Intel's New Uber CPU And The Future Of PC Gaming
10 is better eight, isn't it?
Once upon a time, the launch of a new Intel uber CPU was unambiguously exciting. You'd have the raw appeal of the chip itself, capable of new heights of computational prowess. But you also got a glimpse of the near future for more mainstream CPUs. These days? Not so much. So what to make of the shiny new Intel Core i7-6950X and its 10 mighty cores? Is it remotely relevant to gaming? While we're on the subject, are CPUs generally terribly relevant to gaming, now? And what might recent announcements regards high-performance respins of the Xbox One and PS4 consoles tell us about all this?
In the good old days, Intel had a single CPU socket and, essentially, a single CPU. Often, little separated Intel's top chip from more mainstream processors beyond clockspeed. That meant that any new uber CPU brought with it an unspoken challenge - can you match me by clocking the twangers off my cheaper siblings?
Sometimes the answer was yes and suddenly £150 and a bit BIOS butchery bought you gaming frame rates to match a £500 CPU. Now, everything is just so bloody complicated. For starters, Intel now has two sockets for desktop processors – currently LGA1151 for mainstream systems and LGA2011-V3 for high-end rigs.
The latter is arguably more a server and workstation platform re-badged for desktop use. But that's ultimately academic. What matters is that Intel's platform bifurcation makes for a total disconnect between the aspirational high-end and what most of us can afford.
Making matters worse is the shift from clockspeed to core count as the enabler of major performance increases. In other words, kiss goodbye to overclocking as a method of making cheap chips perform like their big-ticket siblings. You simply cannot make a quad-core CPU seem like something with double or more the core count with a few BIOS changes.
The good old days of turning a cheap chip into a champ with a bit of overclocking...
Of course, the caveat to all this is that games have not historically scaled well across lots of cores. By that I mean that you can add lots of cores to your CPU, but games usually haven't been able to make much use of them. The returns tend to diminish pretty rapidly beyond four cores.
The reasons for that are complicated. But it's certainly true that the incentive for game developers to code for more than four cores is slim to none when Intel's mainstream platform remains capped at four cores and AMD is failing, for the time being, to offer anything really competitive.
Anyway, that's the context for the arrival of Intel's mighty new Core i7-6950X. The headline numbers include 10 cores, 20 threads, a nominal clock speed of 3GHz and a maximum Turbo speed of 3.5GHz. Oh, and it's part of the Broadwell family, which means it's made of 14nm bits and pieces but actually last-gen compared to the newest Skylake chips for the LGA1151 socket like the Core i7-6700K.
So, yes, it's Intel's first 10-core CPU for desktops. But if you care to peruse some game benchmarks, you'll find that doesn't make for gaming greatness. In fact, it's often beaten by the best quad-core chips.
You could argue that's great because it means you don't have to worry about the 6950X idiotic $1,723 (about £1,400 in old money) price tag (for the record, Intel has also launched three further new i7-6000 chips with eight and six cores, the cheapest of which is almost affordable at $434). On the other hand, it does make you wonder if games might simply be different if the bulk of the installed base of gaming PCs had gone beyond four cores. Would there be exciting new things going on with AI, for instance?
Plenty of pins in an LGA2011, but not that many of 'em are relevant to desktop performance
Which brings us to those new consoles. Needless to say, the specification of games consoles has a huge influence on how game devs go about things. At the E3 show at the mo' details of a rather unusual mid-life refresh of the two more performance-orientated lumps, the Xbox and Playstation, are emerging.
Both essentially target virtual-reality capability. That means a huge uptick in pure graphics performance instead of the simple slimming down and lower cost that normally accompanies these kinds of refreshes. As a for instance, the refreshed Xbox, currently codenamed Scorpio, is said to be jumping from 1.23 TFLOPs of pure processing power, to over 6 TFLOPs.
On the CPU side for these consoles, things are less clear. The upcoming PS4 Neo seems to be sticking with those awful AMD Jaguar cores, all eight of them, while the CPU spec for Xbox Scorpio is also sticking with an eight-core design but the identity of those cores is unknown.
There has been speculation that Scorpio might receive AMD's upcoming high-performance Zen cores. But that seems unlikely to me. It would mean eight Zen cores plus a GPU that seems to be roughly on a par with AMD's newly announce Radeon RX 480 on the same chip. Not impossible, but certainly it would represent an unprecedented level of performance for a console. And it would be expensive.
Could Xbox Scorpio sport AMD Zen cores? Doubt it!
Whatever, if anything would motivate game developers to get good at coding for multiple cores, it's the current PS4 and the Xbox One and their fairly feeble CPU cores which are based on AMD's low-power 'Cat' architecture and thus more like Intel's Atom cores than high-performance desktop cores. In practice, developers have found this tricky and generally speaking games still don't scale well beyond four cores.
Then there's Microsoft's DirectX 12 API and the alternative Vulkan API, both of which supposedly help improve game performance scale across multiple cores. But the full implications of those have yet to be seen.
If there is cause for hope for significant CPU performance increases for mainstream PCs, it's AMD's upcoming Zen CPU which might, just might, shake things up. I'm not expecting it to go on sale for about a year, however, so it's not exactly just around the corner.
It's a complicated overall picture, then. Honestly, I don't have a great feel for the extent to which developers are being held back by CPU performance. But CPU performance has certainly stagnated in mainstream PCs over the last five years and instinctively I'm not comfortable with that. Ultimately, I find it hard to believe it's a good thing for gaming and game development innovation.