So recently got a Radeon Pro Duo(Fiji) and was doing some testing with it Vs my old NVIDIA GV100 in TopazLabs Video Enhance to upscale some old cartoons and movies and it seems that the NVIDIA Card gets 0.24FPS therefor taking 2hr 14min to upscale a 20min 720x480 video → 1920x1080, and the Radeon card getting over 2fps and only ETA of 54Mins per episode at the exact same settings and even the same episode.
any ideas why this would happen? ive got CUDA + Quadro drivers installed when testing nvidia and Radeon Pro Modded Nimez Drivers when testing the Radeon card. bare in mind the app doesnt take advantage of mGPU so i doubt its the fact it has dual GPUs on the AMD card that its beating it.
Old AMD cards were based on GCN architecture which was good for compute workloads and far outpaced NVIDIA in those areas. (at the time)
These days I think NVIDIA has some advantages with their CUDA stuff as companies have deployed it into software whereas AMD is still getting hardware level compute supported (I think Maya got something recently)
Interesting! It’s cool to see AMD actually be on top for once, I’ve used the Mi100 and W6800 Radeon pro GPUs for work and benchmarking our CFD and AI workloads and it seems Nvidia always wins when we use our RTX A6000 and A100s so thought that nvidia would still be on top back In the day too!
THIS…
Also GCN lasted insanely long time, so every new driver adds extra improvements to the old cards as well…
This is where this “amd ages better” thing came from.
HD7870 is while not exactly the same, similar architecture to RX580, even in some ways RX5700XT… At this point you have in general 8 years of driver improvements. On top of that the architecture itself being good at compute and here we are…
If someone makes a benchmark today R9 280X would probably me super close to GTX 780, despite being different tiers…
Ah Yes, this is no surprise. AMD used to destroy Nvidia in compute workloads.
In some ways they actually still do, its just that their Instinct cards have gone after a different niche vs Nvidia. So today they are typically used in different sectors. However the recent Instinct card that went into “Frontier” which is now the worlds fastest super computer (by a factor of 5) is also the worlds fastest compute based accelerator (not technically a GPU).
