AMD quietly funded a drop-in CUDA implementation built on ROCm: it's now open-source




The story behind the gentleman from Intel to Amd is interesting.

In french we say it’s “nouveau” information. :wink:

Thanks for sharing. :smile:

1 Like

It’s so freakin’ huge man!
If someone will secretly get a drop-in replacement of PhysX - i swear i’ll throw away my novidya and never look back!

1 Like

Again, I’m thinking maybe BOINC would possibly have good code to help science in some projects … the modifications could add a lot of Graphical processors to some free software operating systems that are contributing only with there Cpus.

A lot of projects accept Gnu/Linux but we can’t contribute freely with our gpus …

I’ll keep an ear open. Maybe the futur has a good path for that :smile:

1 Like

While that’s great news, it’s not like NVIDIA are the cool guys that will appreciate this…

I can easily imagine a future where the API is dynamically generated at runtime with random namespaces (think how JS is used for cancerous ads to avoid static rule blocking). Ofcourse that would make development an absolute nightmare, but who cares? …certainly not NVIDIA :stuck_out_tongue:

We get CUDA support for AMD in Blender quicker then we get Vulkan support.

Curious how much it’s will boost AMD GPUs performance on contrast to Nvidia.

OpenSource NVENC/NVDEC for h.264/265 would also be nice.

Imagine something far worse, API written in JS, with OS running inside of a browser via WASM and it’s all hosted in a cloud, inside of a docker.
And user device acts as just a monthly subscription thin-client.

1 Like

I’m sure THEY :male_detective: :alien: :dragon_face: would love that! :rofl:

P.S. Although it’s unlikely, for performance reason…but still they’d love that :rofl:

1 Like

Hopefully by then people remember how good it was to use an AMD GPU and more are motivated to improve ROCm

That was fast… :rofl: :rofl: :rofl: :rofl:


man… sad. I guess this is what one should expect from them, though.