# Proof of GPU usage — H100 on vast.ai

- Node Hour index at rent time: **$2.69/h** (H100 SXM, quoted 2026-10-02T11:37:55.604Z)
- Rented: offer 40179478, H100 SXM, **$1.7355555555555555/h**, Washington, US
- Requested → running: 174 s; → SSH usable: 175 s
- GPU identity: `GPU-c0464b35-0f50-72f7-dbf9-13b22b54d2b9, NVIDIA H100 80GB HBM3, 550.163.01, 81559 MiB, 00000000:E3:00.0, 1650324009867`
- Workload: bf16 matmul 8192², 3600 s on the GPU (100% utilization, ~700 W, observed via nvidia-smi at 12:07–12:08 UTC). **The final TFLOPS line was lost**: the controller's SSH session dropped before it could read the result. TFLOPS were measured in a separate 15-minute run on the same GPU UUID: see run 20261002T131716Z.
- Billed (estimate): $1.7496 for 3622 s
- Attestation: not available on this host (nvidia-smi conf-compute not supported)
- Log sha256: `4239617fd296dfde1ee68922c64859a74da1a619bfcfe7d5f7b87f4144879083`

Errors: SSHException: SSH session not active (controller lost its session while reading the result; the machine was destroyed by the script's finally block at 12:38:41 UTC)