Closed hliuca closed 2 months ago
While this fix technically works, it's just silencing the default behavior. I updated the API we called for --pcie to the modern gpu_metrics implementation. This is resolved in the develop branch: 25ef420407a93ea29a6f8444a5e5d95033b55bc9 The fix will be in the ROCm 6.1.2 release. Thank you for pointing this out!
It takes around 8.6s to output all values, which is too long for system monitoring. It also needs a big screen to fit all values.
The PR disables a few values, which shortens the output and speeds the query to be less than 0.5 s.