jeremylongshore CoreWeave Upgrade & Migration
GPU Type Migration
# Before: A100
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: gpu.nvidia.com/class
operator: In
values: ["A100_PCIE_80GB"]
# After: H100 (update affinity label)
values: ["H100_SXM5"]
Migration Steps
- Build new container with updated CUDA version if needed
- Deploy with new GPU affinity alongside old deployment
- Shift traffic gradually (canary)
- Validate latency and throughput match or improve
- Scale down old deployment
CUDA Version Matrix
| GPU Type | Recommended CUDA | Driver |
|---|---|---|
| A100 | CUDA 12.2+ | 535+ |
| H100 | CUDA 12.4+ | 550+ |
| L40 | CUDA 12.2+ | 535+ |
Rollback
kubectl rollout undo deployment/my-inference
Resources
Next Steps
For CI/CD, see coreweave-ci-integration.