Hi everyone,

Thank you so much for the comments!

Well, indeed it is a toy application made for my research group just to check which GPU is available. However, I was hoping to make it a little more useful with help and suggestions from the community. I believe I should add "Show HN:" to the title before posting, sorry.

As for why not to use professional tools, I guess we wanted something dead simple.

Also there's the whole "killing a mosquito with a bazooka" point there. If you are making this for your research-group then all the more reason to not depend on a big thing like Zabbix that'd require a fatter server and someone to set up/maintain.

For your purposes this looks like more than enough but if you wanted to hone your Pythonic/Flask/webapp skills a little more I'd suggest you expose the rest of the info exposed by the CLI and also expose the backend API for people that may wanna use your underlying code and integrate into their own stuff (eg something that chooses an available GPU before starting a job).

+1 really appreciated!

Bit of a toy really but the effort is appreciated :). As far as I can see you're only showing memory usage (which I'm not sure qualifies as 'utilisation') while `nvidia-smi` also returns the processor utilisation and temp. Perhaps it'd be worth including those.

nvidia-smi reports the core utilized if even one vector lane is working, it's not super reliable

Which tool are you using for monitoring Nvidia GPUs?

`nvidia-ml-py3` a Python interface to `nvidia-smi`.

I was looking for something more accurate than nvidia-smi, per parent comment.

Cool! We all set this across a spreadout VPN so that our apps can keep track of the GPU availability https://github.com/jolibrain/gpustat_server It does spit JSON out with memory and computer usage.

Is there a reason you wouldn’t use one of the many more comprehensive monitoring solutions e.g. Zabbix?

+1 I will fix it in the documentation.

Thanks!