plasma-umass / scalene

Scalene: a high-performance, high-precision CPU, GPU, and memory profiler for Python with AI-powered optimization proposals
Apache License 2.0
12.23k stars 399 forks source link

Neuron support #836

Closed emeryberger closed 4 months ago

emeryberger commented 4 months ago

Note that this PR changes - in fact, fixes - the way GPU utilization is reported. It is no longer normalized to the number of cores in use, but rather to the total number of GPU cores, which paints a better picture of how well code is using GPU (or here, Neuron) cores.