[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#1056170: libhsa-runtime64-1: ROCr must assume xnack is disabled



Hi Christian,

I think the path forward to closing this bug is to patch rocr-runtime and remove this message. My conclusion from this investigation is that the message is merely informational. It's normal for HSA_AMD_SVM to be disabled and for HMM features to therefore be unavailable.

We should also document how to enable xnack+ on Debian. The xnack+ mode is not something any of us have much experience with, so an informal document on the ROCm Team wiki might be a good starting place for collecting information.

On 2023-11-24 09:05, Christian Kastner wrote:
That doesn't mean we can't do our own experiments, in fact I like the
idea of "forking" unstable with a customized kernel more and more, call
it "unstable-amdsvm" or whatever.

The advantage of having our own APT repo is that it's pretty easy to do
things like that.

Good point. We'll probably also want to enable HSA_AMD_P2P on our infrastructure once we have more software that can take advantage of multi-gpu nodes.

Sincerely,
Cory Bloor


Reply to: