Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions skills/together-gpu-clusters/references/cluster-management.md
Original file line number Diff line number Diff line change
Expand Up @@ -334,6 +334,7 @@ During provisioning, nodes undergo automatic tests:
- **GPU Burn (5 min)** -- stress test for thermal/power issues
- **Single-Node NCCL** -- GPU-to-GPU communication within a node
- **Multi-Node NCCL** -- cross-node GPU communication
- **Storage Performance** -- sequential read/write throughput and data-integrity validation on attached storage volumes

Nodes showing "Tests Failed" are not added to the cluster until repaired.

Expand All @@ -350,6 +351,9 @@ Nodes showing "Tests Failed" are not added to the cluster until repaired.
**PCIe Performance:**
- NVBandwidth: CPU-to-GPU, GPU-to-CPU bandwidth, GPU-CPU latency

**Storage:**
- Storage Performance: `fio` sequential read/write throughput plus a checksummed write/read-back for data-integrity validation against shared and local volumes attached to the cluster. Results include a **Skipped** state (in addition to Passed/Failed) -- on Kubernetes clusters the shared-storage portion reports **Skipped** with message `Shared volume PVC not created` until a shared-volume PVC exists (see the [Storage](#storage) section).

### Node Repair

- **Quick Reprovision**: VM recreated on a random physical node (for software issues)
Expand Down
Loading