Script or app to load GB10 GPU to test for maximum power?

[Larry-SB]

My Spark seems slow and runs very cool, with nvidia-smi indicating ~30 watts of power usage and 50c temperature on LLM running, with over 90% GPU utilization.

What’s a good quick test that would run the GPU at its maximum load capacity to see if some sort of throttling is at play?

[davedgd]

I’d try gpu-burn from the terminal: GitHub - wilicc/gpu-burn: Multi-GPU CUDA stress test · GitHub

It worked well for me for stress testing some cards I bought in the past.

[joshua.dale.warner]

This is the Spark-specific tool to check for the USB-C PD issue.

GPU PD Throttle Check Tool

The suspected PD throttling issue has been consistent enough that I wanted to create a quick tool to check for the failure state. Some noodling around with Claude resulted in: It checks to see if a GPU load runs over or under a default clock speed threshold (defaults to 1400MHz) over a number of samples (defaults to 20 samples spaced at about 0.5s). A GPU in a good state should hit ~2400MHz. Mine in a bad state are showing 500-850MHz, even though the power state is still showing P0. Happy…

Run it. If you have the issue, unplug your device from the wall AND from the brick, wait a bit, plug it all back in and restart.

[Larry-SB]

Thanks, I just ran it and monitored nvidia-smi. It hit 93 watts and 71C and held through the run until it terminated.

[Larry-SB]

Ran the second tool and it looks good too.

PU state at idle:

Clock: 208 / 3003 MHz

P-state: P8

Power: 3.1 W

Warming up GPU (2.0s)…

Collecting 20 samples under load (0.5s interval)…

Threshold: 1400 MHz

  \#  Clock (MHz)  Max (MHz)  PState  Power (W)

───── ─────────── ───────── ────── ─────────

  1         2411       3003      P0       64.0

  2         2411       3003      P0       64.2

  3         2411       3003      P0       64.4

  4         2411       3003      P0       64.6

  5         2405       3003      P0       64.5

  6         2405       3003      P0       64.7

  7         2405       3003      P0       64.7

  8         2405       3003      P0       64.7

  9         2405       3003      P0       64.8

 10         2405       3003      P0       65.0

 11         2405       3003      P0       65.0

 12         2405       3003      P0       65.0

 13         2405       3003      P0       65.1

 14         2405       3003      P0       65.1

 15         2405       3003      P0       65.1

 16         2405       3003      P0       65.2

 17         2405       3003      P0       65.2

 18         2405       3003      P0       65.5

 19         2405       3003      P0       65.3

 20         2405       3003      P0       65.7

────────────────────────────────────────────────────────────

RESULTS

────────────────────────────────────────────────────────────

Samples: 20

Peak clock: 2411 MHz

Average clock: 2406 MHz

Avg power draw: 64.9 W

Below threshold: 0% of samples < 1400 MHz

┌────────────────────────────────────────────────────────┐

│ PASS — GPU clocks look healthy under load. │

│ Peak: 2411 MHz, Avg: 2406 MHz │

└────────────────────────────────────────────────────────┘