Shortly after my friend and I graduated from college and AI started really taking off (maybe a year or two after chatgpt), we decided to build an ML learning server primarily for learning how to leverage these shiny new local AI tools such as ollama and stable diffusion.

We were originally going to build a custom self hosted agent personality that we could interact with, but we ended up getting side tracked a lot of the time. This was no easy task.

We started with one 3090 and an old PC, then eventually grew to two 3090s and a server motherboard. (pictures coming soon)

Thankfully we built this before the RAM and GPU price hikes, it was actually pretty reasonable financially!

Maybe around $350 per GPU ECC 2666MHz RDIMM RAM for 64GB was $100 Open Frame Case $55 CPU Cooler (huge pain to install this one specifically, we had to put so much pressure on it that the motherboard seemed to bend a little, but its still working a year later) ARCTIC Freezer 4U-M Rev. 2 - $60

This test above was for thermal issues we were having. The second GPU would cook the GPU above it and cause the whole system to shut off.

Having it externally with this super long PCIE riser worked, but it all just felt so wrong. We used a small PC case to hold it temporarily like this.

(TODO upload and write more, theres a bunch missing from this)