The k1 node at the top has my RTX 4000 ADA SFF gpu in its factory form. The case is off on that node since, as you can see, it would not fit otherwise. I've been waiting to get gpu-operator working in order to start collecting "before" metrics before I take the gpu apart and replace the cooling with something smaller that will allow it it fit properly. Making this thing calculate pi, I see its temps settle at ~70 degrees C.
I picked up a replacement cooler from n3rdware (nice and heavy for its size) and some PTM7950 for the thermal pads. The PTM7950 is currently in the fridge. If the kids don't eat it, I'll use it later in the installation.
Mmmmm. Forbidden PTM7950 soft serve.
I've replaced the stock graphics card cooler with the #n3rdware alternative. The idle temp with the stock cooler and with the system case open was about 55 C. It looks like the new setup with the case closed has an idle temp of 63 C. I can probably improve on that with a shroud, but let's see how it runs under load first.
Also, I probably should have run the new cooler with the case open to have less variables changing for benchmarks. I may still do that before messing with printing shrouds.
Hmmmmm. This is from my pi calculation benchmark. It seems the new setup runs consistently about 15 degrees C higher than the old setup. That may be expected and fine. I'm trying to find specs for thermal throttle temp for this card. I see a suggestion of 83 degrees, but nothing official.