I would not be so quick to rule out Ryzen. Any next gen console will be on 7nm, so the die area for a Ryzen CPU cluster will already be much smaller than it was on Summit Ridge. Implement it using higher-density libraries and limit the frequency to, say, 2.5-3GHz, and the die area will be down again. Cut the L3 cache per cluster from 8MB to 4MB (like they already did in Raven Ridge), and there's some more die area. (Note the visibly smaller CCX in
Raven Ridge compared to
Summit Ridge.)
Yes, but K12, or the next-gen ARM core, will also be 7nm. It comes down to which solution is going to maximize die area, tdp, and cost per core efficiency which will be clearly in the favor of ARM, on the simple principle that a core architected for a specific, low power envelope, and performance target, is going to be more efficient than one that scales over a wider performance and power consumption range. There will be wasted silicon on that 2.5ghz mobile grade Ryzen. ARM will just be the neater and tighter solution.
The way K12 was described, it was never going to be significantly smaller than Zen. It was described as a "sister core" that had a "bigger engine" than Zen. That does not sound to me like a more die-area efficient design than Zen. Potentially they could
Right, at the same time it was stated it would be 10-15% more efficient which isn't a trivial difference. I expect more than that plus other advantages at the performance and power consumption target. It's true K12's size was mysterious, and I am making an assumption of a somewhat significantly smaller core. I just don't see the point of a same size K12.
At the end of the day, a console is going to be a consumer embedded, thermally limited, small CPU core device. ARM is king of small core. Just made for the application. The fact that Jaguar has no successor is a nod to the fact that ARM and RISC has unequivocally taken the small core away from x86.
Regarding many-core and economies of scale- the more cores you need to integrate into a design, the more you start running into problems with your interconnect. Just look at how messy Intel's bidirectional ringbus got as they scaled up the number of cores in Xeon (ending up with multiple separate ringbuses connected by switches), and how they had to eventually ditch it and go to a more complicated mesh fabric. And just look at the diagram of all the different links between CCX nodes in a full dual-socket Epyc system, and the list of wildly varying latencies. Just slapping more cores onto the die doesn't give you an easy free lunch, it's an extremely complicated scaling problem.
This plays highly into the case for ARM, eliminating those CCX cluster latencies and allowing a true 8 core (maybe 16 at 7nm) processor without CCX latency, and then there is the major advantage of being able to add LITTLE OS and AI (Dynamic IQ) dedicated cores. Compared to having 1 or 2 Ryzen cores just for the OS..
Economies of scale.. if you are buying 800 million or 1.6 billion cores (in case of 16 core) over the life cycle of a console, this is a big billion dollar plus decision. That's why a small die savings/tdp advantage per core is a big deal. If the ARM core is smaller and simpler in its design than Ryzen it has a clear advantage. Even more advantageous if it's a stock ARM core that's simpler to shrink for future nodes and has far greater economies of scale.
Yes, the Switch is using ARM- it's a Nintendo mobile device. They've used ARM since the Gameboy Advance. Mobile is ARM's speciality.
I see a trend.. Nintendo was first in the console space with PPC, Rambus, AMD, and Edram, and the others followed. K12 was about widening the range of application for ARM, moving beyond mobile..
Yeah, emulation would be a possibility- the Jaguar isn't a massively powerful CPU. But given all the work Microsoft have invested in backwards compatibility, writing emulators for their x86 platform, making it a key part of their brand in this generation... it feels like madness to throw that all away and start from scratch again for ARM, for fairly minimal benefits.
Also remember that Microsoft want to sell games across both PC and XBox, in a unified Microsoft store (XBox Play Anywhere). It's a big part of their plan to shift gamers over to buying through the Microsoft store on PC, and having the PC and console run entirely different instruction sets will just complicate matters for them.
Yeah, but Microsoft has also made significant investment in Windows on ARM and converting Win32 programs into Store Apps. Maybe all those X360 b.c. titles could be converted to UWP and be playable across all Windows devices. At some point in the future Windows gaming on ARM is going to be a more significant thing. I see ARM XBOX as a part of a large strategic convergence strategy. Is it going to be messy W.I.P, yes, but that is the state of Windows now.
Windows on ARM and x86 emulation themselves are even a temporary solution until ARM devices are migrated to Windows Polaris/Core OS. That's an insane amount of work going to a transitional migration strategy.
Microsoft actually made a recent change to their store game policy allowing UWP games to have both ARM and x86 containers. The beginning of a strategy is there..
I could see Sony moving away from x86- they have not staked so much on backwards compatibility in this generation. The PS4 Pro is backwards compatible, but arguably that's just the same generation with slightly boosted performance. And they're not trying to encourage developers to cross-develop for PC. Maybe an NVidia SoC with A75 cores and an Ampere GPU would be just the thing for them.
Definitely, and ARM Ares next-gen A75 'big core' successor is unveiled in May at Computex. I would watch that with interest as that might be the next-gen console core right there. NVIDIA is a much more powerful company than it was last time console contracts were signed and the console gaming market much bigger and significant than expected. I fully expect NVIDIA to make a play for a console contract even if they were just going for a high profile PR win.
Lastly..., where is ARM going to be in the future relative to x86 throughout the life cycle of console? ARM may have expanded its reach significantly, or perhaps AMD might not even be an independent business anymore. Ryzen doesn't have a long term guarantee of success of an architecture either.