https://www.phoronix.com/news/Red-Hat-Distribute-CUDA-RHEL
Following Canonical announcing plans to better support NVIDIA CUDA on Ubuntu Linux and make it easier to install as well as SUSE better supporting CUDA along similar lines, Red Hat today affirmed their plans to do the same. Red Hat will be making it easier to use the NVIDIA CUDA stack across RHEL, Red Hat AI, and OpenShift products.
Red Hat will be distributing the NVIDIA CUDA Toolkit directly within their platforms to streamline the developer experience, provide operational consistency to customers, and make it easier to leverage Red Hat platforms with the latest NVIDIA hardware and software innovations.
https://distrowatch.com/dwres.php?resource=showheadline&story=20084
Red Hat has announced a partnership with NVIDIA to bring GPU computing tools to Red Hat platforms, making it easier for developer to access NVIDIA video card features. A blog post on the Red Hat website states:
"Engineers and data scientists shouldn't have to spend their time managing dependencies, hunting for compatible drivers, or figuring out how to get their workloads running reliably on different systems. Our new agreement with NVIDIA addresses this head-on. By distributing the NVIDIA CUDA Toolkit directly within our platforms, we're removing a major point of friction for developers and IT teams. You will be able to get the essential tools for GPU-accelerated computing from a single, trusted source."
The NVIDIA tools will be available on Red Hat Enterprise Linux, Red Hat OpenShift, and Red Hat AI.
See also:
• https://www.redhat.com/en/blog/red-hat-distribute-nvidia-cuda-across-red-hat-ai-rhel-and-openshift
• https://developer.nvidia.com/cuda-toolkit
(Score: 2) by aafcac on Friday October 31, @06:18PM
This is good and all, but the price of the hardware to run CUDA on nVidia hardware is still out of control. And while there are projects like SCALE and ZLUDA, I'm not sure if those are going to be applicable to this.
It's a real shame that OpenCL doesn't seem likely to catch on for these sorts of applications any time soon.
(Score: 0) by Anonymous Coward on Friday October 31, @07:11PM (1 child)
What could possibly go wrong hooking your wagon to a proprietary corporate entity whose primary goal is to satisfy the important people, you know, the stakeholders. The ones whose yachts need upkeep.
(Score: 0, Troll) by Mojibake Tengu on Friday October 31, @08:04PM
I am looking for a cheerful day when NVIDIA buys indebted USA. I am sure I'll live to it.
Anyway, not impressed by funny LLMs either but bf16 NPU hardware is extremely practical for accelerating balanced ternary logic.
Now it is possible to write emulators for very divergent hardware of the far future.
Rust programming language offends both my Intelligence and my Spirit.
(Score: 4, Funny) by The Vocal Minority on Saturday November 01, @03:49AM
On Ubuntu, at least, I'm pretty sure you can install CUDA libraries through the package manager.
The problem has been that in the AI world no (or at least very few) fucks are given about backward compatibility, leading to much hair pulling if you are trying to run older models. One of the impacts of this is the necessity of having multiple versions of CUDA installed and then switching between them as needed - hopefully this is what will be made easier as a result of whatever this announcement is actually about.
(Score: 2, Touché) by jb on Saturday November 01, @08:42AM
What's the point of running a free software OS if you're going to install some big blob of proprietary garbage on it and give that big blob direct access to (at least some of) your hardware?
Looks like IBM (RHEL) and Canonical (Ubuntu) are yet again showing that they just don't understand free software at all.