No GPU boost after driver downgrade (DGX Sparks stopped to warm to ~80 C on inference)

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
,
1

Hello,

I’ve recently discovered how Deepseek V4 Flash could be fast at 2x DGX Spark nodes, so I’ve reached about 40 generation tps and found a note about performance degradation in NVIDIA driver v580.159 so tried to downgrade to v580.142.

at v580.159 I’ve noted DGX Spark could be kinda warm and heated GPU to 79-80 degrees which looks fine

after downgrade to v580.142 I’ve seen 40-50 digress on GPU which looked strange for me and I’ve started to gather frequency and power metrics from nvidIa-smi and found on one DGX Spark I have frequency 507Mhz on second (which is a rank1) - 689Mhz.

I’ve suspected driver/firmware issue and upgraded back to v580.159 but the behaviour still persists, I even restored to factory defaults and set all the things up back but it still remains on 507Mhz and 689Mhz.

During this operations I’ve turned off Secure boot, but has “loaded optimal defaults” before recovery as documentation describes. So currently I’m a bit lost in ideas what this could be.

Any ideas where to dig next?

Thanks!

2

Found an answer in suggested topics: DGX Spark (Grace Blackwell GB10) Performance Drop: GPU Trapped in 15W / 650MHz Loop with 50°C Artificial T.Limit Temp - #9 by aniculescu - turning DGX Sparks from the power physically helped and performance restored. Strange that happened on driver change, however the issue is fixed.