Looks cool!<p>nvtop can actually support TPUs too via <a href="https:&#x2F;&#x2F;github.com&#x2F;rdyro&#x2F;libtpuinfo&#x2F;" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;rdyro&#x2F;libtpuinfo&#x2F;</a> <a href="https:&#x2F;&#x2F;github.com&#x2F;Syllo&#x2F;nvtop&#x2F;blob&#x2F;76890233d759199f50ad3bdb57a0c0988e96fc44&#x2F;CMakeLists.txt#L100" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;Syllo&#x2F;nvtop&#x2F;blob&#x2F;76890233d759199f50ad3bdb...</a>

Look into all-smi <a href="https:&#x2F;&#x2F;github.com&#x2F;lablup&#x2F;all-smi" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;lablup&#x2F;all-smi</a>
It supports all GPUs thinkable including Apple Silicon and many AI accelerator cards.

would be nice to have cpu usage added so I have all in one?<p>currently I use btop which shows basic gpu load along with cpu, network, etc.

Is it capable of exposing metrics in Prometheus format?

Renaming fopen64 to intercept library calls feels like a brittle hack masquerading as &quot;sandboxing.&quot; Why not just upstream this hardware support to nvtop instead of fragmenting the ecosystem?

&quot;NPU&quot; seems to refer to trainium only?

HN

Zml-smi: universal monitoring tool for GPUs, TPUs and NPUs

Comments (11)