PlayStation 5 GPU: 9.2 TFLOPs with 36 custom Navi compute units

Just in time to round out 2019 we have some more specifications on Sony's next-gen PlayStation 5 console, but this time they're more GPU specific concentrating on the RDNA2-based custom Navi GPU that AMD made for both Sony and Microsoft.

Digital Foundry has analyzed a bunch of data and come to the conclusion that Sony is using a custom Navi GPU with 36 compute units, which will be clocked at around 2GHz. However, Sony will reportedly have 3 different GPU performance profiles -- with PS5 games pushing the Navi GPU to 2GHz -- but emulating previous-gen PlayStation games will see the GPU clocks doing some funky things.

This sees the PlayStation 5 and its custom Navi-based GPU rolling out with 9.2 TFLOPs of compute performance -- losing the next-gen console battle to Microsoft and it's new Xbox Series X which will have a 12 TFLOPs custom AMD solution. Sony will still use GDDR6 memory (and not HBM like the current insane rumors are leading some to believe). AMD's current flock of GDDR6-based cards in the Radeon RX 5700 series have 448GB/sec of memory bandwidth, but we could see Sony using some faster GDDR6 memory and seeing memory bandwidth hit 512GB/sec or so.

Sony's new PlayStation 5 will have its GPU clocked for "native" tasks -- aka running next-gen PS5 games, while Gen1 and Gen0 will see the custom Navi GPU dropping its clock speeds and memory bandwidth radically to "emulate" the PS4 Pro and PS4. Gen1 will see the SoC dropping its GPU clocks to 911MHz and its memory bandwidth to 218GB/sec, and just 64 ROPs in order to emulate the PS4 Pro. The regular PS4 emulation mode on the PS5 is 'Gen0' and reduces the GPU clocks to 800MHz, memory bandwidth to 176GB/sec and 32 ROPs.

When it comes to the Xbox Series X however, Microsoft is definitely going to win the next-gen console war. More on that here.