tools

DeepLM Baseline

Benchmark your GPU cluster's real performance in one run.

DeepLM Baseline benchmarks your GPU cluster's real performance in one run. Results are compared against vendor specifications with pass/fail thresholds, which makes it useful for acceptance testing new hardware and diagnosing degraded performance.

What it tests

  • Compute: TFLOPS throughput, HBM bandwidth and thermal throttling detection.
  • Interconnect: NVLink, NVSwitch, PCIe and NUMA health.
  • Network scaling: InfiniBand and RDMA, NCCL, and AllReduce performance.

The source is in the Baseline repository.