What do yo think will be the hardware specifications of PS5 if it arrives arround 2019-2020?

[Trumpstyle]

CrazyGPU said:

Trumpstyle said:

Yes about 2x more gpu performance than xbox one x gpu, but for PS4 it should be upwards 8x. Both next-gen consoles will be above 10 Teraflops and DF have run tests showing NAVI gpu architecture having 50% better TF compared GCN 1.0 (ps4 gpu). Just enough for Sony to do native 4k/60fps for all their PS4 exclusive games.

CPU performance will probably be lower, keep in mind Gddr6 has higher latency than DDR4, they very likely cutting of GameCache and around 3,2ghz clockspeed. So around 4-5x.

As Permalite would love to say, there are not better teraflops, but more efficient hardware centered on gaiming. Teraflop is the same, millon of a million of floting point operations. But the feeding of the execution units improved, with better architecture, caché and bandwith. Anyway, everybody undestand your point there, but I would say that the real performance is improved 50% given the same number of teraflops compared to VEGA 64. Better performance than you predicted a couple of years ago, so you might be happy. 

I don't think anyone could had foreseen that the next-gen consoles would be unreal monsters, my first post was actually navi 9TF but with vega 56 performance so it's quite off. But 10TF for both is close to geforce 2070 super, most optimistics predictions were 12-13TF with vega64 performance so even they were off :)

Pemalite said:

CrazyGPU said:

As Permalite would love to say, there are not better teraflops, but more efficient hardware centered on gaiming. Teraflop is the same, millon of a million of floting point operations. But the feeding of the execution units improved, with better architecture, caché and bandwith. Anyway, everybody undestand your point there, but I would say that the real performance is improved 50% given the same number of teraflops compared to VEGA 64. Better performance than you predicted a couple of years ago, so you might be happy. 

Graphics Core Next is an extremely compute centric design, when AMD designed the Architecture they had forward projections that games would become more compute heavy rather than pixel/texture fill-rate heavy or more geometry demanding.

Obviously history plays out differently... So AMD spent the next 8+ years "modifying" Graphics Core Next in order to bolster geometry throughput, improving the ROPS, TMU's, increasing bandwidth via delta colour compression and more... But GCN just had a ton of architectural bottlenecks that held back it's gaming potential regardless of AMD's attempts.
Didn't help that nVidia was making massive strides in efficiency starting with Kepler, making a big leap with Maxwell and refining that with Pascal, which undercut AMD's efforts even more in comparisons.

In short... The 5700XT's peak single precision floating point capability is down 23% from Vega 64, but the more important aspect of the GPU for gaming... Pixel Fillrate had an increase of 23%, ROP/Texture Mapping Unit/Geometry performance increased by 28.7% (And that is before we calculate in architectural efficiency improvements so likely higher!)... And it's no wonder it's able to beat Vega 64 in Gaming.

In scenarios that are compute-heavy, like GPGPU though? Vega will still beat Navi hands down, but that is just because Vega isn't being bottlenecked by components like the geometry engines or ROPS in those instances and thus it can make actual use of it's theoretical TFLOP potential.


I don't expect someone like Trumpstyle to ditch the "TFLOP" argument anytime soon, but people are slowly starting to wake up to how meaningless TFLOPS are for gaming which is really good to see, if only they knew that a console generation ago!

Hmm let's w8 for Navi 12, it should have 80CU with hbm2e and easily beat your Vega64/7 in compute!

No problems