| CrazyGPU said: Also, let suppose PS5 averages 9.7 Teraflops because of variable frecuency , the XBOX has 25% more compute power. Then you would need 25% more bandwith to take data to the cores. And 560 is 25% more than 448. So maybe there is the balance and getting the expensive 18 Gbs chips to get PS5 to 576 GB/s doesn´t make much sense since 448 is balanced with those 36 shader cores and more bandwith would not translate in that much performance. Just guessing here. |
Lol.. that's cheating :)
First we can't downclock the PS5 GPU and leave the XSX GPU as is. Further more the whole variable frequency doesn't even work that way being that the CPU would ever need to be tasked enough for it to be running at its max clock. I actually expect devs to pretty much lock the CPU to like a 3.2Ghz clock to be able to run the GPU at its max clock under max load whenever they need it to.
But let's get to the numbers, and fairly this time.
PS5: 448GB/s
CPU+ Background OS state = ~ 48GB/s (that's more than what you would get from a PC with dual-channel 2666Mhz DDR4)
GPU = 400GB/s /10.3TF = 38.8GB/TF
XSX: 336GB/s + 560GB/s
CPU + OS = 336GB/s ( this is even waaaay more than the CPU would ever need but nothing can be done here cause the 3.5GB of RAM left over here after the OS has reserved its share would probably get saturated by the CPU bound tasks anyways)
GPU = 560GB/s / 12.1TF = 46.3GB/TF
The PS5s GPU just looks like something that would become bandwidth starved to me. Yes, the bandwidth difference is about the same as the total TF difference (~17%) but this is one area where you want to have as much bandwidth as possible. If they had gone with 16Gbs chips they would have at least had 512GB/s total and if we take the same 48GB/s out for the CPU/OS, they would have ended p with 45GB/TF.
Strange choice if you ask me unless of course, they know something that we don't or the CPU tasks use far less bandwidth than I have listed here.
