Which metrics belong in routine GPU monitoring?
Why B is the answer
Utilization, memory, temperature, power and errors
These show both load and health.
In detail
- Utilization, memory, temperature, power and errors show load and health.
- Set alerts on errors.
Use case
An illustrative scenario, not a real incident.
An operator tracks utilization and errors together, because load alone hides failing hardware.
See it in different ways
flowchart TD A["DCGM"] --> B["Utilization"] A --> C["Memory"] A --> D["Temperature"] A --> E["Power"] A --> F["ECC and Xid errors"]
mindmap
root((Operations))
Monitoring
nvidia-smi
DCGM
Scheduling
Slurm
Kubernetes
Provisioning
Base Command Manager
Sharing
MIG
vGPUmindmap
root(("GPU monitoring"))
n1["Load"]
n2["Utilization"]
n3["Memory"]
n4["Health"]
n5["Temperature"]
n6["Power"]
n7["Errors"]Key point: Monitor load and health.
Why the other options are wrong
- A. Only the date of the last operating system update
- That is not a metric.
- C. Only the GPU driver version
- A driver version is not health.
- D. Only the number of network cables in the rack
- Cable count is not a metric.