Quarkpad

Local CSV check

Log the load.
See the row.

Read one to eight NVIDIA GPUs after the same watched workload. See final temperature, power, utilization, the hottest device, and the spread. The file stays here.

Check a load logRead the watched-load steps

Owner telemetry. Not shop proof, control software, or a safety limit.

One fixed command

Capture. Stop. Check.

Run for at least 20 minutes during the exact workload. Press Ctrl+C. Then choose the CSV below.

No UUID, serial, name, email, or order data. PCI bus ID is not a physical position.

nvidia-smi --query-gpu=timestamp,pci.bus_id,name,temperature.gpu,power.draw,utilization.gpu --format=csv,noheader,nounits --loop=5 --filename=quarkpad-gpu-load.csvReads GPU telemetry. Writes one local CSV.
Read in this browser. Nothing is submitted.
What counts

Twenty. Five. Eighty. Watched.

20 min

Use one load

Short logs stay a recheck.

5 min

Average the end

The tool uses the shared final window.

80%

Load every GPU

One idle device blocks comparison.

Watch

Stop for trouble

Slowdown, fan, cable, airflow, or rising heat ends the run.

Have a before and after?

Compare the matched runs.

Pair the same GPUs, workload, direction, power, and case. A mismatch produces no difference.

Compare two logs
Fail closed

Wrong format? No result.

NVIDIA says nvidia-smi output can change between driver releases. This checker accepts only the fixed six-column file. Read the NVIDIA reference. See the record rules.

Map the physical row