Skip to content

revert: Remove g7e MIG profiles (hold back MIG support for RTX PRO 6000)#403

Merged
mollyheamazon merged 1 commit intoaws:mainfrom
PremiumSpider:revert-g7e-mig-profiles
Apr 7, 2026
Merged

revert: Remove g7e MIG profiles (hold back MIG support for RTX PRO 6000)#403
mollyheamazon merged 1 commit intoaws:mainfrom
PremiumSpider:revert-g7e-mig-profiles

Conversation

@PremiumSpider
Copy link
Copy Markdown
Contributor

@PremiumSpider PremiumSpider commented Apr 7, 2026

What's changing and why?

Why now: The g7e MIG profiles (PRs #390, #391) merged to main on Mar 19 but haven't shipped yet (latest release is v3.7.0 from Mar 2). If we release without reverting, customers would see
MIG profiles for g7e via list-accelerator-partition-type and attempt to use them before MIG on RTX PRO 6000 is fully validated end-to-end. This revert prevents exposing unvalidated MIG
functionality while keeping g7e fully usable in whole-GPU mode.


Removing MIG-specific configuration for g7e instances to decouple MIG support
from the initial g7e release. Instance type recognition and hardware specs
remain intact so g7e clusters work in whole-GPU mode.

Changes:

  • Remove 6 g7e entries from INSTANCE_TYPE_MIG_PROFILES in constants.py
  • Remove g7e MIG config block (4 profiles) from GPU operator
    default-mig-config.yaml

Partially reverts 902e88f (PR #390), fully reverts MIG portion of 51b342f (PR #
391). MIG partitioning for g7e will be re-enabled in a future PR.

Before/After UX

Before: g7e instances support MIG partitioning with profiles 1g.24gb, 2g.48gb,
4g.96gb.

After: g7e instances work in whole-GPU mode only. MIG configuration requests for
g7e are rejected by the CLI. All other instance types' MIG configs are
unaffected.

How was this change tested?

Config-only removal. Verified:

  • YAML remains valid after removal (yaml.safe_load passes)
  • Both code paths using INSTANCE_TYPE_MIG_PROFILES check key membership before
    access — no KeyError risk
  • g7e entries in HyperpodInstanceType enum and INSTANCE_RESOURCES are preserved
  • B300 profiles added after the g7e block are unaffected

Are unit tests added?

N/A — config-only change (removal of entries).

Are integration tests added?

N/A — config-only change.

Reviewer Guidelines

‼️ Merge Requirements: PRs with failing integration tests cannot be merged without
justification.

  • Changes are documentation/config-only delaying g7e MIG without blocking
    release support for g7e

Remove MIG-specific configuration for g7e instances while keeping
instance type recognition and hardware specs intact:

- Remove g7e entries from INSTANCE_TYPE_MIG_PROFILES in constants.py
- Remove g7e MIG config block from GPU operator default-mig-config.yaml

g7e instances remain valid for HyperPod CLI operations in whole-GPU mode.
MIG partitioning will be re-enabled in a future PR.

Partially reverts: 902e88f (PR aws#390), fully reverts MIG portion of 51b342f (PR aws#391)
@PremiumSpider PremiumSpider requested a review from a team as a code owner April 7, 2026 21:59
Copy link
Copy Markdown

@rvikart rvikart left a comment

Choose a reason for hiding this comment

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

Approved

@mollyheamazon mollyheamazon merged commit 3d889ab into aws:main Apr 7, 2026
1 of 2 checks passed
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