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

Enable NFD rule for GPU resource driver Helm chart #68

Open
wants to merge 5 commits into
base: main
Choose a base branch
from

Conversation

oxxenix
Copy link
Contributor

@oxxenix oxxenix commented Dec 9, 2024

No description provided.

@oxxenix oxxenix changed the title Enable NFD rule for GPU resource driver Helm chart [WIP] Enable NFD rule for GPU resource driver Helm chart Dec 9, 2024
@oxxenix oxxenix self-assigned this Dec 9, 2024
@oxxenix oxxenix removed the request for review from softwarerecipes December 9, 2024 10:41
@oxxenix oxxenix changed the title [WIP] Enable NFD rule for GPU resource driver Helm chart Enable NFD rule for GPU resource driver Helm chart Dec 10, 2024
@oxxenix oxxenix force-pushed the gpu-nfd-rule branch 2 times, most recently from cd15249 to e5bd563 Compare December 10, 2024 11:59
@oxxenix
Copy link
Contributor Author

oxxenix commented Dec 10, 2024

@byako, @eero-t, @tkatila PTAL

@oxxenix oxxenix requested a review from mythi December 10, 2024 12:04
Copy link

@eero-t eero-t left a comment

Choose a reason for hiding this comment

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

I think the initial ruleset could be simplified.

Every API exposed to users needs maintenance policy e.g. how long they're deprecated before being renamed or dropped, so exposing unneeded APIs (like labels) should be avoided.

charts/intel-gpu-resource-driver/templates/nfd.yaml Outdated Show resolved Hide resolved
charts/intel-gpu-resource-driver/templates/nfd.yaml Outdated Show resolved Hide resolved
@oxxenix oxxenix requested review from eero-t and tkatila December 16, 2024 11:37
- labels:
gpu.intel.com/product: "Max_1100"
labelsTemplate: "gpu.intel.com/device.count={{"{{"}} len .pci.device {{"}}"}}"
- labelsTemplate: "gpu.intel.com/device.count={{"{{"}} len .pci.device {{"}}"}}"
Copy link

@eero-t eero-t Dec 16, 2024

Choose a reason for hiding this comment

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

I did not notice this earlier, but there are 4 rules for different sets of GPU IDs, each producing the same gpu.intel.com/device.count=<count> label. Which won't produce correct results on node which has GPUs from multiple rule sets.

While this is not a problem for GPU plugin because it does not support nodes with different types of Intel GPUs, DRA GPU driver is supposed to support support such use-case too.

I think it's best just to remove these rules completely for now...

Copy link
Contributor Author

Choose a reason for hiding this comment

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

thanks! I removed these rules

Copy link

Choose a reason for hiding this comment

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

As there are still gpu.intel.com/family label rules for A_Series & Max_Series, for constency there should be also Flex_Series one, or Flex GPUs should be included to A_Series, as they're also "Alchemist" variants.

Btw. nowadays there would need to be also B_Series, but that sounds really bad for the latest & greatest Intel client platform (so such labeling would be very unlikely to be OKed by marketing). I.e. I think the family name of that should rather be e.g. Battlemage, and A_Series (which nobody's going to recognize) should rather be Alchemist...

PS. PCI IDs for these are documented here: https://dgpu-docs.intel.com/devices/hardware-table.html#gpus-with-supported-drivers

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I refactored nfd rules a bit. See the latest commit

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants