JD Russo @quasifrodo@qoto.org

Today’s project: an #ESP32 powered VOC sensor above my cat’s litter box, to try and detect when she’s pooped.

I love her, but she can be a smelly girl. If I can detect when she’s gone, I can automatically dispense some air freshener!

Ya know, I don’t think I’ve EVER regretted doing my due diligence before sending a salty email.

Turns out my secondhand SAS drives aren’t busted — my RAID controller was just overheating with no top cover on the server chassis. Problem, uh, “solved”!

Today’s project: I picked up a proper 12U #server cabinet, so I’ve been moving my #supermicro server and #3DPrinter from my IKEA Lack tables over to it.

Don’t have proper rails for the server because broke #gradstudent, but I’m test printing some brackets to mount it on. Not sure they’ll hold up its weight, TBD..

Also couldn’t resist setting up some #DIY LED lighting inside!

Aaaaand it's working! #proxmox is now seeing my #nvidia K10 #gpu . That was actually surprisingly easy.

Now to set up passthrough to a container, and do some benchmarks

Score!

Now… to figure out how the hell I mount and power this

#homelab #selfhosted #science #computing #gpu #nvidia

Looks like Christmas came early this year! #homelab #selfhosted #server

Today's accomplishment: I have an analysis pipeline with a few parallelizable steps. I typically run my analysis in parallel on my desktop, but I've both been needing more oomph and have been running out of drive space nonstop.

My #homeserver (40 cores/80 threads with quad E4870s) is running ZFS on a RAID0 array, currently just with two 10k SAS drives. By mounting this as an NFS share on my desktop, I can put all my generated simulation data directly on that larger filesystem.

Additionally, my server is running #proxmox, and has a "science" container. I can also mount that NFS share in the science container, giving it direct access to my simulation data, and run it as a #Ray cluster head node. Now, I can do simulations on my desktop, run the single-core tasks on that machine, and farm out the parallel analysis to workers on the server, putting all 80 cores to work!

#hpc #science #clusters #scaling #docker