I am really looking forward to having a viable alternative to NVIDIA. I would love buying Intel or AMD and being able to enable CUDA support on the GPU out of the box.
gaylord_fartmaster@lemmy.world
on 02 Jan 20:14
nextcollapse
Me too. I recently switched from an RTX 2080 to a 7900 XTX, which is way more powerful for games, but local LLM performance tanked without CUDA.
geneva_convenience@lemmy.ml
on 02 Jan 22:38
nextcollapse
NVIDIA skimps out on VRAM needed for deep learning. If AMD could compete it would be huge as they often include more for cheaper.
enabling unmodified CUDA applications to run on AMD GPUs at near-native performance, the ZLUDA atop AMD HIP code was made available and open-source following the end of the AMD contract
Trouble is… HIP doesn’t support all of AMDs GPUs. It’s only 7900s in the consumer line-up.
NoLifeGaming@lemmy.world
on 02 Jan 23:33
nextcollapse
I hope this becomes a thing so we can have more competition
mindbleach@sh.itjust.works
on 03 Jan 04:38
collapse
Nvidia’s anti-competitive monopoly should be broken up over this API.
The fact AMD was visibly scared to even be associated with this project is a big fucking hint to regulators that CUDA is vendor lock-in tool which Nvidia guards at all costs.
apt_install_coffee@lemmy.ml
on 05 Jan 11:53
collapse
Is that why they prevented it from being open sourced? I thought I read a while back that they just wanted to keep the code in-house.
threaded - newest
I am really looking forward to having a viable alternative to NVIDIA. I would love buying Intel or AMD and being able to enable CUDA support on the GPU out of the box.
Me too. I recently switched from an RTX 2080 to a 7900 XTX, which is way more powerful for games, but local LLM performance tanked without CUDA.
NVIDIA skimps out on VRAM needed for deep learning. If AMD could compete it would be huge as they often include more for cheaper.
Trouble is… HIP doesn’t support all of AMDs GPUs. It’s only 7900s in the consumer line-up.
I hope this becomes a thing so we can have more competition
Nvidia’s anti-competitive monopoly should be broken up over this API.
The fact AMD was visibly scared to even be associated with this project is a big fucking hint to regulators that CUDA is vendor lock-in tool which Nvidia guards at all costs.
Is that why they prevented it from being open sourced? I thought I read a while back that they just wanted to keep the code in-house.