[BUG] Temperature and core frequency graph issue

[ezekriSCW]

Describe the bug
During full gpu load job, on 'core_freq vs Thermal' graph, at the end of the job, CPU temperature suddenly falls down et core frequencies have strange behaviour.
Same behaviour noticed with default 600s-runtime duration and with 420s-runtime.

To Reproduce
Steps to reproduce the behavior:

1. uv run hwbench -j configs/full_cpu_load.conf -m monitoring.cfg
   2.uv run hwgraph graph --traces hwbench-out-20241126111954/results.json:DL340:BMC.Server --outdir DL340

Additional context
job graphs respectively with 600 and 400s

[ezekriSCW]

additional cpu pkg vs thermal graph

[anisse]

Something is very weird; notably the fact that the core frequencies are moving towards the end. Some hypotheses:

• your turbostat is not up-to-date (what version do you use?). I'm not sure it's this because I think it should just "not work" in this case
• stress-ng crashed at some point and the benchmark stopped?
• some other issue, maybe hardware-related?

[anisse]

How is this benchmark result in term of performance scalability ?