feat: add RTX 5090 Blackwell GPU support (sm_120)#330
feat: add RTX 5090 Blackwell GPU support (sm_120)#330rishikanthc merged 1 commit intorishikanthc:mainfrom
Conversation
Add support for NVIDIA RTX 50-series GPUs (Blackwell architecture) which require CUDA 12.8+ and PyTorch cu128 wheels due to the new sm_120 compute capability. Changes: - Add configurable PYTORCH_CUDA_VERSION environment variable to control PyTorch wheel version at runtime (cu126 for legacy, cu128 for Blackwell) - Update all model adapters to use dynamic CUDA version instead of hardcoded cu126 URLs - Update Dockerfile.cuda.12.9 for Blackwell with CUDA 12.9.1 base image, PYTORCH_CUDA_VERSION=cu128, and missing WHISPERX_ENV/yt-dlp - Update Dockerfile.cuda with explicit PYTORCH_CUDA_VERSION=cu126 - Add docker-compose.blackwell.yml for pre-built Blackwell image - Add docker-compose.build.blackwell.yml for local Blackwell builds - Add GPU compatibility documentation to README Fixes: rishikanthc#104
| RUN curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp -o /usr/local/bin/yt-dlp \ | ||
| && chmod a+rx /usr/local/bin/yt-dlp \ | ||
| && yt-dlp --version |
There was a problem hiding this comment.
Shamelessly copy/pasta'ing this section from the main Dockerfile.cuda: https://github.com/rishikanthc/Scriberr/blob/main/Dockerfile.cuda#L81-L84
| WHISPERX_ENV=/app/whisperx-env \ | ||
| APP_ENV=production \ |
There was a problem hiding this comment.
Making this sync up with: https://github.com/rishikanthc/Scriberr/blob/main/Dockerfile.cuda#L57-L58
|
Note: I haven't yet tested with all the different adapters, but I feel like it should just work? Would take any advice on what to test locally. :-) |
|
Moving to draft while I doing some local testing. |
|
I'm so dumb. I did not read your PR properly and replied asking you if you can do what you already did xD APologies. Deleting my old comments to save me some embarassment :P And thank you so much for your contribution. Appreciate the help and support :) |
|
Haha! Thanks @rishikanthc! :-D |
2 participants
I came across #104, which led to me believe the most recent release should work with my RTX 5090. However, then I saw #273, and then saw this comment... so, this PR is intended to make
scriberr-cuda-blackwellimage that will work with the newer RTX 50-series. Ideally, that will allow folks with newer hardware to opt-in while not breaking folks with older hardware.Ultimately, I would like to get that image published as part of these releases [so I can selfishly use it and not have to host it somewhere 😜], so I hope this helps in that effort 😇.
Add support for NVIDIA RTX 50-series GPUs (Blackwell architecture) which require CUDA 12.8+ and PyTorch cu128 wheels due to the new sm_120 compute capability.
Changes:
Fixes: #104