As a followup to my previous piece on cloud and the console wars, and in honor of the Xbox’s 10th birthday, I offer the following speculation on the CPU that will power the Xbox 360?s successor.

Nothing beyond the vaguest rumor and speculation are known about the upcoming consoles from Microsoft and Sony. We only recently got public confirmation that next-gen console efforts are, in fact, in progress at both companies. So I have no idea who Microsoft will turn to for the CPU that will power the “Xbox 720?, as the Xbox 360?s successor is popularly called in the gaming press. But I do have some ideas about the kind of processor that might work the best. In short, I think Microsoft should use an SoC that’s based on ARM’s recently unveiled A15 core.

With its next-gen console, Microsoft needs to keep costs low and reach profitability as soon as possible; its unlikely that the company will risk the ire of shareholders by taking years’ worth of losses on its upcoming console, the way that it did with the Xbox 360. So this constraint pretty much rules out a completely bespoke console CPU like the current generation’s Xenon part.

Instead, Microsoft is more likely to use a relatively unmodified version of an off-the-shelf design, the way that it did with the first Xbox, which had a tweaked version of a commodity Intel CPU inside. There are two options for such a strategy: ARM and x86. Let’s take a look at both.

The x86 options

The rumor mill has Microsoft looking to AMD for at least the GPU part of the unit, and the fact that AMD will have a new generation of CPU/GPU “Fusion” parts on the market in the 2013 to 2014 timeframe that the 720 might launch has prompted some speculation that AMD has won the contract for the CPU, as well. Some speculate that the 720?s hardware will be based on AMD’s “Krishna” processor, which pairs multiple, low-power “Bobcat” CPU cores with a DirectX 11-class on-die GPU. The Krishna option is much more believable than other reports, which claim that the console will use the Bulldozer-based “Trinity” processor. The performance of Bulldozer’s launch lineup is fairly mediocre in gaming and 3-D benchmarks, due to the fact that every two Bulldozer “cores” share a single FPU. Given cost, performance, and power considerations, a Bobcat-based design makes more sense; Bobcat cores are smaller, and AMD can use them to put more FPU hardware per core around the on-die GPU.

As for Intel, there are no rumors that I’m aware of putting the chipmaker in the running for the CPU slot in the 720. Since Intel’s Larrabee project bit the dust, it has been clear that the GPU would go to either AMD or NVIDIA, and as I said above all signs are pointing to AMD. There is an outside chance that Intel could provide the CPU for the 720, but Bobcat is probably a better fit for the console than anything in Intel’s lineup. Bobcat is a sort of middle ground on the performance/watt spectrum between Atom and Intel’s mobile Core i5 family. Bobcat’s decode execution resources are close to Atom’s, but the former is an out-of-order design so it gives much better per-thread performance. This spot between Core i5 and Atom is probably the sweet spot for a CPU core in the next generation of consoles.

Incidentally, there is another commodity CPU core option that occupies the exact same performance/ watt spot that Bobcat does: ARM’s A15.

The ARM option

Microsoft might find ARM to be an attractive console CPU option because it will give the company a single architecture across its console, mobile phones, and desktop offerings. With regard to the last, the fact that Windows is being ported to ARM suggests that Microsoft’s next-generation console OS could conceivably be just another Windows 8 flavor. With the same OS across the PC and the console, and the same architecture across PC, console, and mobile, Microsoft could achieve unheard-of levels of cross-platform integration.

If Microsoft goes the ARM route, I would expect them to use the A15 core, which will be entering the market in 2013. As with Bobcat, A15 is a relatively simple core design that is quite similar to Intel’s Atom, except for the crucial fact that it is out of order, which gives it an edge performance. A quad-core A15 chip would make for a low-cost, low-power CPU that gives significantly more performance than the current Xbox 360 CPU.

The ultimate A15-based console chip would, of course, have an on-die GPU. ARM’s Mali T658 is a DirectX 11-class GPU that will be out in 2014, so it or some variant of it is the most obvious candidate for the 720?s GPU.

Or, another option would be for AMD to put down four A15 cores on the same die as an AMD-designed console GPU. This possibility is quite far-fetched, but a report from Semi Accurate suggests that it may not be entirely out of the question. (BSN begs to differ, however.)

So what’ll it be?

As the title of this post suggests, I think ARM’s A15 would be a great fit for the Xbox 720. It fits within the modest cost, performance, and power constraints that I imagine Microsoft will want for the next-gen console. (Again, I don’t believe they’ll repeat the last generation’s mistake of going all-out with something large, hot, expensive, and completely custom.) And the A15 would also give Microsoft architectural unity across every platform where it offers an OS.

But all of that being said, I’ve learned over the years not to fight the rumor mill, and the rumor mill has fingered AMD as the winner this time around. So if I had to bet money on the identity of the Xbox 720?s CPU and GPU hardware, I’d bet that it will be some variant of AMD’s Krishna processor. While Krishna won’t bring the console up to parity with an desktop gaming PC of 2013 or 2014, it will nonetheless fit the bill nicely.

Then again, apart from the rumor mill, the other thing that I’ve learned not to discount is the power of legacy software, which is why PowerPC is still in the running as a possibility for the 720. This report at Semi Accurate claims that AMD and IBM are working on an SoC that combines PowerPC cores with an on-die AMD GPU — in other words, a straightforward iteration of the CPU/GPU combo SoC in the current Xbox 360. So IBM and AMD could theoretically take the current SoC, double the number of cores, add in a DX11-class GPU, and put a high-bandwidth link between the CPU and GPU to get a much faster console that might fit within its conservative cost, power, and performance envelope.

Homepage photo: ed10vi/Flickr