Start with the problem

We need to accept a new cluster

Plan independent HPC and AI acceptance around useful workloads, correctness, recovery and a reproducible handover.

Start with one question

What research must the delivered system support, with what correctness, turnaround and recovery criteria? Agree this before comparing any headline benchmark.

A useful route through the evidence

  1. Choose what each test establishes — A compute result is useful evidence for its scope; it cannot stand in for the research workload.
  2. Measure the communication baseline — Record placement and binding, then add the concurrency the real workload needs.
  3. Qualify GPU communication — Check intra-node and inter-node behavior without substituting a headline collective result for application work.
  4. Include shared-storage behavior — Test bulk and metadata paths with agreed data, layout and concurrency.
  5. Recover useful work — Verify a restart and its recovery objective, not just the existence of a checkpoint file.

Keep the first experiment small

Use a bounded test allocation or agreed maintenance window. Save the exact workload, node selection, software and configuration before changing one variable. A repeatable result from a small test is more useful than a large run whose conditions cannot be reconstructed.

Download the benchmark run record and acceptance checklist. The manifest is an illustrative starting template; fill in actual identities and criteria before using it as evidence.

See how the investigation could look

Read the related illustrative lab scenario. It explains the hypotheses, evidence and decision points without claiming a customer result.

Explore the relevant service, see sample deliverables or read how procurement works. You can use the guides and tools without starting an engagement.