Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Docs: Add xnack to unified memory page #3732

Open
wants to merge 1 commit into
base: docs/develop
Choose a base branch
from
Open

Conversation

adeljo-amd
Copy link

Add sub-section about XNACK (along with a small example) to the unified_memory.rst page.

Copy link

@MKKnorr MKKnorr left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Some small suggestions

@adeljo-amd adeljo-amd force-pushed the docs-xnack branch 2 times, most recently from 0498b4d to 5931eb6 Compare January 29, 2025 14:46
@adeljo-amd adeljo-amd added ci:docs-only Only run Read the Docs CI on this PR documentation labels Feb 3, 2025
Copy link

@randyh62 randyh62 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A few comments, but overall it looks good.

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fir line 13:
space and dedicated physical memory backing it up, for example normal RAM for CPUs and

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For line 49:
the host or another device, on which the memory is located. The page is then

XNACK
-----

On specific GPU architectures (which is referenced in the previous table), there is an
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
On specific GPU architectures (which is referenced in the previous table), there is an
On specific GPU architectures (referenced in the previous table), there is an

Depending on the system, page migration may be disabled by default in which case managed
memory will act like pinned host memory and suffer degraded performance.

**XNACK** describes the GPU's ability to retry memory accesses that failed due to a page fault
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Although it is footnoted in the previous table, I would add here that you need to enable XNACK ability by setting XNACK=1 environment variable.

Name: amdgcn-amd-amdhsa--gfx90a:sramecc+:xnack-

Here, ``xnack-`` means that XNACK is available but is disabled by default.
Turning on XNACK by setting the environment variable ``HSA_XNACK=1`` gives
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here it says that XNACK is enabled with HSA_XNACK=1, but up above it is simply XNACK=1. Which is correct?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
ci:docs-only Only run Read the Docs CI on this PR documentation
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants