What CPU is the best for games? What is the best processor for productivity? What is just the hands-down best CPU?
Stick with us people, we've got all you need to know to find out what processor you should be dropping your cash on.
This year could have been one of the greatest ever for PC processing. After an awfully long wait for new high-end chips from AMD or Intel, verily they both delivered. The result could have been, and should have been, a titanic tussle for desktop supremacy.
Best PC upgrades for gamers
Best CPU: the 8 top processors today
Best gaming laptop: 8 reviewed
Even for those who can't afford premium-priced chips – and let's be honest, that's most of us – the knock on effects would have been huge.
Competition not only improves the breed; it also tends to push down prices. Faster chips and cheaper prices for all, in other words.
As it turned out, neither AMD nor Intel's uber-processors were quite what we were expecting.
AMD's new Bulldozer-based FX processors, for instance, haven't exactly torn up the record books for raw computational grunt. Rumours of upcoming respins could translate into improved performance, of course.
For now, the FX is AMD's new top chip and at least offers an intriguing alternative at certain price points.
Intel's latest range topper, the chip known as Sandy Bridge E, has indeed emerged as the hands-down fastest PC processor you can buy, but it's still not the massive leap forward we had hoped for.
Worse than that, it very much looks like it's a case of Sandy by name, sand bagging by nature.
Put another way, we're convinced Sandy Bridge E could easily be much, much faster, if only Intel was facing stiffer competition. With AMD's Bulldozer not delivering, Sandy E's performance has been capped at a level that's merely good enough. Intel has plenty in reserve should AMD's form take an uptick.
If that seems like a grim assessment of the current state of CPUs, the reality isn't all that bad. New CPUs mean new platforms and in turn that makes now a great time to buy into a high-end system while maximising your long-term upgrade options.
Meanwhile, AMD's failure to take the high end by storm means Bulldozer has been dropped right into the heart of the mid-range melee. Since few of us can actually afford Intel's £820 monsters, maybe we only need AMD to keep Intel honest in the midrange. There's only one way to find out.
Is the PC in crisis?
Certainly the unleashing of new CPUs from AMD and Intel has left us all with a pair of crucial conundrums. Can AMD survive the failure of Bulldozer to sock it to Intel? And just how badly is Intel sand bagging with Sandy Bridge E?
If you reckon we're trying some scaremongering, well, think again.
Bulldozer is a big problem for AMD. Quite literally, the new FX processors sired by Bulldozer are very large. Clocking in somewhere between 1.2 and 2 billion transistors, they are twice as complex as any previous AMD desktop CPU.
That's fine if you're slapping the competition around with large benchmark results. If you've got the biggest performing processor, you can attach an equally large sticker price.
As it turns out, however, Bulldozer is struggling to match the performance of Intel chips with little more than a billion transistors.
Frankly, Bulldozer FX doesn't even compare all that well with some of AMD's own chips packing fewer than a billion transistors.
Bulldozer's problems become especially vexing when you dig down into the architectural details. The big idea with Bulldozer is beating Intel at the multi-threading game. Put simply, there are two ways you can add threads to a CPU.
Most obviously, you can drop in more cores, but you can also allow a single core to process more than one thread.
Playing with threads
The latter option is known as simultaneous multi-threading or SMT for short.
Intel has offered SMT for the better part of a decade, dating right back to the Northwood version of the Pentium 4. In fact, the very first Willamette P4 also had Hyper-Threading (HT) support, as Intel calls it, but it was never switched on.
The early P4 flavour of HT was never that impressive.
It duly disappeared around the same time Intel gave up on chasing high frequencies and changed its focus to power efficiency and multi-core processing with the introduction of the Core 2 family.
However, the basic idea was sound and Intel proved just that when it reintroduced Hyper-Threading with the Nehalem architecture in late 2008. Switching on Hyper-Threading boosted system performance by up to 25 per cent.
Critically, however, that performance increase comes at the cost of very few transistors. Intel's Hyper-Threading is all about using existing resources more efficiently, not bolting on more bits.
In that context, the challenge for AMD was to come up with a CPU architecture that met the dual challenge of increasing the thread count without ballooning the transistor count. And all the while it had to do so more effectively than Hyper-Threading.
AMD's solution is the Bulldozer module.
In simple terms, it's a halfway house between bunging in more cores and Intel's 'multithreading lite' in the form of Hyper-Threading.
A Bulldozer module has a pair of integer execution units, each with its own scheduler. But the rest it effectively shared. So, that's just one decode and fetch and a single floating point scheduler feeding a lone floating point engine, albeit composed of dual 128-bit FMACs.
Now, exactly how effective this would turn out in terms of actual multi-threading performance was pretty much impossible to predict. Predictably, AMD said the Bulldozer was SMT done right, inferring that Intel's HT was SMT on the cheap.
Bulldozer's modular design also threw into doubt the very notion of a CPU core. Did a Bulldozer module count as one seriously hefty core? Or is it a sort of dual-core processor lite? Again, only testing Bulldozer would give us the answer.
Bull-dozy
However, one thing that we were certain about was that a Bulldozer-based chip with four modules and therefore four or eight cores – depending on how you look at it – would end up being a lot smaller in size and transistor count than a conventional eight-core processor.
After all, if it wasn't, AMD may as well have produced a full-on eight-core chip without risking any performance losses associated with Bulldozer's shared resources.
So here's the bombshell.
AMD's six-core Phenom II X6 processor packs precisely 904 million transistors. A quad-module Bulldozer FX chip weighs in at a scarcely believable two billion transistors. Now, it's true that the Phenom II X6 has just 9MB of cache to the FX's 16MB, but that alone is not enough to account for all those extra transistors.
The problem for AMD really hits home when you postulate the kind of performance it might have achieved with a straight forward eight-core update to the Phenom II architecture.
Some simple maths suggests it would be both much faster across the board than a quad-module Bulldozer FX and also smaller and cheaper to manufacture. Meanwhile, hypothetical 10- or even 12-core Phenom IIs would probably fit inside Bulldozer's two billion transistor count and not only blow it away, but very possibly give Intel a run for its money, to boot.
Hell, even Intel's mighty Gulftown six-core chips only swallow up 1.17 billion transistors.
Exactly how AMD ended up with a two-billion transistor chip with at best eight cores and at worst just four, we may never know for sure.
However, one theory mooted by former AMD engineer Cliff Maier is that the explanation involves AMD abandoning hand-tuned circuit design. Instead, Maier claims, AMD has shifted to an entirely automated design process. He also says that circuits designed with automated tools are not only 20 per cent bigger than hand or human-tuned circuits, but 20 per cent slower, too.
Whatever the truth, the upshot is that AMD's FX chips measure 315mm2, which is 50 per cent larger than Intel's Gulftown, a chip that easily outperforms the FX and sells for as much as three to four times the price.
Middling AMD
How much this all matters to you and us depends on the time frame you are considering.
AMD has set FX prices that allow it to compete quite well against Intel's middle to higher performing LGA 1155 processors, such as the Core i5-2500K and Core i7-2600K.
That's all fine and dandy for now. We would advise anyone looking for a chip costing between £100 to £200 to give AMD FX a look.
But in the longer term, letting such a big and expensive chip go for comparatively pitiful prices (remember, Intel's top chips go for anywhere between £400 and £820) cannot possibly be sustainable.
Unfortunately, the fall out may already have begun.
AMD recently announced a round of 10 per cent redundancies affecting pretty much every part of its business. Whether AMD as we know it will survive has become a very real question.
In the meantime, what seems very clear is that another failed CPU design will very probably kill the company. From here on in, every significant product salvo must be on target.
If that's the state of play at AMD, it also provides the context for Intel's most recent CPU launches.
We speak of the new Core i7-2700K and the all-new Core i7-3000 series.
The former takes the tried and tested quad-core Sandy Bridge die and merely ups the nominal clockspeed from 3.4GHz for the 2600K up to (wait for it) 3.5GHz. Not terribly exciting, but easy enough to understand. Critically, the 2700K is only around £20 more expensive than the 2600K. So while you're not getting much more performance, you're not paying a lot extra for it, either.
It's therefore the new Core i7-3000 series that's most intriguing, both in terms of what it might do to high-end PC performance and what it says about Intel's current attitude. All Core i7-3000s are based on the new Sandy Bridge E core.
The first big surprise for us was a total core count – it maxes out at just six.
Not only has Intel already been offering six cores on the desktop for nearly two years. It's been flogging server CPUs with up to 10 cores for a while, too.
Okay, those chips sell for several thousand pounds a pop. But at the very least we were expecting Sandy Bridge E to bring eight cores to desktop computing for the first time.
Playing hide the cores
But the really big surprise came when we saw the die plot image provided by Intel.
Like any plot, it showed the functions of each block on the CPU die in broad brush terms. For Sandy Bridge E, that involves a large lump of cache in the middle, cores down each side and uncore kit including I/O and the memory controller above and below.
The difference, however, was a pair of unlabelled blocks, one each side of the cache memory pool. And it was very clear these blocks were disabled CPU cores.
Yup, It turns out that Sandy Bridge E is an eight-core chip after all. We've seen CPUs with disabled cores before, of course. It's one of AMD's favourite ruses, allowing additional processor models to fill niches, reducing the need to spin up multiple chips and minimising the number of processor dies that end up in the bin.
The thing is, Intel has always been a bit sneery about AMD's core-disabled chips, even inferring that it wouldn't lower itself to such cheap shenanigans. But here we are with an Intel chip with disabled cores.
What on earth is going on? Well, in our view, it's pretty straightforward – Sandy Bridge E isn't running in full eight-core mode because it doesn't need to.
AMD isn't even close to challenging Intel at the top of the performance tables.
Meanwhile, thanks to optimisations to Sandy Bridge cores compared with the previous Nehalem architecture, even with the same core count as Gulftown, the new i7-3000 chips have a modest but unmistakable performance advantage.
For Intel, therefore, it makes sense to ship today with six cores, allowing the use of processor dies with broken cores that would otherwise be thrown away.
Meanwhile, should AMD release a revised Bulldozer chip that unexpectedly raises the stakes, Intel can simply flip the switch on the hidden cores and open up the gap once again.
That's all pretty frustrating.
We'd much prefer to see Intel pushing the envelope, not dragging its feet. On the other hand, you could argue LGA 2011 in its entirety is pretty irrelevant to most desktop users, even to most enthusiasts.
However, in a parallel universe where AMD's Bulldozer was putting Intel under serious pressure, it's easy to imagine Intel releasing a six-core processor for the mainstream LGA 1155 socket and placing it at roughly the same £250 price point as existing LGA 1155 chips.
However you slice it, then, we're all paying a little ourselves for AMD's failure to take the fight to Intel.
In that context, now is definitely one of the trickier times to pick your next processor. The new FX chips are certainly competitive at the price points AMD has chosen. But do you really want to buy into a platform – and even a CPU manufacture – with such an uncertain future?
At the same time, the launch of Sandy Bridge E has just made Intel's LGA 1155 platform look a whole lot less attractive.
The problem, as ever, is that Intel's split socket strategy puts the kibosh on the upgrade path. As it stands, there's no significant upgrade path beyond four cores for LGA 1155. That's something that's not even set to change when Intel releases its 22nm Ivy Bridge processors next year.
Overall, then, it's not an ideal situation. There's not a perfect platform among AMD's AM3+ or Intel's LGA 2011 and LGA 1155 options. And as long as AMD struggles, all things chip-related will remain a little uncertain.
It's a complicated process then, buying a new CPU, so what processor is right for you?
We've checked out the best CPU technology available so there's no need to worry, come with us and find out where to put that hard-earned cash.
Connect with Facebook
No comments