You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We currently list specific driver versions in our requirements (example). Usually it is the initial driver version associated with a CUDA version. However, many of these driver versions are not supported anymore (eg 450 or 460).
[CUDA 11.2](https://developer.nvidia.com/cuda-11.2.0-download-archive){: target="_blank"} with a [compatible, supported driver](https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers)
[CUDA 11.4](https://developer.nvidia.com/cuda-11-4-0-download-archive){: target="_blank"} with a [compatible, supported driver](https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers)
[CUDA 11.5](https://developer.nvidia.com/cuda-11-5-0-download-archive){: target="_blank"} with a [compatible, supported driver](https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers)
[CUDA 11.8](https://developer.nvidia.com/cuda-11-8-0-download-archive){: target="_blank"} with a [compatible, supported driver](https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers)
[CUDA 12.0](https://developer.nvidia.com/cuda-12-0-1-download-archive){: target="_blank"} with a [compatible, supported driver](https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers)
I like the idea of us not needing to frequently duplicate whatever the driver support is from another space. We might want to keep a single location with actual numbers for our easy reference (so we don't need to click through 5 links to see the 5 minimum supported drivers).
Thoughts on keeping the install page with specific driver vers, but all libraries use the compatible, supported driver terminology? The docs repo is a lot easier to push updates to than the libraries is my reasoning.
Thoughts on keeping the install page with specific driver vers
Well, technically, due to enhanced compatibility, any driver >=450.80.02 works with any CUDA 11.x, so listing out driver versions for each CUDA version isn't very useful in my opinion. In other words, currently we list that CUDA 11.8 requires driver >=520 but that's not actually true.
Maybe just list the minimum currently supported driver versions for each major CUDA? IE >=470 for CUDA 11.x & >=525 for CUDA 12.x? We would still need to update these driver versions as they drop out of support (525 drops in December for example).
We currently list specific driver versions in our requirements (example). Usually it is the initial driver version associated with a CUDA version. However, many of these driver versions are not supported anymore (eg 450 or 460).
Since in general, RAPIDS is agnostic to the driver version (CI tests an "earliest" and "latest" driver), we should update our references to driver versions to be more generic and reference a doc such as: https://docs.nvidia.com/datacenter/tesla/drivers/#cuda-drivers
For example:
Here is a good search for all of the usages: https://github.com/search?q=org%3Arapidsai+AND+%28450+OR+460+OR+470+OR+495+OR+525%29+AND+driver+language%3AMarkdown&type=code&l=Markdown
The text was updated successfully, but these errors were encountered: