Is the version specified in NVIDIA CUDA install guide the minimum version, or an exact requirement?

In order to use “RHEL 9.y (y <= 3)”, the table recommends kernel version “5.14.0-362”

This is not the latest version available of that kernel, and as a beginner Linux user, I’ve had trouble getting a downgraded kernel working.

Am I able to download and run the CUDA toolkit and NVIDIA drivers on a kernel version newer than what the table specifies, or does it have to be exact?

I’m currently using CentOS with kernel version 5.14.0-430.

It indicates what the CUDA developers tested things with. It is the recommended kernel version for maximum compatibility/minimum issues.

It’s often possible to use other kernel versions. I have not investigated, but I would not expect your experience with 5.14.0-430 to be any different w.r.t. CUDA as if you had version 5.14.0-362

If you are having good success running things on the -430 kernel, just stay with that. You will be fine as far as CUDA compatibility is concerned.

The main issue you are likely to run into if using a widely different kernel version is in the area of GPU driver install. if you have a successful driver install, your CUDA install will probably work fine with attention to the other aspects of that compatibility table (e.g. gcc version, glibc, etc.)