Objective

Configure PyTorch compilation for RHEL 9.6 with H200 GPU (Hopper architecture).

Work Completed

• Updated build workflow to support RHEL builds
• Configured CUDA architecture: 9.0 (H200 Hopper)
• Set up distributed build with sccache
• Added build artifact handling
• Configured test matrix for 5 test shards
• Implemented self-hosted runner integration

Performance Metrics

• Build time: ~2 hours 16 minutes
• Artifact size: ~3.5GB
• Uses sccache for incremental builds

Deliverables

• [x] PyTorch builds successfully on RHEL 9.6
• [x] CUDA 12.8 integration working
• [x] H200 GPU support enabled
• [x] Build artifacts generated correctly

References

• Workflow: .github/workflows/rhel-build-test.yml
• Build script: .ci/pytorch/build.sh