The case for my build finally arrived so I could finally take it off the box of shame.
This has been cobbled together with as many parts as I could that I already had which is why some of the choices are a bit random.
2x32GB DDR4 3200
3900X
ASUS Pro WS X570-ACE
1TB 980 Pro
Some random 2TB sata ssd’s I had lying around
2x r9700 ai pros
ASRock PG 1600G
Fractal Design Meshify 2 XL
Running proxmox so I can have some sandboxes for agents to play in and an inference vm with the gpu’s passed through. I have p2p working as well which is nice.
Don’t mind the mess in the background, me and horizontal surfaces do not get along.
Edit: forgot to include the box photo
2 Likes
So much room for activities!
1 Like
Maybe some day, I do have an internal yearning for more 
. This is already expensive for what is already effectively a toy, chance of ROI for this other than learning(if you wanna count that) is basically 0.
I said the same thing 5 GPUs ago. I’m a slow learner.
Atleast I got these two when they were 1300, now they seem to be going for 1700+. But I unfortunately will likely follow in your footsteps. I do kinda want a dgx spark or something blackwell based to be able to play on both sides of the isle in my experiments.
You should consider upgrading that CPU to a 5900XT:
Its basically $300 for an easy 25%-50% performance increase. And less power draw.
1 Like
I may consider it but honestly, I would not expect more performance for Inference which is pretty much all this box does minus acting as a tailnode, nginx reverse proxy, and herdr/pi setup. The other node in my cluster has a 5900X in it where I run more of my heavier workloads, CI/CD, game servers and the like.
Edit: I have a particular fascination with using hardware till it doesn’t breathe anymore and being scrappy, 300$ is more money for gpu
