not me. but i can see that scenario. but is depends, what kind of advantages it will offer? we have to remember, that it doesn't targeting casuals.
PS5 Pro updated rumors/leaks & technical/specs discussion |OT| PS5 Pro Enhanced Requirements Detailed.
- Thread starter Gamernyc78
- Start date
Insider Gaming Details a New PSSR Solution (PlayStation Super Resolution) which will be coming to PS5 and Future Consoles:
PlayStation's Spectral Super Resolution (PSSR) is Aiming for 4K 120FPS & 8K 60FPS
PlayStation's new Spectral Super Resolution (PSSR), which will be added into the PlayStation 5 Pro is aiming for 4K 120 FPS and 8K 60FPS.
insider-gaming.com
This math is not possible for the PS5 pro.
For the PS5 you have:
36 CUs x 64 shaders per CU x 2.23 GHz x 2 instructions per clock = 10.275 tflops
Assuming the PS5 pro has working CUs, it will pull
60 CUs x 64 shaders per CU (basically RDNA standard) x 2.18 GHz G x 2 instructions per clock = 16.742 tflops
You have to double this for fp16, but the number we always talked about is fp32.
Not that this matters as teraflops are a single measure amongst an ocean of variables that impact performance.
I'm much more interested in the supposed RNDA 4 features, which I think are just more bull from supposed leakers.
buzz words. fuck off sony with that. 4k120, 8k60fps
Insider Gaming Details a New PSSR Solution (PlayStation Super Resolution) which will be coming to PS5 and Future Consoles:
PlayStation's Spectral Super Resolution (PSSR) is Aiming for 4K 120FPS & 8K 60FPS
PlayStation's new Spectral Super Resolution (PSSR), which will be added into the PlayStation 5 Pro is aiming for 4K 120 FPS and 8K 60FPS.
Insider Gaming Details a New PSSR Solution (PlayStation Super Resolution) which will be coming to PS5 and Future Consoles:
PlayStation's Spectral Super Resolution (PSSR) is Aiming for 4K 120FPS & 8K 60FPS
PlayStation's new Spectral Super Resolution (PSSR), which will be added into the PlayStation 5 Pro is aiming for 4K 120 FPS and 8K 60FPS.
Details here are scarce, but it looks like my prediction was correct:
Here's what they're saying on the first example:
So for the same/similar output image quality in a 4K presentation, the PS5 Pro renders at 2560*1440 (3.7MPix) where the PS5 renders at 3200*1800 (5.7MPix).
We now know Sony is aiming for PSSR to get the same IQ as other temporal upscalers like FSR2 or TSR but at 64% of the base resolution which is quite impressive.
It's bit like saying we're getting FSR2 Quality IQ but with FSR2 Performance framerate.
Again, I’m starting to believe these hacks are purposely posting misinformation to create drama and engagement artificially, or someone needs to explain how they came up with the 33.3 TF.
Also, GPU clocks are lower? Very odd.
Its 33tf in FP32, if yall naysayers want people to take tall seriously on your lowball teraflops claims then you need to list the dual issue teraflops for the current gen AMD RDNA 3 GPU's. Dont just single out PS5 Pro, please enlighten us with the true lower teraflops numbers for the 7900XT, 7800 ect..
Cuda cores X gpu speed X 2 (for single precision) divided by tera
7680 x 2180 x 2 / 1.000.000 = 33,48tf
- 30 WGP = 60 Compute Units
- 64 ALUs per Compute Unit on RDNA: 60x64
- FP32 ALUs in RDNA3 are 2-wide (or double-pumped): 60x64x2
- ALU operations can be MUL (multiply), ADD (sum) or FMAD/MADD (fused multiply add). Since multiply and add are counted as 2 operations, you double the count again: 60x64x2x2
- To get total theoretical FLOP output, you multiply that by the clock. TFLOPs = 60x64x2x2 x clocks.
- So, to achieve the GPU clock for the rumored 33.5 TFLOPs FP32, we calculate clock = (33 500 000) / (60x64x2x2). Therefore clock is 2181 MHz.
It's indeed odd. I think 2181MHz is base clock without boost, like 2GHz is the base clock for the PS5's GPU.
It that's the case and the GPU averages at say 2600MHz, the actual throughput will be closer to 40 TFLOPs.
Game engines would need to be rewritten for that new feature though, right? And with the base PS5 not having it, it is dead weight or maybe not even included to cut costs.
Thanks for the explanation, I knew the math to get to 16 but was not sure why it was 2x that number.
OP
OP
No, it isn't, unless the console has 128 shader per compute unit. I've done the math, just look for my previous post.
Edit: it's 16.7 at fp32 and 33.4 at fp16.
I'm going to quote my post again and clarify: unless the PS5 pro has 128 shaders per CU or has a GPU running at 4+ GHz it will have 16.7tflop at fp32. You double teraflops fp16, not halve.
Isn’t dual-issue very limited in current AMD cards? Unless it works like magic and developers don’t have to rewrite code for it, it is useless on the PS5 Pro, hence, 16 TF is the most accurate number.
The number of instructions in AMD's ISA that support the per-ALU dual-FP32 feature is limited, yes. The guys over at Chips and Cheese couldn't find a way to measure the dual-issue output until they tested Nemes' compute benchmark for Vulkan. I think they don't know what Nemes is using in her benchmark to trigger the ability but the feature is there nonetheless, so it can be used. Assuming it all fits in the caches and registers, of course.
It would probably be dead weight in the PC space except for stuff they fully control like driver-level features (FSR1, AFMF, etc.) or wherever they can hand-tune driver optimizations. Sony however owns the platform and the SDK so they can implement these optimizations directly into the compiler, and they can automate a list of optimizations for their 1st party engines as well as the widely used ones (UE5, Anvil, etc.).
Again, thanks for the info; I need to catch up on what is happening in the PC space.
Anyone get the impression that Inside Gaming might have leaked these things to MLID and then corroborated it afterwards, so they could say it wasnt them that leaked it....
