Haswell: everything you need to know about Intel's latest Core processors
26th Jun 2014 | 13:05
Intel's latest chips uncovered
Intel's new Haswell chips are here. The question is, should you care? As we'll discover, the answer depends entirely on what you want from a PC. Haswell is a new family of 22nm processors from Intel, still known as i3, i5 and i7 inside PCs and on the shelf.
So has Intel succeeded again? As we'll find out the answer is a bit of yes and no. Where Intel has failed us, those failures are entirely intentional. Where it has succeeded, it reflects Intel's current and likely future priorities.
To put it simply, Intel's consumer-level processors have been all about mobility right back to those first Core processors in 2006.
Intel will release a new version of Haswell, called Haswell-E, in the second half of 2014, possibly at IDF in San Francisco to target the lucrative enthusiast/workstation markets.
This new version, whose launch Intel apparently postponed by a couple of quarters, will see the introduction of a new chipset (the X99), Intel's first 8-core, non-server processor as well as DDR4 memory.
All about mobility
But with Haswell, that focus has almost entirely put paid to progress to the pure CPU part of the chip in desktop performance terms. We'll find out more about this bias as we dig deeper into the technicalities of the new Haswell chips. But what was a niggling doubt with previous processors has become screamingly obvious. Intel has done almost nothing to improve the desktop experience with Haswell.
15 best graphics cards in the world
For mobile PCs, Haswell is all kinds of awesome. You could even argue it's the beginning of a new era of computing. Laptops that can survive for a week on a single charge. Touch devices that combine a true tablet form factor with the power of a desktop PC. It's portable, nay pocketable, computing with any compromises.
The work Intel has done with both power consumption and integrated graphics is excellent for laptops and tablets. It's also very promising for a new generation of ultra-compact PCs, including Intel's own NUC as well as set-top boxes, tiny gaming machines, PCs hanging off the back of monitors, you name it, Haswell is going to make it either possible or much better than before.
If that's the upside, the downside is that with Haswell, Intel has taken its disregard for desktop customers to new levels of indifference, maybe even spite. Yes, the desktop is somewhat on the wane by some metrics. And the fact that Intel's main competitor, AMD, has failed to turn up to the fight in several parts of the traditional desktop market hardly helps.
But even taking those factors into account, the way Intel has designed, positioned and priced desktop Haswell processors approaches spite for desktop customers. Somebody's got to say it. So we will.
But hold that thought. First, let's deep dive into Haswell as a technology. Because as you'd expect from an Intel processor architecture, it's a marvel of modern chip engineering. That's actually something all too easy to loose sight of, especially when Intel muddies the message with its marketing machinations. Haswell is a proper bit of kit. The big news involves two key metrics, power efficiency and graphics.
Haswell power efficiency
Intel says Haswell represents the biggest single generational gain in power efficiency in the history of its x86 PC processors. Yup, bigger even than the gains Intel made going from the infamously overheating end-of-line Netburst Pentium 4 chips to the first Core Duo processors. Wow.
So what has Intel done to achieve that and what does it mean for actual devices like laptops and tablets? The process started with Intel changing its targets for power consumption. 35 to 45 watts used to be the design target for Intel mobile processors. Now the target is 10 to 20 watts. And don't forget, we're talking about relatively mainstream chips here, not the very lowest voltage parts.
As for how this is achieved, Intel has been putting the pieces in places over the last few years. A major part of that process has been integrating ever more features into the CPU itself. These days, we have everything from the memory controller and I/O to graphics, all on one chip.
With Haswell, Intel is taking that a step further with the upcoming Mobile-U ultra-low voltage models, which are pretty much SoC or system-on-a-chip designs, complete with PCH functions like USB on the CPU die. That's not actually the case for the desktop and Mobile-M parts that have been announced so far. But it's coming soon. Whilte the Mobile-U parts promise a massive 20x reduction in power consumption in some scenarios, the Mobile-M parts deliver a more modest, albeit still useful, 20 per cent reduction.
Anyway, along with the improved power consumption comes an overhaul of what it means to be an Ultrabook based on Intel technology. Existing Ivy Bridge generation Ultrabooks must offer more than five hours of battery life, for instance. With Haswell, that leaps to nine hours for idling in Windows 8 and six hours of HD video playback. If you can't manage that, you're not an Ultrabook.
In some ways even more impressive is the requirement for seven days of standby including what's known as "fresh data". That means a Haswell notebook can sit for seven days, periodically pulling down email, social media updates like your twitter feed and all that jazz without dying. It's analogous to (but not actually the same as) the scenario with your smartphone when you switch it off – which isn't actually off – but it still receives calls, texts and push email updates.
The way this is enabled involves new sleep states for the chip. These sleep states essentially give you the battery performance of the old suspend-to-RAM state with the responsiveness of sleep states existing PC processors use when the CPU cores are idle. The idea, then, is that a Haswell notebook can sit for a week on a single charge but always be ready with your latest email and updates the moment you power up. Nifty, eh?
Of course, lower power also means you can have a whole new class of mobile devices. We're talking tablets, of course. You could argue that process already started with Ivy Bridge chips.
After all, you can have an Ivy Bridge Intel Core i5 processor in Microsoft's Surface Pro tablet. But with Haswell, the hardware-related niggles that prevent Surface Pro from being a killer product – the extra thickness, the mediocre battery life – could well be banished. Just think, a tablet as thin and light as an iPad, with comparable battery life, but the performance to run all your favourite desktop applications.
Essentially, what we're talking about is a tablet with the ability to replace your laptop, rather than a tablet that's yet another device you have to lug around. And that's a hugely compelling proposition. Lower power or improved performance at existing power levels will also be a boon for a new generation of boxes and systems around the home.
When you look at what Intel has packed into its most powerful Haswell chips, for instance, the idea of a Haswell-powered NUC (that's Intel's tiny Next Unit of Computing box) is super exciting. It might even make for the basis of a cheap gaming box in the style of Valve's Steam Box.
Then there are all-in-ones with articulating touchscreens and batteries that blur the lines between desktops and mobile devices. Put simply, with lower power all sorts of options and possibilities appear.
A key enabler of all this and also the second major upgrade that comes with Haswell is graphics. Graphics performance is in many ways the final frontier of computing. It's also the last major differentiator between desktop PCs and mobile devices. With Haswell, Intel is making its first genuine attempt to plug that gap.
That effort takes the form of Intel's most powerful graphics core yet. There are lots of different versions of the new graphics core, but the underlying technology hasn't changed much. What has changed is the amount of silicon Intel is willing to spend on graphics. It's gone up. Massively.
The quickest existing Sandy Bridge version of Intel's HD Graphics tops out at 16 execution units and 1,150Mhz. With Haswell, that explodes to 40 units and 1,300MHz. Well over double the raw computational performance, in other words.
But wait, there's more. Intel has also added an optional 128MB slab of eDRAM to Haswell. It's a separate slice of silicon, but fitted to the same package as the rest of the processor and actually operates as a shared L4 cache for both the graphics and the CPU cores. But one of its key roles is offsetting perhaps the biggest bottleneck for integrated graphics performance. Bandwidth.
Best CPU: the 8 top processors today
Exactly how much difference it makes to performance is something we'll have to wait and see. We don't yet have our hands on a Haswell processor with the eDRAM chip. But even on paper, it remains a little off the pace of a mid-range discrete graphics chip for raw memory bandwidth.
Anywho, there are a metric tonne of slightly different graphics cores with Haswell. That eDRAM effort with 40 execution units is one of two new cores bequeathed to us via the new Intel Iris brand. So, that's Intel Iris Pro 5200. Take away the eDRAM but keep all 40 units and you have Intel Iris 5100. Keep the 40 units again but lower the clocks a bit from 1,300MHz to 1,100MHz and the Iris bit is dropped in favour of plain old Intel HD Graphics 5000 branding.
Drop to 20 cores, but up the clocks to 1,350MHz and you have Intel HD Graphics 5000, which is what's in all three chips we've so far test. There are further variations, but we suspect you're beginning to lose the will to live. No, we don't know who makes this nonsense up, but we can only assume their primary aim is to make the branding as baffling as possible. At the very least, why on earth aren't all the 40 unit cores called Iris?
But never mind, there's a pretty siginificant part of Haswell we've yet to discuss. Yup, it's the CPU cores themselves. It's indicative of Intel's priorities that the cores come last in this discussion and that there's actually not all that much to say.
Yup, there's a list of tweaks we could go through in forensic detail. Would could explain, for instance, how Intel has massaged Haswell's front end (ooer!) for better branch prediction. Intel has also improved Haswell's instruction level parallelism (ILP) courtesy of upping the szie of various data-level structures, including the out-of-order window and the in-flight loads and stores. Intel has also added a pair of execution ports, bringing the total to eight, and added the obligatory extra floating point instructions, known as AVX2.
But it's still pretty much the same four-wide execution engine as before with the same 14 to 19-stage pipeline as Ivy Bridge. And that means that the vast majority of the time, it performs a very similar amount of work each clock cycle.
Which really leaves clockspeeds and cores as the best way to improve performance. Except Intel hasn't done that. The top Intel Core i7-4770K has the same 3.5 default frequency as its Core i7-3770K predecessor. And the Core i7-2700K that came before that.
Nor has the core count changed. We're still talking about a maximum of four cores for Intel's mainstream socket (which incidentally has changed again, more on which in a moment). All of which means you have to go all the way back to the likes of the In fact, you have to go back to the likes of the Intel Core i7-875K almost exactly three years ago to find a comparable chip from Intel that offers tangibly less performance.
On the one hand, that reflects just how good Intel's CPU cores have begun. On the other, it shows that Intel is no longer prepared to spend transistors on CPU performance. The Haswell Core i7-4770K has roughly twice the number of transistors as that Core i7-875K. But both are quad-core. Intel has put the extra complexity mainly into a graphics core which performance PC owners largely ignore.
Moreover, overall, the 4770K is much smaller than the 875K (177mm2 versus 296mm2), so Intel is just spending less money on desktop chips period, while maintaining the same high prices. Intel, of course, would argue there's the high-end LGA2011 socket for serious desktop performance. But even that is stagnating.
The only other change of note for the Haswell CPU is a small tweak for overclocking. Intel has exposed a baseclock strap, enabling a bit of fairly crude overclocking via the baseclock. But it only applies to K series chips which already have unlocked multipliers, so it's not much of a game changer.
Anyway, the final part of the Haswell puzzle involves the supporting platform. With Haswell comes yet another new socket. Thus, LGA1156 begat LGA1155 and now we have LGA1150. Inevitably, this breaks backwards compatibility. You can't upgrade an existing PC with Haswell chips. You'll need at least a new motherboard.
So, there it is. Intel's new Haswell processors laid bare. Never before have we been in two such starkly contrasting minds about an Intel CPU. For mobile systems, it's seriously exciting. We love the idea of tablets with desktop processor power. For actual desktops, well, you can read our individual chip reviews. But it ain't pretty.
So, Intel's new Haswell processors don't move the game on much as straight-up desktop processors. But they do come with a set of new 8 Series chipsets. Can they redress the balance?
Key chipsets include the Z87 (which replaces current fave the Z77), the H87 and H81. Our main focus is the Z77 and the news is indeed mostly good. It actually involves a slightly less complex PCH chip than the old Z77.
That's because Intel has moved even more functionality onto the Haswell processor and therefore off the motherboard chipset. The display interfaces for the integrated graphics have now largely moved on-die, for instance. On that note, the Displayport interface now supports resolutions up to 3,840 by 2,160. That makes it compatible, for instance, with Asus's new 4K 32-inch monitor. Interesting.
Anyway, the Z87 PCH is actually physically smaller than the Z77 item. But that hasn't prevented an increase is some specifications. All six SATA ports are now the full 6Gbps Monty. Hurrah. There's support for six USB 3.0 ports, too. Hazzah. Elsewhere, it's mostly either nothing new or tweaks and revisions to existing features.
Memory support sticks with DDR3 up to 1,600MHz, there's support for one 16-lane PCI Express graphics card or two eight-lane cards and there's a bunch of revised Intel tech include Rapid Storage, Rapid Start and Smart Connect.
Is it enough to make Haswell exciting on the desktop? Frankly, no. There may be six SATA 6Gbps ports, for example. But even SATA 6Gbps can't keep up with the fastest SSDs. It's also worth noting that early C1 revisions of the Z87 PCH have a bug that causes problems with devices connected via USB 3.0 when the PC wakes up from a sleep state.
Quite a lot of fuss has been made over this, but it's hardly a new problem – devices needing to be physically disconnected and reconnected after wake from sleep is fairly common. In any case, it's been fixed with the new C2 stepping of the chipset. If it's a concern, leave it a few months for C1 motherboards to be purged from retail channels and then pull the trigger.
And what of AMD?
The quest for improved mobility goes a long way towards explaining why Intel's Haswell chips disappoint as desktop processors. But it's not the whole story.
A major contributing factor is AMD's inability to keep Intel honest. It's impossible to know exactly what the world would look like if AMD had more competitive CPUs. But we're confident Intel is capable of a lot more than it is currently releasing in pure CPU power terms.
After all, its mainstream chips have been stuck on four cores for several years. And the physical size (and therefore manufacturing cost) of those same mainstream CPUs has been shrinking. That means Intel can make more money per chip.
And without AMD to force it to give us all something better, you can hardly blame Intel for cashing in on its advantage. It's a business with shareholders and it exists for one purpose. To generate wealth for those shareholders.
What we desperately need, then, is for AMD to give Intel a kick up the behind in the shape of some faster CPUs. Of course, the good news is that Intel's current tactic of mostly standing still is giving AMD a better chance of catching up.
There's still a long way to go, but there are reasons to hope AMD's upcoming Steamroller architecture revision might just get the job done. Certainly, AMD is targeting the main weakness of its current FX desktop processors, namely single-threaded performance.
The problem is AMD's clever modular architecture. Each module arguably shares too much hardware across a pair of execution cores. Steamroller aims to fix that courtesy of a switch to independent decode engines for each integer engine in a Bulldozer module.
Instruction fetch is still shared across a pair of integer units, as is the floating point unit. But Steamroller is significant step in the right direction. It's due out early next year on the desktop. And it's badly needed, for both AMD and the PC market as a whole.
Haswell desktop CPUs: what's available?
Haswell isn't that exciting on the desktop, but if you're in the market for a new CPU anyway, it should be on your list. Exactly what models are available? On the mobile side, there are a bunch of dual and quad-core models. As we go to press the NDA hasn't quite lifted on the dualies, but the quads take in the H and M families, the former packing the vaunted Iris Pro graphics for maximum 3D performance.
So Intel Core i7-4800MQ, 4900MQ and 4930MX, which are the high clocking, but only have the slower Intel HD Graphics 4600 3D core. Meanwhile the 4950HQ and 4850HQ have lower clocks, still four cores and Iris Pro 5200 graphics. If that's not confusing enough, the pricing overlaps a bit. You'll typically be buying them as part of a laptops, so the actual numbers aren't hugely relevant.
On the desktop side, there's a whole hill of quad core i7 and i5 chips at launch, but just the one dual-core model. All but one has Intel HD Graphics 4600, so that's easy enough to remember. For now, anyway. In the mix are a bunch of low power chips, but we'll ignore most of those as they're only value for money in very specific scenarios.
From the bottom, then, there's the Intel Core i5-4750T. Two cores, four threads, 2.9GHz baseclock, 3.6GHz Turbo, yours for about £160. Then there's the 4570, four cores, four threads, 3.2Ghz/3.6GHz and about the same price. And the £190 4670K, again four cores and threads, 3.4GHz/3.8GHz and an unlocked multiplier for overclocking.
Next up, the first Core i7 chip, the 4770. As an i7 it's four cores but eight threads and 3.4GHz/3.8GHz. £250 all in. As for the range topping 4770K, that's £275 in return for four cores, eight threads and 3.5GHz/3.9GHz.
Now why not read our Intel Core i7-4770K review?