← Journal · · Talk sources
Driver, CUDA, and framework compatibility: links for the PyCon talk
Wow, look how many of us there are now 🔥🔥, thanks to everyone who joined the channel, and on that note, here's a batch of useful material))
This Friday, July 26, I'm giving a talk at PyCon about the GPU Operator and how it helps us configure nodes in K8s for working with ML frameworks. I bet a lot of you, when installing CUDA, could never quite figure out the various dependencies: which driver to install? which CUDA version fits? which torch or TensorFlow version is needed for a given CUDA version, and vice versa. And it turns out all of this also depends on the kernel, the OS, and the GPU architecture 🤯🤯🤯
At the talk itself I'll go into all this in more detail, plus of course there'll be stuff about GPU sharing. For now I want to share links to all the sources where you can check these dependencies :) So save this and enjoy 🤝
1️⃣ GPU architectures by driver version
2️⃣ CUDA by kernel (you can change the CUDA version in the link)
5️⃣ TensorFlow by CUDA version
7️⃣ driver by CUDA version on consumer GPUs
8️⃣ driver by CUDA version on datacenter GPUs
9️⃣ NVIDIA framework containers
I couldn't find a source for driver version dependency on the Linux kernel version. If you know one, drop it in the comments 🙏