MS reveal of ballpark of Scarlett

[Pemalite]

trasharmdsister12 said:

DonFerrari said:

Using your best judgement on the gain in efficiencies and architecture, minding that Tflops is quite pointless as well. At which Tflop count more or less for the architecture of PS5/XB4 we would reach about 4x more performance than X1X?

We can do a very rough estimate with some basic math and assumptions. Let's assume GCN 1-2, 2-3, 3-4, and 4-5 all have a roughly 7.5% efficiency advantage while 5 to Navi (RDNA) has 25% as listed above by Pema. 

6 TF * 0.925 * 0.925 * 0.925 * 0.925 * 0.75 = 3.29 TF

So we're looking at roughly a 3.3 TF Navi GPU to reach similar performance to the XB1X, theoretically. 4x that is roughly 13.2 TF. 

This is all extremely rudimentary and depending on other hardware changes made by MS/Sony, I'd wager anything between 12 and 14.5 TF would be the range for 4x. 

That is probably a guesstimate that is as good as any.

One thing we should remember is that... These CPU's are going to be the largest step-up generation over generation that we have seen for a very long time... So some tasks like particle effects and so on could be shifted back from the GPU and done on the CPU, freeing up more GPU resources.

drkohler said:

DonFerrari said:

But 12Tf is probably where the "majority" would expect both systems to hover about and would give the 4x gain of your calculation.

Good luck with that. The new Radeon 5700XT is 9.75TFlops and draws 225Watt (probably a lot more with boost clock temporarily enabled). People want a console SoC with 12TFlop? That's 300W, not even counting the cpu which certainly draws at least 35W (unless it is clocked way below 3GHz).

Some people (particularly on game forums) are simply delusional when it comes to "TFlops".

7nm isn't a refined process yet, it's still early days, RNDA isn't a perfectly balanced architecture either.

A 12 teraflop console GPU is entirely possible at 7nm for 200w~. - But it's actual gaming performance will be substantially less optimal than Navi.