Overview

Not to be outdone by new graphics releases AMD is going back to the APU drawing board with the new AMD A8-3870K processor.

While its flagship FX processors are failing to shine it does seem a little on unfair on AMD that at the other end of the market it has a chip which really ought be cleaning up.

Its Llano Fusion APUs, which combine a multi-core CPU and a Radeon graphics part on one die, are actually rather good.

They may be based on an older processor architecture, but quad core examples like this A8-3870 hold their own against Intel's similarly priced dual core Core i3s in CPU benchmarks.

When it comes to 3D tasks like gaming, the on board HD6550D is simply in a different league to Intel's laggardly HD Graphics 3000 cores.

How could they get better? We'd like a chunky price drop, but that's not going to happen.

Instead, AMD has introduced 'Black Editions' of the chips, recognisable by their 'K' suffix. Just like Intel chips with a K moniker, they come with an unlocked multiplier which makes them more suitable for overclocking.

Now, you might think that anyone who wants to overclock a chip will be looking at something a little more high-end than these very basic processors, but the entire raison d'etre for overclocking is to turn something cheap into an unexpected powerhouse.

An unlocked, gaming-capable hybrid chip for just over £100 sounds like the stuff hardware hackers' dreams are made of. But does it deliver on that promise?

Benchmarks

As expected, even with a 100MHz increase over the older AMD A8-3850, at stock speeds the A8-3870K doesn't distinguish itself fully against the Core i3.

Put the two chips head to head solus and it has the edge, but if you're using external graphics the Core i3 still pulls away in any game that's not massively multithreaded (Shogun 2, tellingly, is).

And what of the overclocking? It's actually a bit disappointing really. At the speeds we achieved using fairly basic equipment, the difference is often not enough to be worth even trying.

DirectX 11 gaming performance

DirectX 11 tessellation performance

CPU rendering performance

CPU video encode performance

Verdict

A stock A8-3870K ships with its CPU set at 3GHz and the graphics core clocked at 800MHz. Our attempts to increase these weren't exactly stellar - it ramped straight up to 3.4GHz/800MHz, but any attempts to go further were unstable enough to be frustrating.

With more fine-tuning and a really good cooler, we've seen reports of much higher clockspeeds, but we're not overly sure it's ultimately worth it for the performance returns.

Plus, it's unlikely that motherboard manufacturers are going to get behind this platform with high end boards full of good, stable overclocking features – because potential purchasers just aren't going to spend the extra they'll cost to make.

Still, since the A8-3870K is already available for less than its non-tweakable predecessor, the A8-3850, the unlocked multiplier is more of a pleasant extra to have.

So even if our dream of using one to outperform a Core i7 isn't going to happen, the A8-3870K is still a great little chip.

The CPU part of Llano was never in much doubt. Based on the same architecture as Athlon II and Phenom II with a die shrink, it's old but still capable.

Because it has four native cores it outperforms Core i3 in tasks like media rendering and so on, although Intel's computational engine is still superior in games – even if its graphics lags far behind.

For a workplace PC, then, the A8-3870K is an exceptional choice. It'll beat a similarly priced Core i3 system at everything.

It gets a bit more complicated if you're after a budget system that can game at 1080p – to use as a media centre, for example.

Even overclocked, the A8-3870K alone isn't an alternative to discrete graphics – although it does come very close if you're willing to sacrifice graphics quality. Battlefield 3, for example, is this far off being playable at 1920x1080 with low image quality settings.

Which are still bloody good looking.

Skyrim at similar settings runs like a dream.

If you're thinking of adding in a mid-range card like the HD 6870 or even something as powerful as an Nvidia GeForce GTX 560Ti, though, Intel just wins through – although it's a draw in CPU limited games like Shogun 2 and Skyrim.

AMD's architecture does have the curious disadvantage of performing slower in CPU benchmarks when discrete graphics are attached too.

But that leaves one feature that we haven't discussed yet which bring us down in AMD's favour.

That feature is Asymetrical Crossfire – the ability to use both the A8-3870K's graphics and a low power GPU, such as an AMD HD 6670, for a gaming experience that's capable of playing most games at 1080p with medium settings.

That's a hell of a thing in AMD's favour.

Enough that while it's not going to be our chip of choice for a workstation or enthusiast games rig, if you want a small, low cost PC which is capable of occasional games at console quality, it's a steal.

We liked

Where it comes into its own is as a flexible platform for a quad core office machine or a flexible and capable gamer with Asymmetric Crossfire.

We disliked

The potential to overclock is a bonus, but really not enough by itself to make us favour this chip over any other quad core Llano.

Final word

Not the budget buster we hoped for, but further's our faith in AMD to make Fusion the low cost platform of choice.

Overview

This is a bit more within the reach of us mere mortals - a Sandy Bridge E CPU that doesn't cost the same amount as some full PCs.

Intel, with even less of a fanfare than it made for the muted launch of the Sandy Bridge E platform, has now unleashed the quad-core iteration of its top-end chips, the Intel Core i7 3820, priced at £260 in the UK and $286 in the US.

But, to be honest, we're not entirely sure why this CPU even exists.

Before this, the cheapest Sandy Bridge E processor was the Intel Core i7 3930K, costing £450-500 in the UK and $650-700 in the US. The crazy expensive £800-900/$1,100-1,200 Intel Core i7 3960X takes its place at the top of the Intel CPU tech tree. Both those CPUs are full-blooded hex-core chips with Hyper-Threading tech enabling twelve threads of processing power.

Except that they're not really hex-core chips at all. The Intel Core i7 3930K and i7 3960X are actually octo-core server chips with a couple of cores disabled to fit into the desktop segment.

This, coupled with the huge price tag, has been our major bugbear with the latest of Intel's top-end chips. We now have eight-core CPUs potentially running in our home rigs, but unfortunately we're not allowed access to the disabled silicon sitting inside our incredibly expensive chips.

A cynic would say Sandy Bridge E represents a great opportunity for Intel to sell broken Xeons to the desktop computing crowd for a tidy profit.

Intel obviously refutes this, making the case that it believes the consumer would rather have higher clockspeeds over those extra cores, and it needed to sacrifice two cores to hit the 3.3GHz speed of the top chip.

Before with the imminent release of its E5 Xeons, Intel has managed to hit 3.1GHz with a Sandy Bridge E-based server chip designed for the LGA 2011 socket.

So what about the Intel Core i7 3820? Is this huge processor the same, is it really carrying around four dead cores in its innards? Little ghoulish really...

Chip architecture

Thankfully, the Intel Core i7 3820 is not just another bust eight-core chip, this time with half its goodness turned off.

The processor is actually a different die entirely, compared with bigger brothers the i7 3930K and i7 3960X. And they most definitely are bigger, by around 1 billion of the wee 32nm transistors.

That gives the Intel Core i7 3820 1.27 billion in its make up. But that means it's still bigger than the previous generation's Gulftown hex-core CPUs, the entire Sandy Bridge lineup and even AMD's chunky quad-module Bulldozer chips.

So the CPU die is rattling around in that massive LGA 2011 package, especially considering it's over 100mm2 smaller than the die size of the other two Sandy Bridge E processors.

The Intel Core i7 3820's cache is also inevitably smaller than its Sandy Bridge E compadres, but compared with the Intel Core i7 3930K's 12MB you're only losing 2MB of L3 cache.

At 10MB, the Intel Core i7 3820 is a little more cache-heavy than the 8MB-toting Intel Core i7 2600K and Intel Core i7 2700K chips.

There is one other key difference between this straight Core i7 3820 and the X and K suffixed CPUs from Intel, and that's the unlocked nature of the other chips.

This processor has a partially locked multiplier, which limits it to 43x compared to the 57x multiplier offered by both the other Sandy Bridge E chips and the top-end K-series i7 and i5 CPUs of the plain Sandy Bridge range.

That ought to seriously limit its overclocking potential, but that's not necessarily the case.

Benchmarks

The Core i7 3820 actually posts some decent scores but it's the Core i7 2700K that makes the best impression thanks to some stellar comparative benchmarks.

In single-threaded applications the straight Sandy Bridge architecture has the edge in our test, showing why the gaming performance is higher as well.

CPU performance

Video encoding performance

Gaming performance

Performance

The Intel Core i7 3820 actually posts some decent scores in our benchmark tests, but it's the Intel Core i7 2700K that makes the best impression, thanks to some stellar comparative benchmarks.

In single-threaded applications, the straight Sandy Bridge architecture has the edge in our test, showing why the gaming performance is higher as well.

So the key battle for the Intel Core i7 3820 is the head-to-head with the Core i7 2700K, the top Sandy Bridge CPU. At £260, they're both priced in the same ballpark (or stadium, if you prefer), and at 3.6GHz vs the 2700K's 3.5GHz they're both around the same sort of clockspeed.

Predictably things are pretty close in terms of raw performance. Importantly, though, not identical.

The Intel Core i7 2700K still maintains a lead at stock clocks over the Sandy Bridge E Intel Core i7 3820. In single-threaded performance, the 2700K is actually quicker than even the i7 3830K, although despite having a higher clockspeed, the i7 3820 offers the same figures as the 3830K in single-threaded performance.

Thanks to the extra couple of cores in the mid-range Sandy Bridge E chip, it takes a lead in the multi-threaded Cinebench rendering and X264 encoding tests. The resolutely quad-core Intel Core i7 3820 lags behind the hex-core chip and, interestingly, the lower-clocked Core i7 2700K, though.

The straight Sandy Bridge's gaming pedigree comes to the fore, though, when we start throwing the World in Conflict and Shogun 2 benchmarks at the different chips.

When the graphics card is taken out of the equation, in the WiC and Shogun 2 CPU DX9 CPU tests, the straight-line performance of the 2700K puts it ahead of both the Sandy Bridge E chips. And that's even true when we beat the i7 3820 with the overclocking stick, too.

But the overclocking performance of the i7 3820 is quite interesting. Despite being only partially unlocked, meaning you hit the 4.3GHz limit straight away, you can push it further given the right motherboard.

Thanks to the BCLK strap, the Sandy Bridge E chips are able to withstand a good tweaking of the base clock. That's still limited to being 100MHz, 125MHz, 166MHz and 250MHz, but it does give you some leeway.

With a decent motherboard, you should be able to get up to the same sort of overclocking performance - around 4.7GHz - as the i7 2700K.

That means that the partially locked multiplier really doesn't impact too much on the overclocking prowess of this new Sandy Bridge E chip.

But the reliance on a decent motherboard is one thing that goes against the i7 3820, especially in a straight fight between it and the 2700K. And that's all down to platform costs.

Verdict

So the Intel Core i7 3820 performs at around the same sort of levels as the top-end standard Sandy Bridge CPUs.

We have to say, that's a bit of a disappointment.

This is supposed to be the serious enthusiast processors for Intel's desktop faithful, but the more mainstream, soon-to-be-replaced Sandy Bridge setup is just as good at the quad-core level.

On core-for-core performance, you're not getting much extra for your Sandy Bridge E money. Of course there are the extra PCIe lanes, which are useful if you're rocking more than two graphics cards, and the extra bandwidth offered by the new PCIe 3.0 tech, but the real value of that is still rather ephemeral at best.

There are also the extra two channels of DDR3 memory in the Sandy Bridge E/X79 combo, but again for most of us the benefits of that extra bandwidth is rather difficult to gauge.

In the overclocking stakes, the partially locked multiplier doesn't do the Intel Core i7 3820 a lot of harm, and you ought to be able to hit the same sort of OC numbers as the 2700K. But not best it.

You need a decent motherboard to facilitate that, and that means spending nigh-on £200 on your board. Combined with the £260-odd you're paying for the chip, that still makes the Sandy Bridge E platform too expensive compared to the far cheaper base Sandy Bridge.

There will be instances where the Intel Core i7 3820 and X79 combo will be beneficial, for the offline 3D rendering workstation or other serious productivity tasks that don't necessarily require heavy multi-threading. But that's a niche of an already minuscule niche.

There's only one PCIe 3.0 graphics card out at the moment and the benefits that interface alone gives you are limited. And the extra memory bandwidth is only useful for a tiny minority of the world's populace.

The fast majority of us, and all PC gamers, can already get enough all-round performance out of a standard Sandy Bridge setup.

The quad-core Intel Core i7 2700K or 2600K are still the CPUs we'd recommend.

We had hoped for a more impressive, compelling outing for the quad-core version of Sandy Bridge E. Sadly Intel already has the best-performing quad-core chip on the market, and as a whole platform the i7 2700K and Z68 chipset look like quite the bargain.

The low price of the Intel Core i7 3820 is pleasing, though, and does mean that if you want to build the basis for an excellent workstation you can pick up a decently-priced CPU now and then splash the serious cash on a hex-core CPU if you can afford, or need, the extra cores at a later date.

The socket will continue with 22nm Ivy Bridge E chips towards the end of the year, so there is at least some onward progression for the X79 chipset. Even if Ivy Bridge on the Z77 platform will put the boot into the i7 3820 in a couple of months time.

We liked

The fact that Intel has put the chip out for the same price as the top-end Sandy Bridge i7 2700K is impressive, and thoroughly welcome.

We're also pleased that the BCLK strap shenanigans help to render the partially locked multiplier a non-issue for the majority of overclockers.

We disliked

Sadly there's no real compelling reason to choose this CPU over the existing Sandy Bridge setups. We had hoped for at least a little bit of a performance lead for the extra cash you'd have to stump up for this platform as a whole, but sadly for the Intel Core i7 3820, the i7 2700K has it licked in most benchmarks.

Final verdict

It's a decent chip, offering the extra bandwidth of the Sandy Bridge E platform for straight Sandy Bridge prices. But the vast majority of us don't need that bandwidth.

Overview

The Intel Core i7 3930K is the cheaper alternative to the top-end Sandy Bridge E processor. But then many second hand cars are cheaper than the top-end Sandy Bridge E processor.

At nearly £500 it's still an expensive CPU, but currently it's the only LGA 2011 processor worth a look.

The Intel Core i7 3960X, that top-end Sandy Bridge E, is a positively preposterous processor. For in excess of £800, you get a chip that's not substantially quicker than Intel's own Core i7 980X of two years ago.

Not at stock clocks, at least. Admittedly, the Core i7 3960X does overclock very nicely indeed and in doing so opens up a gap from ye olde 980X.

But we've reviewed the 3960X elsewhere and deemed it disappointing, moderately sinister (it's prima facie evidence of Intel carpet bagging in response to AMD's failure to bring out a really quick chip) and largely irrelevant to human existence.

This then is the Intel Core i7 3930K and it's not the same chip. Not precisely, anyway.

Benchmarks

You can see below just how small a gap, in performance terms, there is between the i7 3960X and the i7 3930K.

It's also interesting just how close the latest traditional Sandy Bridge chip, the Intel Core i7 2700K, is in general performance terms too, especially in gaming and single-threaded speeds.

3D rendering performance

Video encoding performance

CPU gaming performance

Verdict

Yes, this Core i7 3930K is based on the same, quite colossal 2.27 billion transistor chip, known as Sandy Bridge E. So, it shares most of the same specifications as the Core i7 3960X. That starts with six cores in Intel's latest Sandy Bridge-generation idiom.

Next up we have a new memory controller no fewer than four (yes four, count 'em) channels.

Intel's previous high end processors sported a triple channel memory controller. Even that looked like overkill for a desktop processor. Four channels is getting silly and merely serves to underline the real reason the new Core i7 exists.

It's a thinly disguised server chip.

Whatever the merits of the quad channel controller, it forces the use of a new socket, the monumental LGA 2011. If nothing else, you are getting a satisfyingly massive chip for your money.

What's more, thanks to the 'K' on the end of Intel Core i7 3930K, this lower priced alternative to the Intel Core i7 3960X gets the full unlocked treatment and also benefits from the newly introduced CPU strap, the better to make overclocking a bit more flexible.

At this stage, you may be wondering what on earth the difference actually is.

The answer is twofold. Firstly it's clocked infinitesimally lower – 3.2GHz instead of 3.6GHz, along with a commensurate climb down in the maximum Turbo speed to 3.8GHz. The other bit is less L3 cache memory to the tune of 3MB. The 3930K makes do with 12MB.

Frankly, both of these compromises in the name of cost savings fall into the 'who cares?' category.

In terms of desktop computing, neither is going to make a blind bit of difference to experience your PC delivers. That's reflected in benchmarks that are barely any slower. Even better, the 3930K overclocks very nearly as well as the 3960X. Again the gap is just 100MHz, 4.8GHz on airs plays 4.9GHz.

So, here's the best bit. The 3930K costs over £300 less.

OK, £500 is still a big ask. But the difference in price alone is enough to buy a half decent desktop PC or a cheap laptop.

The point, then, is that this cheaper Sandy Bridge E gives you everything the top chip delivers for a lot less money.

There's absolutely no reason to spend. We're not completely convinced even this truly means the Intel Core i7 3930K is good value for money, but it's still a very fast processor and the chip we'd buy if we had a big budget.

We liked:

The fact that you're getting almost the same sort of performance out of this £500 Sandy Bridge E as the £850 Core i7 3960X makes it a more intriguing prospect.

The other boon of the Core i7 3930K is that it's got the same huge amount of overclocking headroom sitting in that CPU package. You can reach extraordinary speeds out of this architecture.

We disliked:

As it's still based on the same architecture as the Core i7 3960X it's also got the same flaws, namely that it's actually an eight-core CPU with a couple of cores turned off.

And despite the fact it's over £300 cheaper than the top end chip, the Core i7 3930K is still an incredibly pricey processor.

Final word:

Much better value than the 3960X. The obvious choice if you're thinking of the LGA 2011 platform.

Overview

The vanilla Sandy Bridge lineup has a new chip in the Intel Core i7 2700K . As well as populating a new socket with the Sandy Bridge E Intel Core i7 3960X, it has seen fit to give us another brilliant processor, but what makes it different to the Intel Core i7 2600K?

Short answer: not a lot.

Not that long ago, things moved fast in the CPU industry.

Between 2005 and 2010, we progressed from piffling little 100 million transistor single-core desktop CPUs to six-core, billion-trannie mofos capable of implausible feats of number crunching.

During that period, the technology used to knock out computer chips has powered ahead, too.

We've transitioned through 90nm, 65nm, 45nm and 32nm silicon, with 22nm CPUs just around the corner. Along the way, we've picked up all manner of fabulous new chip tech, including SOI, strained silicon and latterly tri-gate and 3D transistors.

It was relentless stuff.

Indeed, Intel has committed itself to rolling out either a new CPU architecture or a new production process every single year.

And yet here we are at the apex of 2011 and 2012 wondering where it's all gone wrong. The evidence comes in the form of both of Intel's new PC processors.

We've proselytised plenty enough on the subject of the new Core i7-3960X. Suffice to say we view its hidden cores as evidence Intel has taken its foot off the gas.

Benchmarks

Intel's new top-end Sandy Bridge is quite some chip. Sadly it was quite some chip when it was the slightly slower-clocked Intel Core i7 2600K. Not a lot has changed apart from the price.

It comfortably hoses the best that AMD has to offer and in the grand scheme of things isn't even the best that Intel can chuck into a desktop motherboard.

Multi-threaded CPU performance

Video encoding performance

CPU gaming performance

Verdict

This new chip is the Core i7-2700K. In many ways it's a much more interesting processor than the Intel Core i7 3960X.

We don't mean that in a technological sense. Deep down, the 2700K is nothing new.

It's the same four-core Sandy Bridge generation die seen in several existing chips, including the Intel Core i5-2500K and Core i7-2600K. The number of four-core Sandy Bridgers is legion.

However, the 2700K is the new de facto king of Intel's line of LGA1,155 models.

For us, it's the LGA1155 socket that's really relevant to PC enthusiasts and gamers, not the highfalutin', server-derived LGA 2011 platform and its quad-channel silliness.

The 2700K, then, is the fastest chip any mere mortal is likely to run in his PC any time soon.

Unfortunately, what it ain't is a big step forward over the existing Core i7-2600K. You suspicions will first be aroused by the pathetic clockspeed bump from 3.4GHz to 3.5GHz.

That's right, it's clocked fully three per cent higher. Needless to say, at stock clocks, the 2700K achieves absolutely nothing of interest.

Yes, it's a very nice little quad-core chip. Yup, it has the edge on AMD's ostensibly eight-core flagship, the AMD FX 8150 Black Edition, across the board. And yey, it absolutely hammers said AMD chip in any benchmark that majors on per-core performance rather than multi-threaded throughput.

All of which means the 2700K's only hope of giving something we haven't already got is overclocking.

What'll she do, mister? The answer during our testing, and in the context of air cooling and a modicum of extra voltage, is 4.8GHz. A very good result, we think you'll agree. But not materially better than you can expect from most Intel Core i7 2600K processors. Again, the game doesn't move on.

That said, the hefty 1.3GHz gap between what the 2700K is nominally clocked at from the factory and what it will do with a bit of tweaking is perplexing.

Why on earth doesn't Intel give us a 4GHz chip?

We liked:

You can't argue with the fact that this is a fantastic processor. It's way in advance of what AMD's top processor can manage.

There's also a huge amount of overclocking headroom nestled in that unassuming CPU enclosure.

We disliked:

Sadly there's absolutely no difference between this chip and the cheaper Intel Core i7 2600K apart from a higher price tag and a CPU multiplier upped by one notch.

And every single 2600K out there could manage that on the reference cooler.

Final word:

A great chip, but barely any better than the 2600K. Hardly worth Intel's bother bringing it out.

Overview

When is a six-core PC processor not a six-core PC processor? When it's AMD's new FX 6100 Black Edition.

Long before AMD released its fancy new FX chips, we had a feeling a fisticuffs was brewing over the definition of what constitutes a processor core.

Now the FX has arrived and the gloves are off.

It all comes down to the radical new approach to simultaneous multi-threading (SMT) taken by AMD with the new Bulldozer architecture.

Intel's been doing SMT for a bit in the form of Hyper-Threading and fair to say Chipzilla's take can be called SMT lite.

It hasn't spent a lot of transistors in enabling each of its processors cores tow run two software threads in parallel. So a multi-threading performance boost of around 15 to 20 per cent is a pretty decent outcome.

AMD reckoned it could do better and verily cooked up the Bulldozer module. This little techno-beasty packs a pair of execution units. But much of what you'd normally consider a processor core is shared.

So has that Bulldozer module got enough power to give this six core processor a chance?

Benchmarks

The very least you'd hope from this latest 'six-core' processor from AMD is that it would beat the previous generation of hex core chips, here represented by the AMD Phenom II X6 1100T.

Sadly it's well off the pace the previous generation sets.

Most tragic of all is just how weak it is in gaming terms. The older processor has it licked.

3D Rendering performance

Video encoding performance

CPU gaming performance

Verdict

On paper it's hard to say whether a Bulldozer module it one hell of a single-core solution or truly deserves to be counted as two cores.

In practice, however, it's all too clear.

A single Bulldozer module doesn't deserve dual-core status. For proof, observe how this triple-module, supposedly six-core, AMD FX 6100 Black Edition chip compares with AMD's own Phenom II X6 1100T.

That's a true six-core chip and one that isn't exactly famed for world-beating per-core performance. Put simply, the 1100T owns and it owns in every metric save memory bandwidth.

As if that wasn't enough, comparisons with Intel's quad-core chips make the AMD FX 6100's six-core status look even sillier.

The non Hyper-Threaded Core i5-2500K mugs it in every test. Meanwhile, the Core i7-2700K is in another league.

What's more, in terms of overclocking, having one less active module compared to the FX 8150 doesn't appear to deliver much benefit. We didn't get significantly higher clocks out of the 6100.

If that makes the AMD FX 6100 sound pointless, it's not quite as bad as that.

AMD has priced it keenly. At £128, you can make an argument for it as a cheap multi-threading chip. The allure becomes even greater when you factor in the possibility that you might be able to flick a BIOS switch and enable that hidden Bulldozer module.

The same applies doubly to the dual-module FX 4100. However, there are absolutely no guarantees and it will take a few months and a bit of forum trawling before we have a feel for how successful FX buyers are finding their module-unlocking exploits.

Until then, we'll put a hold recommendation the AMD FX 6100 Black Edition.

At stock clocks and with the final module hidden, it's not terribly exciting. However, if it turns out that most of all 6100s will happily run with the final module enabled, it might just be worth a roll of the dice.

If that happens, we'll be more than happy to upgrade the 6100's status to buy.

We liked:

In multi-threaded terms the AMD FX 6100's performance isn't bad, still lagging behind the previous top hex core CPU from AMD.

We disliked:

Again it's the weak single-core performance of the Bulldozer architecture that makes for one massive technological sad-face.

We'd like to say positive things about the possibility of unlocking those dormant cores, sadly though we simply can't say how likely it is to happen.

Final word:

An interesting option, but only if the module-enabling prospects turn out good.

Overview

We've been waiting for AMD's brand new architecture for ages, and then Bulldozer turned out to be fairly underwhelming. Can the cheaper, quad-core AMD FX 4100 Black Edition rectify any of its wrongs?

Multi-threading and parallelism are all the rage. Put simply, more cores equals more fun, right?

Not, as it happens, when it comes to games, as seen in the AMD FX 8150. Despite the hype, the fact is PC games still don't scale all that well across multiple processor cores.

On the one hand, that's a disappointing state of affairs given the massively multi-core hoopla the likes of Intel have been pedalling for the last five years or so.

On the other, it also presents an opportunity for fanatical gamers to save a few quid.

After all, if it's just games that gets you going, we'd argue there's only so many cores you actually need.

Consider, therefore, the AMD FX 4100.

We can't be absolutely sure about this without official confirmation, but we reckon it's based on the very same two-billion transistor processor die as the range-topping FX 8150.

The difference is that two of the 8150's four Bulldozer modules have been nuked from orbit.

Benchmarks

The benchmarks don't make for particularly good reading for the AMD FX 4100 Black Edition.

Sure the multi-threaded application performance shows improvements over the equivalently-priced Intel Core i3 2100 Sandy Bridge CPU, but the single-core and gaming performance is rather poor.

That said the Bulldozer FX 4100 processor does overclock, which sadly the Intel chip cannot.

3D Rendering performance

Video encoding performance

Gaming performance

Verdict

What we can say for certain is that the FX 4100 is based on AMD's latest 32nm Bulldozer architecture.

It has two modules and therefore a grand total of four pseudo cores. It runs at 3.6GHz nominally, sports a 3.8GHz Turbo frequency and packs 12MB of L2 and L3 cache memory, combined.

And it's happiest in an AM3+ socket paired with DDR3 memory up to 1,866MHz.

But here's the real kicker. It's yours for just £100.

So, that's healthy clocks, four threads and AMD's very latest CPU architecture. Quite an attractive package. But not, as it turns out, terribly fit for gaming purpose.

The problem is the new AMD Bulldozer microarchitecture. If anything, its strongest suit is multi-threading. And that means you want as many of those modules switched on as possible.

What Bulldozer doesn't deliver, however, is decent per-thread performance.

For proof, first observe the single threaded benchmark results. The 3.6GHz (3.8GHz with Turbo) FX 4100 may be clocked significantly higher the similarly priced 3.1GHz Intel Core i3-2100, but the latter is absolutely off the map in terms of single or few-threaded performance.

The FX 4100 needs three minutes and 44 seconds to churn its way through Cinebench R10 in single-threaded mode. The i3-2100 needs just two minutes and 51 seconds.

Of course, what really matters is actual game performance and here Bulldozer's per-thread gutlessness duly rears its head.

The 67 frames per second average the AMD FX 4100 Black Edition achieves in the World in Conflict test may sound pretty reasonable, but it obscures the fact that the frame rate can, and does, drop well below 30 when the on screen action goes into overdrive.

Meanwhile, the merely dual-core Core i3-2100 cranks out 80- frames per second and generally keeps cooler under pressure.

In our testing, the 4100 didn't deliver significantly better overclocking headroom than the full eight-core Bully.

We liked:

To be honest, apart from the price, there really isn't a lot to like about this even weaker spin of the Bulldozer architecture.

We disliked:

We'd hoped for a lot more from this brand new AMD Bulldozer architecture, but the weak single-core performance, behind even the old Phenom II processors, makes us sad.

We could have forgiven it some of those failings if the gaming performance wasn't just as poor.

Final word:

The best that can be said about this dual-module Bulldozer, then, is that it's not far behind its triple and quad-module brethren in games.

If only they weren't all off the pace.

Overview

A new high-end processor from Intel is normally a cue for much rejoicing. After all, who doesn't like exciting technology and the promise of epic new levels of performance? In that context, the all-new Intel Core i7-3960X is as snazzy as they come.

It's a properly new chip, not an upclocked respin of an existing design. It even comes with a new socket and chipset, known respectively as LGA 2,011 and Intel X79.

But there's another side to the story of this chip, otherwise known as Sandy Bridge E. And it's symptomatic of a broader problem with the PC platform.

The story starts with a history lesson - the origins of multi-core PC processing.

It all began when Intel realised its fascination with frequency was on the verge of failure. The chip in question was the infamous Pentium 4 processor.

In its first and second generation iterations, Pentium 4 looked pretty clever. It rapidly scaled from a little over 1GHz all the way to 3GHz.

Then the die shrink to 90nm arrived and the wheels fell off.

That was June 2004 and the first significant date on our journey from a frantic fight for frequency to what increasingly looks like multi-core malaise.

Intel's response was to cobble a pair of Pentium 4 processors together and create the Pentium D dual-core processor of 2005. The Pentium D was then superceded in summer 2006 by the much revised dual-core Core 2 family, which in turn begat the quad-core Core 2 in early 2007.

Just like Intel promised, we were on a fast track to multi-core computing.

In hindsight, that's when the problems began. It took a further three years for Intel to release the six-core Gulftown desktop processor, sold initially as the Intel Core i7-980X. And here we are today with the launch of Intel's latest and greatest, a chip that defenestrates Gulftown's Westmere architecture in favour of Intel's fancier Sandy Bridge design.

But it's still only got six cores.

What, exactly, is going on? Things become even more intriguing when you discover Intel already sells server processors with up to 10 cores.

That's right, 10 cores.

Then you look at the die shot that Intel has released for the new Core i7-3960X and realise this new six core isn't quite what is seems. In fact, it's a little bit fishy.

Time to find out more.

Architecture

First, let's address the elephant in the room.

There's no mention of it in Intel's official documentation, but the image Intel has provided of the new Intel Core i7-3960X's die makes one thing perfectly clear: this is an eight-core PC processor with two cores disabled.

For now, we can only speculate on Intel's strategy, even if we do have a pretty solid hunch. But more on that in a moment.

Architecturally, the Intel Core i7-3960X is part of the the Sandy Bridge family. That makes it a 32nm processor with thoroughly revised internals compared with the previous six-core Gulftown processor. That, too, is a 32nm chip, but its Westmere-vintage architecture is merely a die shrink of the earlier Nehalem family.

Not that Nehalem, Westmere and Gulftown are stinkers. Quite the contrary.

The point is that the Core i7-3960X and its sister Sandy Bridge E chips bring Intel's high-end processor offering up to date.

Predictably, therefore, many of the best bits of Sandy Bridge already seen in processors for the LGA 1,155 socket make an appearence.

That means everything from Intel's latest iteration of Smart Cache, Hyper-Threading and Turbo Boost are all in the mix. Likewise, you get support for the latest SSE4.2 instructions, along with AVX and AES extensions.

There are also a number Sandy Bridge E-specific extras that don't apply to previous Sandy Bridge chips and are largely related to the chipset and peripherals.

Most obvious is the new LGA 2,011 socket, which has a big increase in pins over the Intel's existing LGA 1,366 high-end desktop socket. The explanation is an upgrade from a triple to a quad-channel memory interface.

Now, Intel's LGA1,366 platform was hardly memory starved in the desktop environment. What's more, since Intel is sticking with six cores on the new i7, increased demand for bandwidth is likely to be marginal.

But here's the thing. Just like the LGA 1,366 platform, LGA 2,011 is really a server and workstation solution that Intel has given a quick buff, changed a few names on and chucked onto the desktop.

Four channels of memory is handy when you're talking about multi-socket systems with as many as 10 cores per socket. For a single-socket desktop platfrom that maxes out at six cores, it's overkill.

Anywho, other changes in the shift from LGA 1,366 on the old X58 chipset to LGA 2,011 on X79 include a single-chip chipset, PCI Express moving onto the processor die and an extra four PCI Express lanes for a grand total of 40.

Plenty, in other words, for multi-GPU action in just about any flavour you fancy.

The SATA ports also get an upgrade to 6Gbps. But there's no native USB 3.0, which is a bit of a bummer.

However, the really critical areas where the Core i7-3960X deviates from previous Sandy Bridge processors are twofold and not entirely unrelated. Again, we're talking about the uncore stuff.

Most notably, Sandy Bridge E is a pure CPU, there's no integrated graphics and that means no Quick Sync video encode acceleration. That's not an absolutely critical omission, but it is a little hard to swallow in the context of the 3960X's premium pricing.

Quick Sync is a nice feature and something you get even on the lowliest of LGA1,155 Sandy Bridge chips.

Sandy Bridge E's final change up involves overclocking.

Intel has introduced a new divider between the baseclock and the cores, known as the CPU strap. The net result of it is that baseclocks of 100MHz, 125MHz, 166MHz and 250MHz are available. Now, for an unlocked chip such as the 3960X, that's not a terribly big deal. You have full access to the standard CPU ratios on a per-core basis.

But for the more affordable partially locked members of the Sandy Bridge range, specifically the upcoming quad-core Core i7-3830, it makes all the difference and ensures the LGA2,011 platform offers something extra for enthusiasts.

Benchmarks

Unfortunately, AMD's new Bulldozer-based FX processors aren't even in the same ballpark as the new Intel Core i7-3960X for performance. So it's Intel's own Core i7-980X that provides the competition. With similar clock speeds and the same core count, it a battle of architectures.

As our results show, the Sandy Bridge architecture is better across the board than the 980X's Westmere-vintage transistors.

Question is, is the gap big enough to justify an upgrade? When you factor in the Intel Core i7-3960X's impressive overclcoking headroom, it might just be.

CPU multi-threaded rendering performance

CPU single-threaded rendering performance

CPU video encoding performance

Memory bandwidth

CPU gaming performance

Platform power consumption

Overclocking performance

Performance

Out of the box, one thing is clear. The new Intel Core i7-3960X is in a class of its own.

That, of course, isn't terribly surprising. AMD's new pseudo eight-core FX processors are well behind the old Gulftown Core i7 processors on performance. So, they were hardly likely to give this new Core i7 anything to worry about.

The contest, therefore, is Intel versus Intel.

Does the new Core i7 represent a significant step forward over the old king? The answer, inevitably, is yes and no. More specifically, it's a question of whether you're talking about stock frequencies or running overclocked.

The official numbers give the Core i7-3960X a nominal frequency of 3.33GHz, with a top Turbo speed of 3.9GHz. In our testing, it typically ran at 3.6GHz regardless of how many cores were heavily loaded.

Now, the fastest Gulftown processor is the 3.46GHz Core i7-990X.

In other words, the new i7 doesn't bring a major frequency boost. Nor does it add extra cores. So any performance advantage is going to be largely architectural.

For our yardstick, we've commandeered a Core i7-980X. It's slightly slower than the 990X at 3.33GHz, but the difference is pretty marginal. Kicking off with the x264 HD video encode test, it turns out that the new Core i7-3960X tears through at 49 frames per second while the old timer manages 43fps.

Elsewhere, it's a similar story. In Cinebenches R10 and 11.5, the Core i7-3960X achieves 27 seconds in the former and cranks out 10.54 points in the latter. The 980X's numbers are 32 seconds and 8.64 points.

In World in Conflict, which remains one of the most demanding games in terms of CPU load, it's 93fps for the Core i7-3960X and 88fps for the 980X.

All of this means the new chip is appreciably but not dramatically faster.

If you already had a 980X, you'd hardly feel compelled to upgrade.

Unless, that is, you compared the available overclocking headroom. Our 980X sample chip is capable of clocking up to just 3.9GHz. But in our testing, the new beast is good for 4.8GHz. And that's on plain old air cooling.

At that frequency, the Core i7-3960X is spewing out 63fps in the x264 HD video encode benchmark. At 3.9GHz, the 980X is good for 51fps.

Factor in overclocking, therefore, and the 3960X's performance advantage roughly doubles.

Verdict

What to make of Intel's new flagship processor and platform, the Intel Core i7-3960X?

On the one hand, it's clearly the fastest PC processing solution money can buy. By some metrics, such as memory bandwidth, it absolutely blows away even Intel's own six-core Gulftown processors such as the Core i7-980X.

It also brings a healthy improvement in overclocking headroom. We've little doubt that stable systems running at 5GHz or more will be possible with water cooling. Platform upgrades that include more PCI Express lanes and a faster SATA interface are also welcome.

However, in more realistic tests of application performance, the gap to Intel's outgoing flagship processor and platform is rather more prosaic.

What makes this frustrating is that the Core i7-3960X is very probably capable of the kind of massive performance boost we expect from a new high-end Intel processor.

Deep down, the specs don't lie.

Intel quotes the transistor count at 2.27 billion. That's nearly double the number of transistors in a Gulftown processor, despite the same nominal core count. Admittedly, the Core i7-3960X has 3MB of additional cache and a few more on-die features. But that's not enough to explain away all of the additional one-billion-plus transistors.

The explanation, of course, is that the Core i7-3960X is actually an eight-core processor with a pair of cores disabled.

When you're paying the thick end of £750, it's very hard to accept the idea that Intel is holding something back.

The question that immediately follows is whether there will be any scope for core unlocking, as with processors made by AMD. In our view, it's extremely unlikely. For that to be the case, Intel would have to be under extreme pressure from AMD.

In reality, the opposite is true.

The lack of Quick Sync is another downer. There are good architectural reasons why Quick Sync doesn't appear. But that's Intel's problem, not the customer's.

And customers of the Core i7-3960X have every right to expect full access to the product they've paid handsomely for.

We liked

If you want the fastest processor on the planet, look no further.

Most impressive is the additional overclocking headroom Intel has delivered over the old six-core chip. Platform upgrades including more PCI-E lanes and SATA 6Gbps are welcome, too.

We disliked

Simple, it's those hidden cores.

We want them switched on and we want them switched on now. Until that happens, the Core i7-3960X will remain a deeply, deeply unsatisfying chip, especially in the context of its sky-high pricing.

Final verdict

Ultimately, the reason why the Core i7-3960X looks like it does comes down as much to the performance of Intel's main competitor, AMD, as it does anything else.

Had AMD's new Bulldozer architecture been the game changer we had been hoping for, it's hard to imagine Intel would have released the Core i7-3960X with a pair of hidden cores. It would have run the full eight cores.

We feared AMD's ongoing inability to challenge at the top of the performance tables would eventually lead to Intel sand bagging. With the Core i7-3960X, it's finally happened. And that's a shame.

AMD FX-8150 - Overview

We've waited a long time for AMD to release a brand new processor architecture, but finally the AMD FX-8150 has arrived.

This first chip is the vanguard of the somewhat tardy Bulldozer technology and is this top-of-the-line AMD FX chip, code-named Zambezi. This is the full-fat, eight-core AMD super-chip running at a not inconsiderable 3.6GHz straight out the box.

The FX moniker isn't a new one for AMD chips.

The last time we saw it used for its high-end parts was in the 90nm Athlon 64 FX-74 in late 2006. It's been reborn this year to cover the first in what AMD hopes will be a long line of Bulldozer-based CPUs.

The AMD FX CPUs represent the chip maker's first real new architecture since the exciting times of the first Athlon back in 2003.

And it is very much a new architecture; AMD has started from scratch with the design of the Bulldozer modules, taking a very different approach to what makes up a CPU core than anyone else.

We'll explore the depths of that architectural change later, but the real key is the use of that word 'module'.

Each of these modules holds the essential makeup of a standard dual-core processor, sharing certain non-timing sensitive parts.

The AMD FX-8150 has four of these Bulldozer modules and AMD claims that makes it the world's first eight-core desktop chip. It's tough to ignore the sickly sweet scent of fudge here, though this hardware-based solution ought be a lot more effective than Intel's mostly software-oriented HyperThreading.

The Bulldozer module represents the future of AMD processors going forward.

Speaking with Bernard Seite, an AMD Technical Advisor, last month he told us that the Bulldozer modules are likely to last as the basis for its CPU range for the next 5-7 years.

At the moment we've only seen the roadmap for the next couple of years, with the Piledriver update coming next year, Steamroller coming in 2013 and Excavator in 2014.

Yes, we know it sounds like a joke, but AMD's latest roadmap really does read like the Urban Dictionary Karma Sutra.

But hopes are still high for these new AMD FX processors, especially after it managed to snag the Guiness World Record for highest CPU clockspeed in September.

We saw this top CPU here running at well over 8GHz.

That is impressive, but realistically it's just numbers. The world record didn't demand the machine to actually run any applications, it just needed to boot into Windows and report it's clockspeed. It also didn't need to be running all its cores. The world record was actually only broken by a single Bulldozer module in operation and under serious liquid helium cooling too.

The precedent is there though and means the AMD FX-8150 ought to be one hell of an overclocker's chip. What that will mean in real-world applications though we'll soon see.

So how does the new AMD FX CPU stack up then? Can it give Intel's Sandy Bridge a run for its money and can it make AMD a performance chip manufacturer once more?

AMD FX-8150 - Architecture

The AMD FX-8150's new Bulldozer technology is a fairly radical departure from the standard model of processor design.

The big news is the modular design housing the constituent components of a dual-core CPU in one.

It's also AMD's first 32nm desktop CPU. Sadly that does make the chip seem a little behind the times with Intel set to release it's first 22nm processor in Spring next year.

But like Intel's shift in process size the change to this modular design allows the AMD engineers to squeeze more performance into a smaller space, and therefore also cut the costs of manufacturing.

Essentially the idea is to share parts of the module with lower utilisation, such as the Level 2 cache, Fetch and Decode components, while high utilisation parts, such as the Integer pipelines and Level 1 data cache, are separated out per core.

According to AMD that should give each Bulldozer module around 80% of the performance of a standard dual-core CPU.

It's this squeezing of two cores into each module that means AMD can produce an eight-core CPU for less than the price of Intel's top Sandy Bridge quad-core.

And seeing as competition on price rather than performance is more of AMD's concern than Intel's that's something it had to do with the AMD FX CPUs.

Still, like the AMD FX-8150 the Core i7 2600K - that top Sandy Bridge chip - can run eight processes in parallel.

Intel doesn't call it an eight-core chip, instead uses its HyperThreading technology to split the four cores in its die into eight processes. There is a certain amount of hardware in the Sandy Bridge die to make the HyperThreading magic happen, but mostly it's a software-oriented model.

Intel's way of spreading out four cores into eight processing threads operates mainly by effectively managing instructions going into one core and separating that out into two parallel threads for the operating system.

Turbo

Again like the Intel chips, the AMD FX CPUs use a form of on-the-fly overclocking to boost performance when there's spare capacity available.

The AMD Turbo Core technology has been used in the Phenom II CPUs before, but has been specifically enhanced for the Bulldozer architecture. Now it can use the Turbo Core to offer increased performance when all cores are active should the TDP headroom exist for it to do so.

The AMD FX-8150 has a standard clockspeed of 3.6GHz, with a possible Turbo Core available to it to allow it push up to around 3.9GHz.

When only a single core is needed though then Max Turbo comes into play.

This essentially takes the available TDP headroom for the entire chip and focuses it onto a single core giving it the ability to push even further in terms of clockspeed.

On the AMD FX-8150 that means it's theoretically possible for the chip to hit 4.2GHz as standard when only one thread is needed.

But what about the platform?

Despite the delay to the actual CPUs themselves the motherboard platform was actually launched a fair while back. The AMD 9-series motherboards have been doing a roaring trade out there even without the chip to back them up.

There's precious little difference between the 8-series and 9-series motherboards save for one vital ingredient; the AM3+ socket that supports the new AMD FX CPUs. That also means they aren't the priciest boards in existence.

Still, to get the best out of your Bully CPU, at least in overclocking terms, then dropping some cash on a decent board should be well worth the money.

To this end AMD is shipping out the Asus RoG Crosshair V Formula with each review kit.

That's a brief overview of the technologies behind AMD's latest chip architecture, and mighty impressive it looks on paper. But how does it perform in the real-world? Overclocking records aside, what will happen when you drop this chip into your motherboard at home?

Let's take a look.

AMD FX-8150 - Benchmarks

The first benchmark below is possibly one of the most telling.

Taking the eight-cores of multi-threading out of the picture you can see how the individual cores actually stack up. Running in single-threaded mode shows the FX-8150 cores actually running slower than the hexcore Phenom II it's replacing.

With the multi-threaded benchmarks though the FX-8150 starts to look more interesting. However the gaming benchmarks tell a worrying story.

When it's just relying on the GPU the story is much the same across the four CPUs we tested. DiRT 3 is shown here, but in Just Cause 2 and World in Conflict all four processors spat out roughly the same performance figures at the top resolutions.

Take the GPU out of the equation and the Sandy Bridge chips stretch out ahead.

Things aren't too pretty in terms of multi-GPU performance either.

In overclocking terms though the FX-8150 is a success. Stably running at over 4.7GHz is impressive and really pushes it ahead of the Core i5 2500K.

If the chip had been released at the 4GHz it can easily manage we might've been looking at a higher score...

Single-threaded performance

Multi-threaded performance

CPU gaming performance

High-end gaming performance

Multi-GPU performance

Overclocking performance

Platform power draw

AMD FX-8150 - Performance

All the clever architectural tricks in the world count for nought if the performance isn't there, so how does the top-end AMD FX chip stack up against it's rivals?

The short answer is not as well as we might have hoped.

AMD really needed this chip to be at least a rival for the current generation of Sandy Bridge CPUs, especially with a new production process and accompanying chips arriving in the first half of next year for Intel.

With the high clockspeed and nominal eight cores, we had hoped to see the AMD FX-8150 taking the resolutely quad-core Intel Core i5 2500K to task.

Indeed with that chip artificially hobbled by the switching off of HyperThreading you'd think it wouldn't be much of a contest.

In fact it's a lot closer than Intel could have dared hope, and certainly a lot closer than AMD would have wanted. Especially in the multi-threaded benchmarks, and that's quite a surprise.

In the heavily multi-threaded Cinebench R11 test the AMD FX-8150 comes away with a decent 5.98 index score, but with only four threads of processing grunt against the FX chip's eight the i5 2500K manages to get awfully close at 5.90.

That test is also interesting when you compare the previous top AMD chip, the Phenom II X6 1100T.

That's a full six-core CPU, but is running on the older 45nm hardware and at a slower clockspeed too. Yet that older brother still manages to put up a score of 5.88.

The eight-threaded Core i7 2600K however is streets ahead of all three.

Still, those extra threads come into account with the X264 video encoding benchmark. Even at stock speeds the FX-8150 actually starts clawing back some ground on the i7 2600K, beating the quad-core i5 2500K easily as it does.

Unfortunately the actual cores in the Bulldozer modules look like the weak link in the multi-threaded chain.

A quick check on the single-threaded performance of the FX-8150 highlights that weakness.

Using the single-threaded Cinebench R10 benchmark you can immediately see the problems the FX-8150 faces. Even the ageing Phenom II cores run faster than this brand new Bulldozer core.

And the two Sandy Bridge chips demonstrate just what an advantage their cores have over the AMD processors.

Gaming

The gaming benchmarks are similarly telling.

On the whole at the high end of the graphics spectrum, at the high resolutions, it's all about the graphics hardware. As long as the chip's not getting in the way then the GPU can operate with impunity.

Take the GPU out of the equation however and things get more interesting.

World in Conflict is a great CPU benchmark in gaming terms as different CPU hardware genuinely can show a difference. We ran the benchmark at the lowest resolution and lowest graphical fidelity, while retaining CPU-heavy operations such as physics, just to make sure the CPU was taking the load.

The two AMD CPUs behaved much the same, while the Sandy Bridge chips were well over a third faster.

We were hoping to run the Shogun 2 CPU benchmark too in order to see how things stood on a more modern engine. Sadly the game refused to load for the Bulldozer chip...

The AMD FX-8150 though is meant to be a rather capable gaming chip, and with a high-end GPU it delivers performance every bit as good as the Intel chips.

So there's nothing holding it back in gaming terms.

What about multi-GPU though? The FX-8150 ought to be a better bet for dual-graphics setups thanks to the two native x16 PCIe lanes within the platform.

And if you just took DiRT 3 as the only gaming benchmark worth a damn then all would indeed be rosy.

Using the AMD-sponsored title it garners an extra 3FPS over the competing Core i5 2500K. Not an impressive lead, but a lead nonetheless.

In the DX10 World in Conflict and Just Cause 2 benchmarks though the FX chip lags behind, in the case of WiC by over 25FPS.

There's no problems with the GPUs, or how they're talking to each other, as the GPU-centric Heaven 2.5 test ignores all the other hardware and just gives a score based on what the GPU is capable of. In both AMD and Intel setups the Heaven score was identical.

AMD will argue that legacy software is not where you'll see the benefits of the new hardware.

It may have a case, but still the vast majority of games released today are not DX11 games, they're DX10 and in some cases DX9.

The OC

But there is at least some good news for AMD though, and that is all down to the fact it has finally created an overclocking-friendly processor.

We've been used to 1GHz+ overclocks from Intel chips for a while now, and with the AMD FX-8150 we've managed to get the same scale of improvements.

With a Corsair H100 water-cooling block we were able to push the FX-8150 to an impressive 4.73GHz. With the same cooler we could only get 4.6GHz out of the two Sandy Bridge chips.

When you're pumping out that sort of performance then the benchmarks start to look a lot more impressive.

Now with the CPUs at their maximum overclock the Cinebench R11 test shows the FX CPU easily outstripping both the Phenom II X6 1100T and the Core i5 2500K.

The i7 still remains way out in front, though in the X264 test only by a single frame per second on average.

AMD FX-8150 - Verdict

Inevitably the performance of these chips is where they are going to be judged in real terms.

And somewhat inevitably it's something of a disappointment.

It's somewhat inevitable as AMD simply doesn't have the vast research budget of its Silicon Valley rival, Intel. We don't want to start making excuses for AMD though, and the FX-8150 is by no means a bad chip.

It is definitely the fastest CPU AMD has on the market right now.

Take Intel out of the equation and we'd be lauding the Bulldozer architecture as a truly remarkable thing.

The problem is Intel is most definitely in this equation and we've had this sort of performance, for around this sort of price, since we first clapped benchmarks on Sandy Bridge.

But Bulldozer remains the future for AMD.

And there is genuine hope on the horizon. The Trinity APU for example will be running the enhanced Piledriver architecture alongside discrete-class graphics in the same die.

For now though the top-end, eight-core AMD FX CPU struggles to keep pace with Intel's middling, ageing, cheaper and resolutely quad-core, Sandy Bridge i5 2500K.

And that's the big problem.

The FX CPUs are almost competitive with their Sandy Bridge rivals, but still can't beat them.

The 2500K is a cheaper chip and represents a better bet for gamers. If you want multi-threaded prowess too then for only another £50, at worst, over the FX-8150 you can pick up the awesome i7 2600K.

This is though only one of the Bulldozer-based chips AMD has launched, and the lower-end CPUs may actually be far more worthy.

The 3.1GHz FX-8120, for example, is available for around £165 – less than the 2500K – and should still have the overclocking chops thanks to the unlocked nature of the entire FX range.

Topping 4GHz with that CPU could turn it into a really very good gamer's chip with added multi-threaded extras to boot.

Still there is more to come from Intel, with another LGA 1155 Sandy Bridge reportedly on its way and the brand new, ultra high-end Sandy Bridge E is soon to touch down too.

We liked

The modular architecture is impressive and ought to pay off for AMD further down the line.

As it is the FX-8150 is an impressive overclocker's chip. Hitting 4.7GHz gives the chip one hell of a boost in performance terms.

We disliked

Unfortunately the chip's just not competitive enough against its rivals.

At stock speeds it struggles against the non-HyperThreaded i5, even in some multi-threaded applications. And while it keeps pace with the competition in gaming terms it loses it when you come to adding in extra GPUs.

Even AMD ones.

Final word

We can't help but feel disappointed with the lack of performance progress the FX-8150 represents. It's not a bad chip, but we wanted more.

AMD A6-3650: Overview

AMD's flagship model of its new A-Series range of desktop Llano Accelerated Processing Units (APUs) – the AMD A8-3850 – launched last month. At the same time, AMD announced a cheaper, less powerful APU, the A6-3650 – which we have now jammed into our test bench.

If you haven't heard about AMD's Fusion technology before, then the term 'APU' might be a new one to you. The APU is a hybrid chip bringing together both the CPU and GPU and sticking them onto a single die.

Before your eyes glaze over and the moaning starts about integrated graphics being about as much use as a chocolate teapot, hold on, because in the A-Series of APUs, the graphics performance is really rather startling. But more about that later.

This isn't the first incarnation of AMD's Fusion technology, but it is by far the most powerful.

The first Fusion APUs (codenamed Ontario and Zacate, versus the current codename, Lynx) had very low power ratings (Ontario 9W, Zacate 18W). They were aimed at the ultra-thin notebook market where, it has to be said, they didn't exactly set the world alight.

Likewise the first of the Llano chips, the A8-3500M, was aimed at the notebook market, although it made a much more favourable impression this time around.

AMD A6-3650: Specifications and performance

The engineering in AMD's new desktop A-Series is a bit of a tour de force for the design teams, because there's an awful lot going on inside a fairly small package.

Built on a 32nm process, the die of the AMD A6-3650 measures just 228 sq mm, but inside it packs a fully-featured quad core processor, a DirectX 11 GPU and a fully-featured Northbridge with a transistor count of around 1.45 billion.

With all this going on, it's quite some feat to keep the thermal design power (TDP) at just 100W.

To put that into some kind of perspective, a Phenom II X4 940, for example, measures 258 sq mm just as a processor on its own. On the other hand, the Phenom design has something the A6-3650 lacks – namely L3 cache.

The Llano Lynx parts have no L3 cache, and have to make do with just 4MB of L2 cache – 1MB per core. This puts the CPU part of the jigsaw more closely related to the latest Athlon II processors, such as the AMD Athlon II X3 450.

The A6-3650 runs at 2.6GHz out of the box – 300MHz slower than it's more powerful cousin, the A8-3850. And, like the A8-3850, it lacks the Turbo Core, although this will be part of the other two A-Series APUs that have yet to make an appearance – the A8-3800 and the A6-3600.

These chips have slower clock speeds, of 2.7/2.4GHz and 2.4/2.1GHz, respectively, but come with a TDP of just 65W.

When it comes to the graphics core, the A6-3650 uses the slower AMD Radeon 6530D chip compared to the A8's Radeon HD 6550D.

The HD 6550D is comparable to the discrete AMD Radeon HD 5570, while the HD 6530D is more like an AMD Radeon HD 5550, with a core speed of 443MHz, 320 Radeon Cores (AMD now calls the unified shaders Radeon Cores, in much the same way as Nvidia's labelled its CUDA Cores), 16 texture units and eight ROPs.

All of which means that the HD 6530D can deliver around 284 gigaflops of graphical processing power.

This may be a lot lower than the A8's 480 gigaflops but it still dwarfs Intel's HD 3000, which can only summon up a mere 125 gigaflops.

Oh yes, and the Intel part is only DirectX 10.1.

AMD A6-3650: Benchmarks

If you are used to integrated graphics stuttering along with single figure frame rates in most things you test them with, then the GPU in the new Lynx APU's will come as a bit of a shock.

Add in a value end card like the AMD Radeon HD6670 for dual graphics and you have some highly playable frame rates.

AMD A6-3650: Verdict

The one thing that really stands out with the AMD A6-3650 is its overclocking ability, particularly at its price point.

We were able to get our review sample to 3.04GHz (117MHz x 26) without any problems and without resorting to tinkering with any other settings in the Bios. Adjustments to the APU voltage settings should see it go much faster, but even so that's a 400MHz+ increase over the stock speed.

Of course it's not only the CPU part of the AMD A6-3650 die that's overclockable – the GPU is too.

You can go from the 443MHz of the HD 6530D up to the 600MHz that is the stock speed of the HD 6550D in the A8-3850, without any problems and with complete stability.

That sort of overclocking, however, is always going to be dependant on the particular slice of silicon in your machine.

It's also good to report that finally after all these years we have integrated graphics that actually offer some decent frame rates. Once you've recovered from that shock bit of news, the other point to take note of is that the GPU also supports DirectX 11.

Sadly, both the decent frame rates and the DirectX 11 support are not available with Intel's latest HD 3000 graphics – those of Sandy Bridge fame – so extra brownie points to AMD then.

There is also the added advantage that you can chuck in a cheapish discrete AMD graphics card, and by enabling the dual graphic mode you can achieve some very playable frame rates.

We did have a couple of odd moments during testing when using a graphics card in dual mode with the integrated graphics where the screen would go blank so you didn't quite know where you were in the boot cycle. Then suddenly it would wake up to show you that the OS had loaded fine, but on a couple of occasions we lost patience and rebooted.

This was probably due to the immaturity of some of the motherboard drivers, and was annoying rather than alarming.

The only other thing of concern is the longevity of the platform itself.

With Bulldozer and the Bulldozer-based APU, Trinity, fast approaching on the horizon, the worry is that the FM1 socket will be here today and gone tomorrow. Remember what happened with Intel's Socket 1156 when the new Sandy Bridge technology appeared?

Getting one of these new APU's isn't quite the future proofing move it might appear to be at first glance then.

We liked

There's no doubt about it, with these new Lynx desktop chips, AMD has seriously upped the ante when it comes to integrated graphics performance, especially at the price the A6-3650's been pitched at.

The dual graphics option is also interesting, because it allows for some fairly intensive gameplay without having to break the bank for the discrete graphics card.

One of the nice surprises of the APU is its ease of overclocking and how far it will go without having to get your hands too dirty in the Bios.

Another plus point is its support of both the latest JEDEC memory frequency standards over and above the 1,333MHz base point, at 1,600MHz and 1,866MHz.

We disliked

Aside from the slightly immature drivers around at the moment, there is a worry that the FM1 socket won't be around for very long.

Verdict

We were impressed with the AMD A8-3850, and similarly the cheaper A6-3650 doesn't disappoint either

While Intel has the performance end of the CPU market sewn up, AMD has been looking out for the little guy, focussing squarely on the mainstream market and cheap chips with competitive processing chops.

The Sandy Bridge family might be your go-to guys at the moment for CPU upgrades thanks to Quick Sync and Turbo Boost technologies, but if every penny counts, AMD's Phenom II chips squeeze every drop of performance out of a middling budget.

The compromise though is some architecture that's getting rather long in the tooth. The X4 980 Black Edition is no exception.

Pound for pound, it delivers excellent performance. HyperThreading though is off the menu on the AMD side so you're relying on four threads and four cores to slice through processing tasks. How much of a problem that is depends on your CPU's workload – video encoding and design apps make good use of multi-threaded processors, and although games have been slow off the mark, we've seen some big titles recently that do work with multiple cores and threads to speed up rendering tasks.

That X4 architecture's a bit old now – it was after all AMD's first quad-core architecture. Intel's 32nm chips are superior in a straight fight, but with the X4 980 BE you still get 6MB of L3 cache and an out-of-the-box clockspeed of 3.7GHz – AMD's fastest ever quad-core.

But of course, it isn't a straight fight. At £140, the 'enthusiast' X4 980 is priced against Intel's entry-level Sandy Bridge chip, the i5 2400. That means it has a 500MHz faster clockspeed, but doesn't have the Turbo Boost and Quick Sync chops of the i5. A stalemate, roughly.

That double-edged sword of older technology gives the AMD chip another selling point though. It's an easy upgrade. Any Socket AM3 motherboard will support it, so you can reinvigorate two-year-old machines with this chip. It's certainly an easier, and cheaper, upgrade than the jump to Sandy Bridge.

Benchmarks

Rendering performance
Cinebench R10: Seconds: Quicker is better
Phenom II X4 980 BE (Stock): 59
Phenom II X4 980 BE (OC'd): 51
Phenom II X4 975 (Stock): 60
Phenom II X4 975 (OC'd): 51
Phenom II X6 1090T (Stock): 47
Phenom II X6 1090T (OC'd): 39

Encoding performance
X264 HD v2: Frames per second: Higher is better
Phenom II X4 980 BE (Stock): 20
Phenom II X4 980 BE (OC'd): 23
Phenom II X4 975 (Stock): 19
Phenom II X4 975 (OC'd): 21
Phenom II X6 1090T (Stock): 26
Phenom II X6 1090T (OC'd): 32

DX11 gaming performance
Just Cause 2: Frames per second: Higher is better
Phenom II X4 980 BE (Stock): 47.93
Phenom II X4 980 BE (OC'd): 52.41
Phenom II X4 975 (Stock): 48.98
Phenom II X4 975 (OC'd): 51.34
Phenom II X6 1090T (Stock): 45.58
Phenom II X6 1090T (OC'd): 44.89

Enemies within

Here's where things get weird though – it's not Intel that's muscling the X4 980 BE out of the market… it's AMD itself.

Its focus on the sub-£200 CPU market has led to a product range that caters for literally every budget. Can't go a penny over £140? X4 980 it is. Won't budge an inch on £135? X4 975 then. In fact, the range is so overpopulated that some CPUs start pushing out others.

Ladies and gentlemen, meet then the Phenom II X6 1090T. It's a six-core, 3.2GHz brute that's also happy in any AM3 board. It shreds the X4 980 Black Edition to bits in processing tasks… and incredibly it's £15 cheaper.

While the X4 can't live with the X6 in video encoding, it does still have the edge in gaming thanks to that faster clockspeed. What's more, they both overclock well – topping 4GHz on air. So which to buy?

We're back to multi-core support again. If you're happy getting the most out of a single core in your applications, the X4 980 Black Edition is looking good. As games step up their multicore support though, the X6 has the potential to blitz the 980 for less cash.