Hold the load
A short spike is not a steady run.
Manual owner record
Run it.
Write it down.
Record the board, fan, load, room temperature, power, and result. The card calculates rise above ambient and flags trouble. Nothing is uploaded.
Manual owner entry. Not automated telemetry, shop proof, or a safety controller.
One exact setup
Build the receipt.
Complete the short watched-load steps first. Continue only after the temperature settles. Use the final five-minute mean and stop for any flagged problem.
Do not enter a name, email, order number, serial number, or private message.
Comparable minimum20 min 5 min 80% Stop
Twenty. Five. Eighty.
Use the final mean
Compare temperature rise above ambient.
Keep the GPU busy
Lower steady load needs a rerun.
Do not chase a number
Fan trouble, thermal slowdown, or a rising temperature ends the run.
Have the CSV?
Check the load logSkip the averaging.
Read one to eight GPUs locally from the fixed NVIDIA load log. Have two matched runs? Compare them locally.
Need physical order?
Open the row checkMap the whole row.
Compare two to four positions after the same watched load. Plan first with the airflow budget.
What this proves
Read the proof gatesAn owner report. Nothing more.
A clean receipt can help support and decide what the shop tests next. Only matched shop-controlled telemetry can create a public result. See the fixed record rules.