After racking, why validate power and cooling before heavy tests?
Why B is the answer
To confirm that the facility can sustain full GPU load
Full load draws peak power and heat.
In detail
- Full load draws peak power and heat.
- Heavy tests come after.
Use case
An illustrative scenario, not a real incident.
A team validates power and cooling first, because full GPU load draws peak power.
See it in different ways
flowchart LR A["Rack, power, cooling"] --> B["BMC and OOB"] B --> C["Firmware"] C --> D["OS and drivers"] D --> E["Single-node tests"] E --> F["Multi-node tests"]
sequenceDiagram participant F as Facility participant B as BMC participant W as Firmware participant O as OS and drivers F->>B: Power and network B->>W: Update through out-of-band access W->>O: Install the OS and drivers O->>O: Confirm GPUs, then run diagnostics
flowchart LR S0["Rack installed"] --> S1["Validate power"] --> S2["Validate cooling"] --> S3["Run tests"]
Key point: Validate the facility before heavy tests.
Why the other options are wrong
- A. To pick the correct NCCL version for each of the racks
- NCCL version is unrelated.
- C. To reduce the number of GPUs in each of the servers
- It is not about GPU count.
- D. To set the correct Slurm partition for each user
- That is a scheduler setting.