process, but this would be asking for trouble because designs as complicated as a GPU are never 100 per cent correct on the first attempt. What's more, given that a set of masks could cost $1million, testing the design on real silicon would be prohibitively expensive.

Instead the design is verified and emulated - a hugely processor-intensive operation that requires supercomputing resources. Verification involves testing individual blocks with perhaps thousands of tests per block. Each time anything fails, the design team backtracks to correct the errors and then performs a full set of simulation tests to make sure the remedial action hadn't broken something that previously worked correctly.

Once all the individual blocks are operational, the team moves to emulation. This means exercising the GPU as a whole, but given the amount of processing time needed to simulate a multi-billion transistor chip, these tests might initially be nothing more complicated than drawing a single pixel.

In addition to functional testing, emulation also ensures that the chip meets its requirements in terms of processing speed.

4. Making the masks

With the simulation out of the way the designers know that the circuit connections are correct, but so far, with the exception of the top-level floorplan, no thought has been given to where the components go on the chip. This is carried out using a special CAD package, driven by the floorplan.

This largely automated process places each component and routes the copper tracks that will ultimately connect them all together. The culmination of this process is a major milestone referred to as 'tape out', and marks the transition from design to fabrication.

Since AMD is a fabless semiconductor company, this is also the point at which it hands the baton to TSMC, its chosen foundry for GPUs. Before any chips can be manufactured though, the foundry needs to create a set of photographic masks that will be used in the photolithography - one for each of the many layers by which the circuit is built up on the chip.

Using the data supplied at tape out (which can be thought of as images of the patterns on each layer), the masks are created as a patterned layer of opaque metallic chromium on the surface of quartz glass.

How GPUs are made, continued...

5. Photolithography

Photolithography is the key to many subsequent steps involved in making the GPU, and although we're going to introduce it here, it will be used over and over again as the circuit is built up, layer by layer on the silicon wafer. It involves applying a patterned mask to the surface of the wafer so that subsequent chemical processes only affect those areas with gaps in the mask.

The letters (a) to (e) in the following description correspond to the steps in the diagram. First, a layer of photosensitive material called photoresist is applied on top of any layers that have already been created (a). This is done by putting solution on the wafer and then spinning it so the solution spreads into a thin, even layer.

When the solution has dried, the wafer is exposed to ultraviolet light (UV) through one of the masks (b). This process changes the chemical composition of the photoresist where the mask allows the ultraviolet light to pass through. The wafer is immersed in a tank of developer that dissolves away those portions of the photoresist that had been exposed to the UV light (c).

With a partial layer of photoresist now in place on the wafer, it's possible to carry out a chemical process that will only affect the wafer in those areas where the resist has that removed - we'll see an example of how this works in step 6 (d).

With the chemical process now completed, the remainder of the photoresist can be removed from the wafer using a solvent (e).

The silicon wafer will contain hundreds of individual chips, or dies to give them their correct name, so the exposure stage above (b) is carried out several times - once for each die - with the wafer being moved relative to the mask and the optical system between each exposure.

6. Patterned oxide layer

We saw in step 5 that the layer of photoresist forms a suitable barrier to many chemicals, thereby allowing a chemical process to be carried out only on portions of the wafer as defined by a mask. Other processes - most notably those involving hot gases - would destroy the photoresist, so a different type of resist is needed.

In these cases, a patterned oxide layer, otherwise known as a sacrificial oxide layer (because it's subsequently removed), is used as described in the following steps (a) to (d). Again, the letters relate to the diagram.

The wafer is covered with a layer of silicon dioxide, which completely coats all existing layers (a). The processes described in step 5 are now carried out (b), the chemical process referred to in part (d) being the dissolving of silicon dioxide using hydrofluoric acid.

The end result, therefore, is a partial layer of silicon dioxide in the pattern of the required features. The necessary chemical process is carried out - this will affect only those portions of the wafer where the patterned oxide layer is missing (c). The remnant of the oxide layer is removed, again using hydrofluoric acid (d).

7. Creating the transistors

A MOSFET (the type of transistor used in GPUs) is an electronic switch. In other words, it's an electronic component that uses a signal on one circuit to control the flow of current in another. This is the most fundamental requirement in digital electronics.

Here we see how an n-channel MOSFET is created, but p-channel MOSFETs are also required. They differ only in that one has n-type material where the other has p-type, and vice versa. In the following description, changes are made selectively to parts of the wafer by use of either a layer of photoresist as described in step 5, or a patterned oxide layer as described in step 6.

The wafer is bombarded with phosphorous ions that implant themselves into the silicon through the gaps in the photoresist to create so-called wells of n-type material. This is a modified form of silicon that has additional electrodes to carry an electrical current.

Next, two smaller islands of p-type material are created within the n-type wells - these form the two electrodes known as the source and the drain of the MOSFETs. After this, a very thin insulating layer of silicon dioxide, just a few molecules thick, is deposited on the surface of the silicon between the source and the drain.

This is done using chemical vapour deposition (CVD), a process that takes place in a furnace filled with gases to chemically modify the silicon.

Finally, again using CVD, a layer of silicon is applied over the oxide layer to create the MOSFETs' third and final electrode, which is called the gate.

8. Connecting everything

We now have a wafer comprising several dies, each of which contains billions of transistors, but to convert these from isolated components into a working circuit they have to be connected using copper tracks.

First, an insulating layer of silicon dioxide is applied to the wafer so that the interconnecting tracks don't short all the MOSFETs together. Next, holes are etched in the silicon dioxide so that connections can be made to the MOSFETs' electrodes.

Then, trenches in the shape of the tracks are etched into the silicon dioxide before a layer of copper is applied by electro-plating. This covers the entire surface of the silicon dioxide, and fills the trenches and the holes to make contact with the MOSFETs.

Finally, the excess copper is removed using a process called chemical-mechanical polishing so that copper only remains in the trenches and holes. A single layer of copper interconnections isn't nearly enough to create a viable circuit. Since it isn't possible to connect everything in a single layer without making shorts, additional layers are used, each created in the same way as the first copper layer.

Increasing the number of layers can reduce the size of the chip, and some layers have to be dedicated to providing power, so ten or more layers isn't unusual in top-end chips.

9. Testing

The initial manufacturing processes are now complete to the extent that the wafer will contain several hundred dies, although typically not all of them will be functional.

The next task is to test the dies one at a time. This is carried out using a sophisticated piece of equipment called a wafer prober, which makes electrical contact with microscopic pads on the dies, ensuring correct registration by use of optical image recognition techniques. In this fully automated process, the machine remembers which of the dies passed the test.

Typically, for a state-of-the-art chip, the yield (the percentage of working dies), is in the 50-60 per cent area, although this will often improve as the manufacturing methods mature.

10. Packaging

Once the testing process is complete, the wafer is sawn up into individual dies and the non-functional ones are discarded. The final step is to take the working dies - tiny rectangles of silicon that are far too flimsy to be used as regular electronic components - and package them into what most people think of as a chip, ready to be soldered onto a circuit board.

This involves bonding the die onto a substrate, and making connections between the minuscule pads on the die and the somewhat larger solder bumps that will eventually be used to make the electrical connections on a graphics card.

A final functional test is carried out to ensure that the packing has been successful, and this miracle of digital technology is then ready to be shipped to the graphics card manufacturing facility.

We'd like to express our thanks to David Nalasco, Senior Technology Manager at AMD, who explained the intricacies of the design process to us.

The death of DirectX

Time was, the latest version of DirectX would render previous versions outdated and redundant, but lately things have changed.

The first version of DirectX 9 appeared way back in December 2002 (older than most Justin Bieber fans, then) and yet we're still seeing the most eagerly anticipated big budget releases built around DX9 almost a decade later.

It isn't as though DirectX 10 and 11 have nothing to offer. You only have to load up Crysis to see what's possible when a development studio goes to town with the new features a graphical API has to offer. The jungles of Crysis offered up motion blur, post-processing and lighting effects that we simply hadn't seen before. It caused quite a stir, you might recall.

Things looked bright in the early days of DX10. Bioshock's idiosyncratic graphics style made good use of the API, and Far Cry 2 made war-torn African tundra look appealing. No one matched Crysis' DX10 smorgasbord though.

Perhaps they were put off by the fact that no one had a PC that could run it. Developers seemed to be reigning in on DX10. Instead of making a game take a giant leap forward into post-console visuals, DX10 releases looked increasingly similar to those made with old man DX9.

There were bigger problems emerging, too. DirectX 10 was only available to gamers who'd taken the plunge and upgraded their OS to Vista. Many were perfectly happy with XP and felt they didn't need it; others bought it and found a myriad of problems from front end to driver support. DX10 then, was bound to an operating system few liked. The gaming world lost its appetite for Microsoft's new API.

The journey of an API

In late 2009 DirectX 11 was released. It had a lot of wrongs to put right, and a lot of mistakes to avoid. This time it was tied to an operating system that people did like, and showed impressive potential for gaming graphics and performance… and yet we still haven't seen the standout DX11 title. What we have seen are some impressive DX9 games, and even the odd OpenGL title.

Throughout the nineties, DirectX and OpenGL locked horns. They scrapped to be the dominant application programming interface (API) to replace the proprietary interfaces that proceeded them. Each wanted to be the API that graphics card manufacturers would write driver support for, and that game developers would use as their toolset.

SGI's OpenGL had the upper hand until the end of the decade, thanks to an alliance with id Software and the Quake engine that powered Half-Life in a heavily modified form, not to mention its subsequent blockbuster sequels to the Doom and Quake franchises.

With each new game, the id Tech engines expanded their graphical features in parallel with OpenGL, and proved enticing engines for other developers to build their own games with.

As the decade ended, Microsoft muscled OpenGL off the API throne with DirectX 8. For the first time, it not only matched the latest OpenGL, but surpassed it by including vertex and pixel shader support.

Microsoft also launched its Xbox, hitting the shelves late in 2001, which ran on an operating system similar to Windows NT, and used a Windows API- DirectX 8.1. This opened the door for cross-platform releases. Developers had a shared toolkit across the PC and Xbox platforms that made programming easier.

Anyone who tried to play Resident Evil 2 or the like on PC will remember the calibre of cross-platform games - or console ports - developed with the console in mind, then re-coded for PC before this point. The vast majority were all but unplayable; controls were poorly translated to our beloved keyboard and mouse and game code was unstable and glitchy.

So, unsurprisingly console games never sold particularly well when they finally made their way onto PC.

The game that bucked that trend was Halo. It was originally developed for PC and Mac, but when Microsoft bought Bungie Studios they saw it made sense to use the buzz the game was getting to boost their fledgling Xbox. As a result, we had to wait two years for Halo to make its way to the PC, grinding our teeth as Xbox owners banged on about how freakin' good it was.

When it actually was released on PC in 2003, it came with DirectX 9.0 support. Perhaps Bungie felt bad about making the people who'd supported their game from the start wait so long, but the fact was the PC release offered more than the original console release - shiny new DX9 visuals that weren't possible on Xbox.

Halo's PC incarnation wasn't perfect. Frame rates were low and performance was often choppy - but it was playable, mouse-friendly and pretty, and that release set the protocol for cross platform releases through the next decade.

Sharing the DirectX API across platforms made for less buggier games. DirectX 9 was a quality API that allowed a lot of flexibility beyond standout visuals; OpenGL had been well and truly muscled out of the action while Microsoft sank its fingers into the warm pies of PC and console gaming.

Hard history lessons for DirectX

The current generation of consoles also proved to be pivotal for DirectX, but this time for the worse. The Xbox 360 supports a modified build of DirectX 9.0c with Shader Model 3 support. Sony's PlayStation 3 natively uses a flavour of OpenGL.

Here in PC land, we're on DirectX 11, which is no big deal to us since we can easily stick in a new GPU and unlock a new generation of graphics technology, but on consoles, the native API is ruler for life. That means if a developer works hard on integrating DirectX 11's tessellation and multi-core rendering into a PC title, they then have to code it back out for the console version.

Whether or not that developer has built their engine from the ground up, that's an expensive and time-consuming undertaking. Given the small profit margins PC titles can expect to generate, it's simply not worth it for many game studios. Is that why we've seen even big-budget blockbusters like Crysis 2 and The Witcher 2 stick with DX9.0c?

Bartlomiej Wronski, graphics programmer for The Witcher 2 explains: "I think the main reason not all developers - including us - provide DX11 support is that supporting two totally different rendering APIs requires much more work and testing. It also creates pipeline difficulties. We would have had to decide which features have fewer fast DX9 fallbacks, which are rendered in lower quality, and which are totally dropped and this would have meant even more work for our artists, programmers and testers.

"The Xbox [360] uses an API similar to DX9, with only few extensions from DX10/11, like texture arrays or hardware tessellation, so to create a multi-platform engine, we had to base ours on DX9. We understand it's an ageing API, and that DX10 and 11 have new features and performance - it came down to balancing these and other considerations."

But there's more to consider than consoles, says Wronski. It's not as if every PC gamer has a DX11-ready machine:

"The main reason games still implement DX9 is to support the many players whose PCs run Windows XP and older graphic cards. Fortunately, DX9.0c with shader model 3.0 is still a pretty powerful API, which allowed us to create a rendering engine capable of creating beautiful locations, characters and scenes."

So developers are faced with a fairly small user base when developing a DX11 title, compared to the prospect of a cross-platform release and reaching gamers with dusty old XP machines by working with DX9.0c. The goalposts are constantly shifting, though, as gamers are able to upgrade their machines.

This makes for an uncertain future for DX11, as Wronski says: "While DX11 is already a very good and fairly mature API, many of its capabilities aren't widely used yet, so it's hard to speculate about new versions. All its cool features like geometry shaders, texture arrays and dynamic hardware tessellation are still awaiting broader implementation. Direct Compute could also help a lot: vendor-independent GPU computing is a feature not only graphics can benefit from, but also particle and physics simulation."

The number of DX11 gamers is growing, and they will dominate the market soon, so including it is an obvious, important next step in engine development.

"We believe that it may be up to the next generation of consoles. The usage of APIs in future games depends on the capabilities of the next-gen console GPUs," says Wronski. So from a developer's perspective, they'll blink when gamers do.

When there are enough DX11 end users open-mouthed for flashy games, it becomes commercially viable to make use of DX11 features - and if eighth gen consoles support DX11 natively, that viability more than doubles.

When Microsoft released Windows Vista, it's fair to say it took a bit of a kicking. There were driver issues, front end gripes, performance problems, all reflective of the enormous changes under the bonnet. But perhaps Vista's biggest problem was that Windows XP still worked just fine.

People weren't chomping at the bit for a new operating system, they were settled into XP as one settles into a comfortable chair. The incentive to jump from that comfortable chair onto the Indian bed of nails that many regarded Vista to be was DirectX 10. XP would never get DX10 support, if you wanted to play DX10 games, you were forced to buy Vista. That was something new for Microsoft.

Previously, new versions of DirectX had been unbound to any particular operating system. You also needed a new DX10 graphics card, and that was no small change. Suddenly, you'd gone from comfy old XP and a healthy bank account to an unfamiliar OS and empty pockets. Why put yourself through it?

Well, for one reason, Crysis. Remember how quickly things moved during the emergence of 3D gaming? How you'd sit there, gazing in wonder at Doom, wondering what kind of transcendent visuals Doom II would have? Well, Crysis looked like that to our 2007 eyes. It was simply one imperial unit of graphical quality higher than anything else.

It was clear Crysis was going to need a brute of a PC to play, and importantly it generated enough enthusiasm from PC gamers that they went out and bought those brutish PCs. It was only when the dust had settled after Crysis that gamers found faults with DX10 itself.

First, only Crytek seemed capable of drawing those cutting-edge visuals out of the API. It had set the bar high, but other developers didn't get near it. Second, DX10 wasn't a performance-enhancing API. In fact, all games that shared the API seemed to also share terrible frame rates. Gamers that bought into DX10 were hugely disillusioned.

Current climate

It would be understandable then if PC gamers had their trepidations about DX11. This time it isn't bound to a Windows OS, in this case Win7 as its predecessor was, but it does require a new generation of graphics card and the subsequent outlay.

And yet, the trusting, doe-eyed PC gaming community, never one to boycott a company or product based on even the slightest mishap or glitch and then spend months furiously typing obscenities about that company or product in their forums long after everyone stopped caring, did have faith in DX11. And they did buy into it.

Matt Ployhar is the president of the PC Gaming Alliance, Senior Product Planner at Intel and former Microsoft employee with experience in MS Games Studios and the Windows/DirectX team through the DX10-11 development phases. He's as informed as anyone on Microsoft's API, and explains that there are in fact plenty of DX11 gamers ready and waiting for DX11 games.

"The DX10 to 11 GPU-discrete Install base alone is now sitting somewhere around 250 million unique PC gamers across every segment and geography. The problem for DX10 and 11 isn't the Total Available Market (TAM) or install base, which incidentally is larger than all the seventh gen consoles combined, the problem is that it's more analogous to being like a largely untapped oil field."

So unlike CD Projekt's vision of the PC landscape, Ployhar sees DX11 as a graphics revolution about to happen if developers would only tap that oil field. So, why aren't there more DX11 games in development?

"Good question," says Matt Ployhar. "There should be. DX11 would likely be the most robust option available presenting the least amount of compromise to one's vision for a game.

"The real problem is probably the lack of incentives, carrots, whatever you want to call them, for making PC games in general. Your console manufacturers can spend 100s of millions a year to single digit billions for a specific platform; and that equivalent spend doesn't occur on the PC."

It seems that the only thing holding DX11 back is time, rather than any failing with the API itself. Perhaps it's predecessor DX10 didn't deserve its bad rep. It paved the way for DX11, for starters:

"Getting people to switch to a new API is never an easy thing to do. It took several years for Microsoft to get the foothold and market share it did with D3D. This obviously didn't happen overnight. Nor should we expect it to happen with DX10-11 and beyond," says Ployhar.

"DX10 didn't really fail. (The API) was going up against a large established DX9 TAM and Install base. DX9 is very mature in terms of support, familiarity, tools, and so on. Once DX10 - which has matured into DX11 - becomes more mature and prevalent we'll start seeing the ocean turn in favour of what is effectively the DX10 code path"

OK everyone, take a knee and let's gather our thoughts. DX10 got a tough time because it came tied with Vista, and at a time when the PC's financial ecosystem was - and is - being ransacked by piracy. Thus, cross-platform releases made for safer releases and that marginalized DX10 coding.

However, DX10 grew into the beautiful butterfly that is DX11, and it seems gamers are ready and waiting to see what developers can achieve with it. The past few years might have changed the way we look at DirectX, but it's still as important to PC gaming as ever.

Hong Kong technology company Sapphire has slapped its proprietary Vapor-X technology onto an overclocked AMD Radeon HD 6850 graphics card, according to a release from the company.

The card ships with clock speeds boosted to 800Mhz for the processor, and 1100Mhz for the memory, which Sapphire says gives users "a considerable boost in performance."

To cope with the overclocked clocks, Sapphire has added a Vapor-X cooler to the card. This uses the laws of thermodynamics and evaporation to cool the card with liquid in a sealed chamber.

Sapphire's Vapor-X technology was first introduced in 2007, and we've been rather impressed with it - it seems to actually work.

As well as the dark arts of Vapor-X, Sapphire has included 10-phase power control, black diamond chokes and its TriXX software for overclocking.

To Eyefinity and beyond

On the back of the card lurk the standard double-DVI ports, as well as one HDMI and two mini DisplayPort connectors. If you're lucky enough to have four monitors, these can be connected to the DVI and DisplayPort connectors for eye-popping Eyefinity vision.

It also supports DirectX 11 with hardware features such as an enhanced tessellation unit, HDR texture compression and 3D stereoscopic support.

The Sapphire Vapor-X HD 6850 has a UK release date of September, with its price to be confirmed.

Component and computer manufacturer Asus has unveiled its Republic of Gamers MARS II graphics cards, which ship with dual Nvidia GeForce GTX 580 chips.

The ever-modest Asus claim that the card is the world's "fastest, most powerful graphics card", and that it sets "new records in video and gaming excellence".

Each graphics processing unit runs 22% faster than reference GeForce chips, and Asus says this will let the card run DirectX11 games in resolutions higher than 1080p. Asus has also crammed 3GB of GDDR5 video memory onto the card.

Ever keen to worship the gods of thermodynamics, Asus has cooled the cards with two 120mm fans, which are said to move 600% more air than reference designs - which is also great if you're thinking of teaching a canary to sky dive.

Push the button

A turbo fan button is incorporated on the card itself, which cranks both fans up to 100%, regardless of driver settings. The button is on the side of the card though, so you'll need to open your case to access it, or use a really long stick.

If that's not enough, Asus has used 21-phase Super Alloy Power technology, which combines a 21-phase power architecture with a special alloy mix for a "15% performance boost", and also more than doubles the lifespan of the card.

The card ships with Asus's GPU Tweak utility, which makes overclocking nice and easy, as well as a "laser-carved, sequentially-numbered aluminium plate" which certifies the card's limited edition status, and is sure to impress the ladies.

Top graphics cards for 3D gaming: 8 reviewed

Stereoscopic 3D might seem straightforward, but you'll need a lot of boxes ticked if you want to enjoy it on your PC. As well as a 3D-capable monitor with a refresh rate of 120Hz, your computer needs a graphics card that's powerful enough to handle 3D content.

To generate 3D images, the graphics card has to produce the graphics twice, creating a slightly different image for each eye. This extra workload can put a lot of strain on the card.

Most modern cards include 3D capabilities - especially recent Nvidia cards, which support Nvidia 3D Vision - so finding one that can display 3D content isn't a problem. The trick is finding a graphics card that can display 3D without any noticeable drop in image quality.

You don't want to spend money on a card only to find that the frame rate plummets when you turn on 3D effects, leading to choppy, stuttering movies and unplayable games.

Upgrading from two dimensions isn't the only reason to buy a new graphics card. More powerful GPUs make for better graphics all round. While modern computer games benefit most from a powerful card, other applications will also show an improvement. Video editing in particular benefits from a muscular card, because many movie-making tools now take advantage of technology like Nvidia's CUDA, which hands more processing tasks to the graphics card, freeing up the PC's processor for other tasks.

Getting the right graphics card can make a huge impact on your computer, and we'll show you some of the best.

EVGA GTX 460 1GB - £139
www.evga.com

MSI GeForce GTX 560 Ti - £176
http://uk.msi.com

Zotac GeForce GTX 570 - £260
www.zotac.com

Zotac GeForce GTX 580 - £385
www.zotac.com

Asus GeForce GTX 590 - £580
http://uk.asus.com

Sapphire Radeon HD 6950 - £181
www.sapphiretech.com

HIS Radeon HD 6970 - £265
www.hisdigital.com

Asus Radeon HD 6990 - £501
http://uk.asus.com

EVGA GTX 460 1GB

Nvidia has had a long history of creating superb mid-range cards, from the 6600 GT to the incredible 8800 GT, and the GTX 460 can sit happily among such illustrious forebears. Its launch price was £200, but you can now easily pick up an overclocked 1GB version for as little as £140.

If any graphics card can be described as a hero, surely this is it. It no longer leads the market in performance, but it's not as far off the pace as you might imagine from its age. It's the oldest and slowest model in this test, but its price keeps it competitive.

When you consider that it's nearly £50 less than the cheapest HD 6870, and with performance that is just shy of the AMD card, it's quite a bargain.

Read TechRadar's full EVGA GTX 460 1GB review

MSI GeForce GTX 560 Ti

Just as the GTX 570 was beginning to replace the GTX 480, leaving the GTX 580 on its own at the head of the single-GPU pack, the GTX 560 Ti appeared to take over from the GTX 470. That's a lot of numbers, but it does make sense - especially when you look at the performance of those GTX 460s, even in today's highly competitive market.

It would make no sense for Nvidia to kill off one of its best cards, but it's not quite as simple as just another transistor-level tweak of the GTX 470's design. The GTX 560 Ti is more of a halfway house between the GTX 470 and GTX 460; it has 64 fewer CUDA cores and eight fewer ROPs than the former.

To make up for this shortfall in GPU spec, it comes out of the box with the fastest core clockspeed we've seen in an Nvidia GPU since the FX 5xxx series.

Read TechRadar's full MSI GeForce GTX 560 Ti review

Zotac GeForce GTX 570

This card represents the second tier of Nvidia's Fermi lineup. It's an impressive performer, and a little cheaper than AMD's top single card, the HD 6970. It has more in common with the Nvidia GTX 480 though, with almost the same basic GPU layout.

The GTX 570 has the same 480 CUDA cores and 60 texture units, but has 40 ROPs compared to the GTX 480's 48. The 570 also has a smaller frame buffer, at 1,280MB compared to the previous generation's 1,536MB. It packs the same transistor-level re-engineering as the GTX 580, though.

As such, it's a much more efficient, cooler graphics card, and its performance metrics will be almost as good if you're running a sub-30-inch panel.

Read TechRadar's full Zotac GeForce GTX 570 review

Zotac GeForce GTX 580

Let's not beat around the bush; Nvidia's GTX 580 is the best single-GPU card on the market. It held sway over all the others across every one of the benchmarks we ran.

This is due to the GF110 GPU at its heart. Fermi has proven to be the best GPU architecture Nvidia has ever manufactured, and with this second generation (the GTX 5xx series) it's gone from strength to strength.

The first Fermi, the GTX 480, was the top card of its day, but not without issues. These were the amount of power it needed and the amount of heat it generated. For the GTX 580, Nvidia's engineers went back to the drawing board and tweaked the design right down to the transistor level, tailoring the smallest parts to give the best performance where needed, and the best efficiency where speed wasn't the goal.

Read TechRadar's full Zotac GeForce GTX 580 review

Four more graphics cards to choose from

Asus GeForce GTX 590

Nvidia has been the market leader in graphic card technology for so long, when it says it's reached the pinnacle of engineering with the fastest single-PCB (printed circuit board) graphics card it's ever made, you sit up and take notice. It's quite a boast, and it sounds amazing on paper - one board with two of the most powerful GPUs Nvidia has ever made combined on it.

However, compromises have had to be made to get the chips to run happily on a single card, and to stay within the 375W limit imposed by a pair of eight-pin PCI-e connectors and the motherboard PCI-e bus. That means we don't quite have a pair of the GPUs ripped straight out of Nvidia's flagship single-GPU card, the GTX 580.

Instead of the core clock of 772MHz and a shader clock of 1,544MHz, the GTX 590 houses a pair of chips running at a seriously toned-down 607MHz core clock and 1,215MHz shader clock. The memory clock has been dropped from 2,004MHz to 1,707MHz.

Read TechRadar's full Asus GeForce GTX 590 review

Sapphire Radeon HD 6950

The Sapphire Radeon HD 6950 appears to be the finest graphics card AMD has produced in a long time. In the sub-£200 market, it's as good as you're going to get, and it's a graphics card that offers much more than any of the opposition in its price bracket. The HD 6950 comes close to the HD 6970 and the GTX 570 in its plain state.

The 2GB GDDR5 frame buffer means it can more or less keep pace with Nvidia's secondtier Fermi card, although it lags behind slightly for the most part. It does the same with the HD 6970, but that's to be expected from a card that's so much cheaper.

The sub-£200 price tag is a fantastic psychological coup, making this an affordable yet powerful option. That impressive performance is all down to the Cayman GPU. As we've already mentioned, the Cayman contains the same core as the GPU powering the more expensive HD 6970, though with a lower clockspeed and a few other components switched off at a software level. This leads to a problem...

Read TechRadar's full Sapphire Radeon HD 6950 review

HIS Radeon HD 6970

AMD's fastest single-GPU model is a tough one to recommend. It's certainly the quickest AMD has released - the big problem is the second-tier card the came out alongside it - the HD 6950.

The Cayman XT at the heart of the HD 6970 is essentially the same GPU you'll find paired in the Antilles chip of the dual-GPU HD 6990. It's clocked faster than the stock specs of the Antilles, but thanks to the Antilles Unlocking Switch (also known as the Screw Your Warranty Switch by those who choose to use it), the HD 6990 can run with its GPUs at the same speed as the HD 6970.

You can get two HD 6970s for about the same as one HD 6990. So far, so good, but what of the HD 6950 we mentioned?

Read TechRadar's full HIS Radeon HD 6970 review

Asus Radeon HD 6990

AMD pipped Nvidia to the post, releasing its dual-GPU, top-end graphics card a few weeks before the GeForce GTX 590. The HD 6990 is the fastest single-PCB graphics card the Texan brand has ever made, and houses two of the cores from its current fastest single-GPU card - the HD 6970.

The Cayman XT from that card has been doubled up and redubbed 'Antilles', and comes with slower clock, shader and memory speeds out of the box. You're not taking as much of a hit in clockspeed as with the GTX 590's GPUs, but that's because the Cayman XTs run a little slower than the GF110 GPUs in the GTX 580s in the first place.

There is a key difference here, and that comes from the little Antilles Unlocking Switch. Like the Cayman-powered cards that came before it (the HD 6950 and HD 6970), the HD 6990 comes with a dual-BIOS switch hardwired into the PCB. This comes with an attractive yellow warning sticker covering it, which warns you that as soon as you flick the switch, you effectively invalidate your warranty.

Read TechRadar's Asus Radeon HD 6990 full review

The award winners

Most of the cards we tested here do the job well, and upgrading your PC with any of them will give a serious performance boost. However, there are a couple that you should avoid to get the best possible 3D graphics on your PC, and there are others like the Zotac GeForce GTX 580 that stand out from the crowd.

Things become a little more confusing when you consider using multiple graphics cards in SLI or CrossFire mode. This allows two or more cards to work together to boost performance. Although it might seem slightly unintuitive, it can be cheaper to buy two graphics cards to get the same performance as one expensive card.

If you're really after the best graphics and your computer can handle it, then this is the way to go.

Editor's choice

Zotac GeForce GTX 580

The GeForce GTX 580 is Nvidia's flagship card for a reason - its performance in our tests was extremely impressive, and it can handle the latest games and 3D content with remarkable ease. It might be a bit on the expensive side for a graphics card, but it's money well spent. If you have the budget and the space to spare on your motherboard, then a pair of GeForce GTX 580s will serve you well and effectively future-proof your PC for many years to come.

Verdict: 5/5

Value award

Sapphire Radeon HD 6950

While the GeForce GTX 580 has the best performance for a single card, the Sapphire Radeon HD 6950 easily wins the best value award. On its own it's a great performance card, but when you also take into account the overclocking potential of this graphics card, you have a fantastic product for under £200. Upgrading to 3D is expensive enough, so it's heartening to see such good value for money.

Verdict: 4.5/5

AMD's focus on its Fusion accelerated processing units could see the processor giant ditching its low-end Radeon graphics cards.

The Fusion architecture combines the CPU and GPU on a single chip, negating the need for a dedicated graphics card.

Interim chief executive Thomam Seifert told analysts and investors that "parts of this business will be cannibalised and the low-end discrete GPUs will be replaced with Fusion-type products."

Long-term

Seifert didn't set a timescale for the move away from discrete graphics, but did say he was looking at the long-term future of the company.

AMD bought graphics card company ATI in 2006, and immediately started working on the Fustion platform. The acquisition saw the ATI name gradually replaced with AMD's on graphics cards.

A move away from discrete graphics cards and chips makes sense - but only if it's timed correctly. Adding a graphics card - no matter how cheap - to your PC is a guaranteed way to boost performance, particularly in gaming.

But as Fusion takes off low-end GPUs will be rendered redundant. The question is if Fusion will take off, and whether or not AMD can afford to lose the cheap and cheerful end of the graphics market.

Seifert pointed out that AMD doesn't make a huge revenue from its low end graphics card, and seems confident that the move will work out. "This is all goodness for us," he said, "because it replaces low-cost margin revenue with high-gross margin revenue."

AMD has launched the Radeon HD 6990M – claiming that it is the world's fastest single mobile graphics processor.

With laptops increasingly becoming our daily computers the need for powerful graphics processors is increasingly obvious, and the AMD Radeon HD 6990M is apparently 25 per cent faster than its nearest rival.

"There's always been a belief that when it comes to mobile computing you need to make performance compromises. Today AMD demolishes that myth," said Matt Skynner, corporate vice president and general manager, GPU division, AMD.

"The AMD Radeon HD 6990M GPU, which not only packs AMD Eyefinity technology with unprecedented specs, also provides full Microsoft DirectX 11 and Stereo 3D support. Bottom line, this processor is epic and it's here - now."

Significant upgrade

AMD is pointing out that the Radeon HD 6990M represents a 'significant' upgrade from its predecessor the HD 6970M, with DirectX 11 support and 3D support ticking the requisite boxes for gamers

"As the PC gaming technology leader, AMD is dedicated to delivering the fastest technology and industry support needed to help create the best possible gaming experience on the PC," added AMD.

"Whether it's driving up frame rates, extending quality settings or pushing clock speeds8, the AMD Radeon HD 6990M GPU is the ideal notebook for hardcore gamers looking to run the industry's most demanding DirectX 11 games such as Dragon Age 2, DiRT 3 and Shogun 2."

Nvidia's GTX 590 is not a card that needs to be overclocked for gaming performance. Two GF110 GPUs, as found on the GTX 580, working in unison will kick any game's behind, and the idea of being short of frames with the card's stock clock settings is wild and preposterous.

Graphics cards compared: what to look for

There's one component in your machine that will be superseded by faster and more powerful versions quicker than any other, and that's the graphics processor.

The graphics card is the supermodel part of any modern gaming PC. It, more than anything else, makes your games look beautiful and run as smoothly as a baby's velvet smoking jacket.

So, if you're a gamer and you've got a bit of cash to spend - it's time you looked at some graphics cards compared and rated.

After all, forking out the lot on a new graphics card is the best way to up the frame rate of your favourite games and make the girls think you're attractive, right?

Unfortunately, that's not necessarily the way the graphics game works. The top-performing rigs are always the most well balanced. Put a Ferrari engine into a Mk1 Golf GTi and it will go fast, sure, but you'll fly off the first corner you come to and probably make someone an orphan.

To a certain extent, it's the same with a PC. Obviously it's not going to career off the desk and decapitate a passer-by, but using the fastest graphics card in the world for your rig isn't always going to make it the best machine.

There are a number of key considerations to think of before spending an obscene amount of cash on a GPU, or a pair of GPUs. Number one is, naturally, can you afford it? But we'll skip the personal economics lesson for now.

The second issue is, can you get the most out of it? This is a multi-part question, though, and covers the size of screen you're running, possible bottlenecks in your system (such as OS, CPU and memory) and whether your power supply can handle the extra draw that comes with a top-end graphics card.

Then it all comes down to which card, or combination of cards, suits you best. This is the key point – do you go for a single, high-performance model or opt for a (possibly cheaper) multi-GPU solution? It's all a bit complicated, isn't it? But have no fear: we've handcuffed ourselves to the test-bench and won't move until we've tried the top 20 graphics setups around today.

Both Nvidia and AMD have released their most expensive and powerful graphics cards ever in the last few months. With the launch of the GTX 590 (Nvidia) and Radeon HD 6990 (AMD), the top of the market has never looked so good. But how do you get the most out of such outrageously priced components?

You want to make sure that, if you spend the best part of £600 on a graphics card, you're wringing as much potential as you can out of it. The one thing a top graphics card such as the dual-GPU GTX 590 or HD 6990 wants is a high-end display setup. A good, high-resolution desktop is going to push your expensive silicon to up its game, and will make sure you get the most out of its pixel-pushing prowess.

Buffering up

Both of the latest dual-GPU cards from the big-name manufacturers come with a frankly huge amount of graphics memory. The GTX 590 has a chunky 3GB of GDDR5 and the HD 6990 a positively bloated 4GB. It's these large frame buffers that help enable the cards to output at the crazy resolutions afforded by the multi-screen setups AMD and Nvidia are championing.

Just how much the frame buffer means is shown by the GTX 460. That appeared in both 768MB and 1GB trim, with the former version significantly worse at higher resolutions than its memory-laden brother. Either of them will breeze through any game you throw at it if you're running at resolutions such as 1,280 x 720, and even up to 1,680 x 1,050. But as soon as you go past the 1080p resolution of 1,920 x 1,080, lower-end cards such as the 768MB GTX 460 will start to struggle.

This is what you get when you drop £600 on a graphics card, then: the knowledge that it will cope at pretty much any resolution you throw its way.

Five stars

When AMD released the HD 6990, it came out blowing the trumpet of quint-screen gaming. For those who aren't aware of the phrase we just made up, it means five screens with a possible resolution of 5,400 x 1,920. That's five 1080p screens running in portrait mode, connected together.

The GTX 590, on the other hand, can only cope with three screens. It is, though, the first time Nvidia has released a single-GPU card that can output to more than two screens. Either way, it's a huge amount of screen real estate to play with.

If your graphics card can cope with that scale of resolution madness then you're surely onto a winner. You do have to be careful, though, because there aren't many games out there capable of taking full advantage of such screen size. Most can cope with the 30-inch panel's native resolution of 2,560 x 1,600, but few are able to deal with multi-monitor setups.

Take a look at the Dragon Age 2 screenshot AMD supplied with its HD 6990 press material – with half of your HUD on screen one and the other on screen five, you're going to get whiplash fast.

Some games have been coded with the likes of AMD's Eyefinity panoramic technology in mind. They offer the chance to keep all the HUD elements on the central screen while still affording you the ultra-widescreen format. Five screens might be pushing it a little, but a three-screen setup capable of 1080p and above in portrait mode would be hitting the sweet spot.

There's one more thing to be aware of with a portrait setup, and that's the screen technology used. You'll need to make sure the monitor you choose has a decent viewing angle; something the traditional TN panel can't offer.

Ideally, then, you'll need to splash out on the more expensive IPS panels. Otherwise, the vertical viewing angle, which becomes the horizontal angle when put in portrait mode, will seriously impede your image quality.

Space to burn

So, once you've made sure the display setup is in place, how do you ensure that your rig is up to the task? The first thing may seem obvious, but it's all about space. High-end graphics cards or multi-card setups need a lot of that.

The Radeon HD 6990 is the same size as the HD 5970 and, as such, is one of the longest cards you'll find on the market. It's worth making sure your chassis is capacious enough to cope with the extra few centimetres the big cards need.

On the next tier down, the HD 69xx cards are fairly chunky too, with the Nvidia models coming in a bit shorter and more likely to fit. Testing on a mid-size CM Storm Scout chassis, we found we could easily fit a GTX 580 inside. On the AMD front, though, the fastest card that would fit the case was the HD 6870.

That's not the only consideration, though. If you do manage to squeeze an HD 6990 into your case, you'll have to make sure there's enough airflow to avoid the thing suffocating and its silicon brains melting out of the choked vents. Top-end cards are, essentially, electric fan heaters. The amount of heat they generate can be directly attributed to the amount of power they need to run, so something like the GTX 590 is going to push out a lot of heat.

The new dual-GPU cards from AMD and Nvidia have both been set up with a symmetrical cooling design. Unfortunately, that means that just as much hot air is blown into the case as is vented out of the back.

The problem isn't quite so bad with multiple graphics card setups because, as with a single-GPU card, you can vent directly out of the back. Multi-card rigs will need to have space to breathe too, especially between the cards. If they're stacked on top of each other then you're going to end up with some seriously hot silicon.

The next concern is just how much power you're going to need to keep your graphical behemoths juiced up. Multi-GPU cards and SLI or CrossFire arrangements need an extraordinary amount of power just to get them up and running.

With something like the GTX 590, Nvidia recommend a 700W PSU as a minimum. If you're looking at the HD 6990 or a dual-GPU GTX 580 setup then you're probably going to need an 850W power supply.

Before you go out and spend the cash on a 1,500W PSU, though, it's worth remembering to make sure you've got enough spare capacity, because an under-used high-end supply is far more inefficient than a lower-capacity unit running at the same power draw.

So a 700W supply drawing 500W is going to be much more efficient than a 1,500W PSU drawing the same 500W. PSU inefficiency means wasted energy, and that means excess heat. With high-end graphics cards, that can be dangerous.

Letting the side down

As well as checking ratings to cover the amount of power a particular card needs, there are other concerns. The latest dual-GPU cards make serious demands of your PSU's 12v rail.

Both the HD 6990 and GTX 590 have been designed to stay below the 375W maximum the twin 8-pin PCI Express power connectors and motherboard PCIe bus can supply, but when you factor in overclocking, that goes straight out of the window.

The HD 6990 comes with a BIOS switch that boosts the core clocks to the same settings as the HD 6970 GPUs the Antilles chips are based on. This increases the power draw to insane proportions, and can overload the 12v rail in lesser PSU lights.

We spoke to one of Enermax's PSU engineers about this at launch, and he explained that customers need to consider two things when purchasing a PSU for top-end graphics cards: "The PSU should be equipped with OPP function, and should have a strong 12v line, or lines."

OPP means 'Over Power Protection', and this technology shuts down the PSU if the graphics card is demanding more power than the 12v lines can give. So you don't get a cascade failure effect if the PSU falls over.

Another key consideration is your CPU. If your processor can't cope with the pace of your new graphics giant then you'll end up with a bottleneck in the system. This means that your expensive bit of silicon will be sitting there twiddling its transistors, waiting for the laggy CPU to catch up. A modern, multi-core CPU ought to be able to cope with anything a new game can throw at it, though.

Granted, a lowly AMD Athlon II dual core will still give you a bottleneck, as will a sub-2GHz Intel Core 2 Duo, but any Phenom II quad or Core i5 should keep things ticking over nicely.

Double trouble?

So, you've made sure your chassis can house a beefy graphics setup, you've sorted out a power supply that won't collapse and turn the insides of your gaming rig into pricey slag and you've got a decent processor at the heart of your machine so the graphics don't get ahead of themselves. Now, what sort of high-end graphics setup do you want?

The options at the pricey end of the market are fairly wide ranging. At the very top you're looking at CrossFireX and quad-SLI setups, with four of AMD or Nvidia's top single-GPU cards being thrown together on one motherboard. That will create a graphical array more than capable of the highest resolutions games will realistically play at.

It's also entirely possible that it will create a black hole. In your wallet, that is. It's this sort of setup that makes the single PCB, multi-GPU cards more attractive, and seem like better value. That's quite a feat considering they cost about £600 themselves.

The benefit of these single-card solutions is that they take up far less space inside your machine. And if you're an Nvidia fan who's not blessed with an SLI-ready motherboard, you can now run a twin-GPU setup. If you do have an SLI-ready motherboard, though, and one that will happily support quad-SLI, then you could drop a pair of GTX 590s in for some serious graphical action.

The paucity of SLI-certified motherboards is only one of the problems with multi-GPU arrays, though with AMD's Bulldozer boards looking increasingly likely to support SLI, that situation could change.

But, still, driver problems persist when you add extra graphics processors. More often than not you'll find that the latest games don't work, or that you don't see one of your GPUs at launch. Then there's a wait for the driver issues to be sorted out by the relevant manufacturers.

This isn't an exclusive problem to either Nvidia or AMD – they're both as bad as each other now. A fast, single-GPU card will save you headaches down the line, then, as well as a fair bit of cash. But if you absolutely must have the highest frame rates then multi-GPU options are the way forward.

Looking down the spectrum and pairing up a couple of lower-end GPUs, such as a couple of GTX 460s, will net you performance that will make anyone who forked out for a GTX 580 a little green around the gills. Why stop at two GPUs, though?

The performance of twin cards has gone through the roof in recent years, to the point where we're close to doubling the performance. But, unfortunately, adding in more cards is very much a case of diminishing returns. With the performance boost you get, the second card is totally worthwhile. A third or fourth addition, though, will barely register.

We tested a pair of GTX 590s and a pair of HD 6990s and it was rare indeed that we saw a significant improvement in performance. Heaven 2.5 was the only benchmark that used the four GPUs properly, nearly netting double the performance. The rest of our benchmarking suite ably showed just why SLI and CrossFire performance is so hit and miss.

Aliens vs Predator only showed an increase with Nvidia's GPUs, Lost Planet 2 gave extremely poor returns and DiRT2 went backwards when we dropped in a second HD 6990.

Up the resolution

Finally, it's important to consider what you're going to be using your high-end graphics setup for: gaming. The problem at the moment is that because of the cost of games development, most publishers are focusing their efforts on the consoles. That's where the big profits lie.

The PC is more powerful than any console, a fact shown by all the recent TV ads for multi-platform games, clearly using PC footage to make sales. Unfortunately, though, PC games are still being held back by the lowest console denominator – it's simply not worth the developers' time to push a game for PC hardware.

So we come back round to the start, and what it is that's going to really take advantage of your serious graphics hardware. Namely, large resolutions. At normal resolutions of 1680 x 1050 and 1920 x 1080, PC games are still quite a way from really pushing top-end GPUs. It's only when you jump into the upper echelons of 2560 x 1600 and above that these expensive beasts start to come into their own.

Should you go out and buy a seriously expensive graphics setup? If you have a hankering for some crazy screen real estate, then hell yeah. If not, you might not need to part with £1,200 for a pair of GTX 590s.

There's still a wealth of graphical options lower down the scale, though. We've benched 20 different setups to give you the lowdown. Turn the page to find out what we've discovered in testing.

10 top gaming graphics cards compared

1. Asus GeForce GTX 590

So here it is, the pinnacle of Nvidia engineering and the fastest single-PCB graphics card the Santa Clara company has ever produced. It's quite spectacular – we're talking about one board with two of the most powerful GPUs Nvidia's ever made jammed on it.

Obviously, compromises have had to be made with the chips to get them to run happily on a single card, and to stay within the 375W limit that comes from a pair of 8-pin PCI-e connectors and the motherboard PCI-e bus.

Read the full Asus GeForce GTX 590 review

2. Asus Radeon HD 6990

Pipping Nvidia to the post came AMD, releasing its dual-GPU, top-end graphics card a few weeks earlier than the competition. As with Nvidia, the HD 6990 is the fastest single-PCB graphics card the Texan brand has ever made, with performance to match.

It almost looks as though both cards have come from the same design school, with so many similarities between them it borders on the suspicious. Like the GTX 590, the HD 6990 houses two of the cores from its current-fastest single- GPU card – the HD 6970.

The Cayman XT from that card has been doubled up and redubbed 'Antilles', and comes with slower clock, shader and memory speeds out of the box.

Read the full Asus Radeon HD 6990 review

3 Zotac GeForce GTX 580

Nvidia's GTX 580 is the best single-GPU card on the market. There's no caveat – across our entire benchmarking suite, it held sway over all others. It's down to the exemplary GF110 GPU sitting at the heart of it.

Fermi has proven to be the best GPU architecture the green company has ever manufactured, and with this second generation, the GTX 5xx series, it's gone from strength to strength.

The first Fermi, the GTX 480, was the top card of its day, but wasn't without issues. The problem was the amount of power it needed and the quantity of heat it generated.

Read the full Zotac GeForce GTX 580 review

4. HIS Radeon HD 6970

The fastest single-GPU model out of the AMD stable is a tough one to recommend. As the first of its Northern Island codenamed graphics cards, it's certainly the quickest AMD has come out with.

The big problem, though, is the second-tier card released alongside it – the HD 6950. But we'll come to that later.

The Cayman XT at the heart of the HD 6970 is essentially the same GPU as paired in the Antilles chip of the dual-GPU HD 6990. It's clocked faster than the stock specs of the Antilles, though – through the shaky-ground magic of the Antilles Unlocking Switch (also known as the Screw Your Warranty Switch) the HD 6990 can run with its GPUs at the same speed as the HD 6970.

Read the full HIS Radeon HD 6970 review

5. Zotac GeForce GTX 570

The GTX 570 represents the second tier of Nvidia's Fermi architecture lineup. Coming in a little cheaper than AMD's top single card, the HD 6970, it's an impressive beast. As much as the GTX 580 was the GTX 480 engineered properly, the GTX 570 is a lot more than just the GTX 470 done right.

It has more in common with the GTX 480, coming as it does with almost the same basic GPU layout. The GTX 570 has the same 480 CUDA cores and 60 texture units, but drops eight ROPs compared with the GTX 480's 48. The 570 also has a smaller frame buffer, at 1,280MB against the previous generation card's 1,536MB.

Read the full Zotac GeForce GTX 570 review

6. Sapphire Radeon HD 6950

We understand that, in pure performance terms, the HD 6990 and HD 6970 have the HD 6950 beaten. But in our eyes, this is the finest graphics card AMD has produced in a long time. In the sub-£200 market, it's as good as you're going to get, and it's a card that offers much more than any of the opposition.

In its plain state, the HD 6950 comes close to the HD 6970 performance-wise, and to the GTX 570 too. The 2GB GDDR5 frame buffer means it can keep pace with Nvidia's second-tier Fermi card, although for the most part it does lag behind slightly. It does the same with the HD 6970, but that's to be expected from a card that's much, much cheaper.

Read the full Sapphire Radeon HD 6950 review

7. MSI GeForce GTX 560 Ti

Just as the GTX 570 was replacing the GTX 480, leaving the GTX 580 out on its own at the head of the single-GPU pack, here we've got the GTX 560 Ti retiring the GTX 470, not the GTX 460.

Yes, that's a lot of numbers, but it does make sense, especially when you look at the performance of those GTX 460s, even in today's market. It would make no sense for Nvidia to kill off one of the best cards it's launched since the halcyon days of the 8800 GT.

But it's not quite as simple as just being another transistor-level tweak of the GTX 470 design. The GTX 560 Ti is more of a halfway house between the GTX 470 and GTX 460.

Read the full MSI GeForce GTX 560 Ti review

8. XFX Radeon HD 6870

Once you drop below the price point of the competing HD 6950 and GTX 560 Ti, you get into the value segment. That's not value as in 'value' toilet paper – it's all about the price, and not necessarily the performance.

With the HD 6870, you'd only be looking at this card if you really couldn't afford to spend a pound more than its listed price. That's not to say it's a bad card. It compares favourably with the marginally more expensive GTX 560 Ti, though when we first saw it at launch, the combination of relatively high price point (£200 at release) and older technology meant it got a fairly poor reception.

Read the full XFX Radeon HD 6870 Ti review

9. EVGA GTX 460 1GB

Over the last 12 months, if there's been one card you could call the a hero, it's most definitely the GTX 460 1GB. Nvidia has had a long history of creating superb mid-range cards, from the 6600 GT to the incredible 8800 GT, and the GTX 460 can sit happily among such illustrious forebears.

Based on the same Fermi architecture as the hot-and-heavy GTX 480, the GTX 460 carries the torch into the lower end of the price spectrum. At launch it was priced at £200, the same as the GTX 560 Ti when it was first released. However, thanks to the passage of time, you can now pick up an overclocked 1GB version for as little as £120.

Read the full EVGA GTX 460 1GB review

10. Sapphire Radeon HD 6850

Thanks to the shift in pricing that followed the the AMD HD 68xx series, the HD 6850 has come into its own more in the intervening months. Now available for as little as £132, it's about the same price as the GTX 460, and just about had it pipped to the post in most of our benchmarking suite.

Like the HD 6870 before it, it's based on the same Barts GPU, a reworking of the GPU at the heart of AMD's last generation of DirectX 11 graphics cards, the HD 58xx series. In real terms, though, this card was brought in to replace the excellent HD 5770, despite what the naming structure may have you believe.

Read the full Sapphire Radeon HD 6850 review

Bench analysis: Making sense of an awful lot of numbers

Because of the awesome amount of graphical power at our fingertips, with this latest batch of DirectX 11 cards all of the benchmarks below represent their performance at the very highest single-screen resolution possible; 2,560x1,600.

With the really intensive DirectX 11 benchmarks you can see a clear divide between the top contenders for the top GPU crown and the alsorans. Heaven 2.5 though shows a nice, smooth gradient as you move down the power order, but you can see with the incredibly demanding Metro 2033 that there is a definite step where the cards simply cannot provide enough graphical grunt to run the game at such high resolutions.

Even the GTX 570 fails to get into double figures in SLI. It's here the HD 6950 shows its muscles, keeping pace with the GTX 580 in this most demanding of benchmarks.

The multi-GPU performance is interesting too, with only the synthetic Heaven 2.5 test showing consistent GPU scaling with up to four GPUs. And talking of multi-GPU the power of twin budget cards, like the GTX 460 and HD 6850, proves a match for single, pricier options.

And the winners are...

Nvidia GeForce GTX 580 AMD Radeon HD 6950

With so much choice in the graphics card arena it can be tough to make a definitive call on what the best cards are for what purposes. There are some standout performers in the test and some weaker options which you can almost discount immediately.

At the top end though things are incredibly tight. In terms of pure, out of the box performance you'd have to side with the GTX 590 as the three-screen hero single card. But with the frankly obscene performance of the GTX 580 in SLI, and despite the price-tag, we'd have to go for the separate, faster cards in SLI over the single-PCB, multi-GPU GTX 590.

AMD's HD 6990 comes in behind them, though with the ability to power up to five screens through its EyeFinity tech. With the, admittedly warranty-voiding, Antilles Unlocking Switch making overclocking the card as easy as flicking a switch (and hoping your PSU doesn't fall over) the HD 6990 actually does have a lot going for it.

Neither multi-GPU card though looks good in quad-GPU trim. Neither camp's drivers are really up to snuff, with only the Heaven 2.5 benchmark actually scaling properly. And driver problems, even with two GPUs, can mean a wait, post game release, for your cards to be supported.

Which makes a single GTX 580 a fantastic proposition on its own, it's fast, though does lack proper multi-monitor support. So Nvidia's top card becomes a double winner in single and SLI flavours.

AMD's top single-GPU card, the HD 6970, while being a decent card in of itself, is actually hampered by its own brethren. The HD 6950, in its default, vanilla flavour is almost as fast as the top AMD single-GPU card and for almost £100 less. Combined with the outstanding performance of the GTX 570 makes the 6970 impossible to recommend.

In the mid-range there really is only one winner, and that's the HD 6950. On its own it's a great performing card, keeping pace with both the HD 6970 and the GTX 570. If you factor in the unlocking potential and subsequent overclocking chops that turns it into a monster of a card for under £200.

You can also pick up two of them for the same price as a single GTX 580 and give Nvidia's top GPU a darned good thrashing while you're at it.

So if you're looking at a single, high-resolution monitor then Nvidia's GTX 580 will come without any driver hassle and keep your games looking great. For a serious multi-screen setup it has to be an AMD card, and personally I'd opt for a pair of unlocked HD 6950s, though the HD 6990, PSU-willing, will deliver incredible performance.

Funny isn't it, how CPUs get all the cooling attention? There they are, sitting in blissful tranquillity beneath 8-inch heatsinks without a care in the world, while their GPU friends sat next to them sweat their bits off, desperately trying to keep up with Metro 2033.

Graphics card upgrade guide

The graphics card is the racehorse of your PC components stable. It's a high-value add-in board that's traditionally done one thing and one thing only: let you play games with all the latest graphical wizardry.

Increasingly though, the graphics card is becoming far more than just a gamer's luxury. With architecture improving year on year, 3D graphics aren't the only thing your discrete GPU can do.

It can now be used to enhance your web browsing experience and enjoyment of high-definition media, let you explore your creative side with enhancements for productivity software and even help cure terminal diseases through projects like Folding@home.

So as well as producing some stunning visuals, your graphics card can also help save lives.

In the last decade, GPUs have been following in the footsteps of the CPU market, with increased core and thread counts. Speed in MHz or GHz is no longer the only measure of a chip's power, whether it's a GPU or a CPU.

What counts now is the number of cores, and how much data the chip can process at any one time. In CPU terms, the maximum on the desktop is six cores and 12 threads, and a full-fat 12 cores in the server space.

The top-end Nvidia GPU – the GeForce GTX 580 – has 512 CUDA (Compute Unified Device Architecture) cores. Meanwhile, the AMD Radeon HD 6970 has 1,536 shader processors, all of which are simple processors capable of taking on tasks such as video encoding, where some simple parallel processing is needed to enhance speed.

Nvidia was first to take this on with its CUDA cores, which let programmers write code in industry-standard languages such as C++. This code is run using all the shader processors (or CUDA cores) in Nvidia's GTX 8-series onwards.

Microsoft's latest update to its graphics API – DirectX 11, has done a similar thing with its Compute feature, which enables general purpose applications to run through a DirectX-capable GPU rather than taxing the processor.

If the GPU is becoming ever more powerful, why is there such doom and gloom around the discrete graphics card market? According to Intel and AMD, the future is fusion.

Are integrated graphics the next big step in the great graphics war?

There are many reasons to be upbeat about the future of discrete graphics cards. There isn't going to be a new games console release for another couple of years now, and the mid-range cards of today are far superior to anything the Xbox 360 or PS3 contain, so the PC is the platform to go for if you want to see the top releases looking their best.

Integrated graphics (the graphics processing power that traditionally comes with your processor chipset combination of CPU and motherboard) are catching up, though. They're changing as well – moving from the motherboard onto the CPU. All the big boys are getting involved.

First there was Intel and its Arrandale processors, which packaged a GPU and CPU on the same chip. Then came the company's Sandy Bridge, with its fully integrated processor graphics.

AMD has recently released its first Fusion board to the world, housing a tiny CPU and GPU setup – the first new CPU architecture we've seen from the company in years.

At this year's Consumer Electronics Show, held in Las Vegas in January, Nvidia announced Project Denver, its own collaboration with ARM to create a powerful desktop CPU with Nvidia's GPU architecture built right in. This may not shake up the high end of the discrete graphics market – after all, the latest 3D games are still going to need a power-hungry graphics card sitting in that PCI Express slot – but the value end of the market is going to change.

Processor graphics will be more than capable of coping with high definition video, encoding and casual gaming, so why would you choose to spend £50 on a separate card that will do the same job?

That said, times move quickly in the graphics card market, and tomorrow's £50 GPU will make processor graphics weep. AMD and Nvidia will be launching a slew of low-end cards to prop up their latest HD 6xxx and 5xx series respectively.

The high end will probably see the biggest battle. Nvidia's GTX 580 is currently top dog, but AMD is due to release its dual-GPU Antilles behemoth in the next few months, possibly at the CeBIT show in Germany. Details are scarce, but if AMD follows the example set by its previous dual-GPU releases, you can expect two Cayman Pro GPUs wired into one slice of AMD-red PCB.

Those are the chips powering the superlative Radeon HD 6950, and will make for one hell of a card.

Don't expect Nvidia to be keeping quiet, though. When we spoke with Tom Petersen, the company's Director of Technical Marketing, at the secretive preview of the GTX 580 last year, we asked if he expected to see a dual-GPU Fermi card any time soon.

He explained that, now the thermal issues seen in the first high-end Fermi card (the GTX 480) had been solved in the GTX 580 and GTX 570, there really wasn't a barrier any more.

So pretend to be surprised when Nvidia announces a GTX 595 just as AMD starts to get excited about its Antilles card.

Three top graphics card choices

Zotac GTX 580 AMP
Performance
Price: £480
Info: www.zotac.com

On the basis that money is no object in your search for graphics perfection, you'll be hard-pressed to find a more impressive pixel-pusher than Zotac's recently launched, overclocked GTX 580 AMP.

This souped-up version of Nvidia's GPU is the fastest thing on two PCIe power cables. Based almost entirely on the first Fermi card, the GTX 480, it's undoubtedly what the brand wanted to release originally.

The GTX 480 had a cutdown version of its low-yielding GF 100, with one streaming microprocessor turned off. That meant a lowly 480 CUDA cores instead of the full 512 we were expecting.

The GTX 580 came out of nowhere last year with the full complement, plus nifty power and cooling advances. So it's quicker, cooler, quieter and far more power-efficient. In short, it's just better.

The AMP version is ever so slightly overclocked, but will also give a little more headroom should you wish to push it further. At these speeds though, you won't need to for a few years at least.

Verdict: 4.5/5

Asus GTX 460 Top 768MB
Budget
Price: £130
Info: www.asus.com

We've already seen the stock GTX 460 768MB, and now it's the turn of the overclocked cards in the shape of Asus' GTX 460 768MB TOP edition.

The GTX 460 looks set to be the most successful iteration of the Fermi architecture that Nvidia has released to date. That's mainly thanks to a redesigned chip, still based on the same technology that made the GTX 480 such a blisteringly fast, and hot, card.

This new GF104 GPU is a far more streamlined chip compared to the fairly bestial GF100.

It still has the same basic premise running through it, but more cores have been squeezed into fewer streaming microprocessors (SMs) and more texture and special funtion units have been jammed in there too.

Verdict: 3.5/5

Read the full Asus GTX 460 Top 768MB review

Sapphire Radeon HD 6950
All round
Price: £228
Info: www.sapphiretech.com

AMD's Radeon HD 6950 is the must-have card of the moment, its price tag hitting the sweet spot in terms of cost/performance ratios.

The card is based on AMD's latest Cayman GPU, and with its redesigned approach to tessellation, offers some serious competition for the far more expensive GTX 570. It's also the only card under £250 that can take on the tessellation-heavy Metro 2033 at an eye-bleeding 2,560 x 1,600 resolution and still come out smiling.

The Cayman GPU's twin tessellation engines make the HD 6950 an excellent DirectX 11 card. On the DX 10 benchmarks it loses ground to the new GTX 560 Ti from Nvidia, but the AMD card has the better scores in the newest titles and comes with an impressive trick up its sleeve.

With a simple BIOS flash you can upgrade your HD 6950 and turn it into an HD 6970 – a £270 card – for free. That's not an overclock; it's unlocking dormant parts of the GPU and setting them free. That makes it the card of choice right now.

Verdict: 5/5

How to overclock your graphics card

With any sort of gaming, the biggest performance boost you can get is by making a change to your graphics card. However, that doesn't mean you need to kick your existing card into touch and dip your hand into your wallet to fund a quicker one.

Overclocking your graphics card is the quickest, cheapest and (most likely) easiest way to wring a little extra performance out of your current GPU. It was originally seen as a way to make older components competitive for longer, whereas it's now an expected consideration when you're choosing any new graphics card.

It's also not as dangerous as it once was; it's fairly difficult to brick a GPU with the simple overclocking you can do in Windows.

Generally speaking, you'll be lucky to get a 10 per cent increase in performance, but those precious few extra frames per second can mean the difference between almost unplayable choppiness and smooth, slick action.

To get started, all you need is a software tool like MSI's Afterburner, a graphics stress test such as Heaven 2.1 and a bit of patience. With a few cards, including the Radeon HD 6950 we've already talked about, it's possible to make some BIOS tweaks to garner a boost in performance.

The HD 6950, for example, runs the same GPU as the HD 6970. All that's different is that the chip was chosen to go in the lower-price card and has some features turned off. Normally, these changes are hard-wired out of the GPU, but AMD decided to implement the castration with software pincers only, making for easy reversal.

This is an extremely rare state of affairs though, and we haven't seen similar things happen in the GPU game for years.

Another way to improve things is by taking the more heavy-handed approach of adding a second GPU to the equation. Multi-GPU graphics have come on leaps and bounds in the last couple of years. Now we're seeing performance with a second card hit the 2x boost we'd always hoped for – and even more in some cases.

The difficulty is with the motherboard. You need one that supports multi-GPU and, unfortunately, that's a tougher, more expensive journey if you want to go down the Nvidia SLI route.

However, most boards will support AMD's CrossFire technology as standard on extra PCIe lanes, so that's becoming far more popular.

Easy overclocking

1. The setup

MSI's Afterburner has an impressive hardware monitoring display. Detach this from the main console and it will give you a better view.

Start up the Heaven 2.1 benchmark in Windowed mode, at a lower resolution than your desktop, but with high settings. This will stress your GPU as you apply the overclock to help you judge stability. If there are no graphical glitches on-screen, the OC is stable.

2. The push

You'll need to approach the memory and GPU overclocking of your card separately to judge the maximum overclock on each. Start with the memory and push the slider up in 5MHz increments.

After each step, watch the Heaven benchmark closely for artefacts. Keep doing this until you see problems, then step back the OC until they go. Reset the memory slider. Repeat for the processor clock.

3. The stress

Once you've discovered the top overclock for the GPU and memory clocks, set both sliders to their maximums and watch the benchmark. You'll have to knock back the relevant sliders to create a stable OC.

Pixel-sized artefacts mean there's a memory problem, and bright colour blocks represent GPU issues. When there are no artefacts, your OC is stable. Now stress test it in fullscreen mode to make sure.