Scratchpad is fast RAM, as fast as cache, but without the automated data fetching of cache. You have to manage what data is in there yourself.
Next Generation Hardware Speculation with a Technical Spin [2018]
- Thread starter Tkumpathenurple
- Start date
-
- Tags
- 5 bytes
- Status
So, even the gamecubes cache is like this groupshared mem in compute shaders?
I'm not sure the industry would respond favorably to Sony releasing a console with PS2 style exotic hardware in 2019/2020. At least not when every other console maker has moved towards standardization. Unless it's effortless to get better performance with the hardware, of course, but that wasn't the case with PS2.
...I hope.Turing is not only a RTX GPU architecture, it is a great rasterizing GPU architecture. The best actually...
and funny the ordered prefix sum is on AMD gpu, it is used in Dreams. You just need to read SIGGRAPH 2015 Alex Evans presentation for verification:
I am not sure the functionnality is exposed inside Direct X
If we had the AMD patent inside the GPU of AMD next generation console and primitive shader is as good as Mesh shader... It will be a great generation.
and funny the ordered prefix sum is on AMD gpu, it is used in Dreams. You just need to read SIGGRAPH 2015 Alex Evans presentation for verification:
I am not sure the functionnality is exposed inside Direct X
If we had the AMD patent inside the GPU of AMD next generation console and primitive shader is as good as Mesh shader... It will be a great generation.
Last edited:
Tkumpathenurple
Veteran
Potentially very interesting... potentially....
anexanhume
Veteran
Some of my fondest memories are goldeneye and Mario Kart. I don’t know how I dealt with 1/4th of a 480i image in a FPS, but I did. I’m on board with bringing it back.
I am worried the micropolygon patent is probably post GCN based on the date of filing. That and their “Super-SIMD” design.
I assume devices naturally means PSVR2. I wouldn’t get too excited beyond that as an additional device.
It is just a wish, I am not sure it will be inside next generation GPU...
IIRC shader model 6 did not include all he functionality he was looking for from reading his other comments. But I could be wrong and he could have been referring to a separate topic. This one, GPU driven rendering, is fairly specific.
then again, his topic does is not explicit in using things like executeIndirect. IIRC he covers the topic though, but found it not capable enough yet.
anexanhume
Veteran
Source had reliable Zen info.
So according to https://wccftech.com/amd-confirms-new-7nm-radeon-graphics-cards-launching-in-2018/ the 7nm Vega itself could be a 20.9 tf monster at 336mm² in size which is half the the size of Turing gpu and even smaller than Pro's 348 mm2, 7nm really does wonders I suppose. What would this translate to a potential PS5 gpu? 13-14tf likely?
sky's the limit I guess if you want to look at it that way.
Something to consider is that the 2070 is actually slower than a 1080TI and the 1080TI has 11.5TF of power.
Why this is an important consideration is because Nvidia has dedicated silicon for RT, and it only exists in the RTX line which is 2070+.
So in theory it's possible that PS5 (have something greater than 11.5 TF) could come out and only have its eyes set at a solid 4K machine, but that's all it would be able to do, and it's clear that nvidia is willing to give that crown up once it started walking down the RT path.
Not entirely sure if having more FLOPs is the answer here is all i'm saying.
I'm definitely leaning heavily towards next gen having some RT capabilities, at the very least, I expect this for Xbox Scarlett.
anexanhume
Veteran
Of course, the usual grain of salt for WCCFTech rumors.
Vega 64 achieves 12.665 TF in 486 mm^2, giving a TF/mm^2 of 0.026.
These numbers would be a TF/mm^2 of 0.062, which is a growth of 2.39x.
TSMC has a shrinking factor of 70% for 16nm->7nm. Keep in mind this is likely 16FF+ to 7nm SoC, not 7nm HPC. 2.39x is a roughly 58% area reduction, iso clock, so these numbers seem plausible to me.
Say a next gen APU has 200mm^2 available for graphics and memory controller. That gives 12.44TF, assuming they can hit the same clock. This also assumes Navi has a similar density to Vega. GloFo 14nm and TSMC 16FF+ are pretty close from what I remember.
Last edited:
Weren't there rumors of PS5 SDKs containing 20TF GPUs?
uhh really? link? I'd read that.
So last time I saw an alpha kit for PS4, I was told was just a single GPU card in the PC that he would not open for me. That being said, it wasn't a ground breaking GPU. If they were to give out a single GPU today, I'm surprised it would be able to hit 20TF considering what's available on the market right now.
Honestly, I can't remember where I read it. I'm not even 100% sure if it was a single or dual GPU configuration that was mentioned. If I find either, I'll link them here.
anexanhume
Veteran
I can only find circular forum references.
Example: https://www.gog.com/forum/general/ps5_rumors_amd_navi_7nm_architecture
Could be the same rumor, but I'm leaning more towards SemiAccurate or TechRadar when I heard this rumor.
- Status
