The die-shot of Sony’s PS5 SoC has finally been analyzed by a third-party, revealing some unexpected details, while confirming the rest. For starters, it looks like the GPU on the console isn’t a full-fledged RDNA 2 design, and instead uses the RDNA 1.1 design which is basically the same as RDNA 2 with the exception of the ROPs.
Update: While the WGP design does seem closer to RDNA 1 along with the render backends, it’s unclear whether the FPUs on the PS5 are the same as the ones on the consumer-grade Zen 2 chips. Either way, the FPUs appear to have been shaved with half the register size in comparison to a standard Zen 2 core.
The lack of Infinity Cache on both the consoles is no surprise as that would have made both the consoles economically unfeasible. Other than the Render Backend, the arrangement of the Dual-Computer Unit (WGP) subarray is also similar to the older RDNA 1 design rather than RDNA 2. I’m guessing RDNA 1.1 was identical to vanilla RDNA in this respect.
The Zen 2 CPU cores on the PS5 are a bit surprising. It looks like the Floating Point Pipeline has been cut-down into 128-bit units rather than 256-bit which is the standard with both Zen 2 and Zen 3. Although this design should still support AVX256 and wider FP instructions, it will be a fair bit slower than the PC-grade Ryzen 3000 and 5000 CPUs. This appears to have been done to reduce power consumption, albeit at the cost of performance as many modern games utilize AVX2 code.
As you can see in the above image, the FP capabilities of the PS5 console are roughly on par with the older Zen core, rather than the Zen 2 architecture. It’s unclear whether the Xbox Series X features full-fledged Zen 2 cores or a custom design similar to the PS5.