I'm putting together a server for AI experiments. I'm newish to that area. The server is based on a Gigabyte T181-G20 chassis.
Dual cascade lake cpus, 6230 20/40 core. 4 V100 gpus.
The budget is still a little limited. it has 6 channel ram, so i intend to fill all 24 slots. It'll be either 8GB or 16GB dimms. Should i bias towards more total ram, but slower, 2133, or less total ram, but faster, 2933 or 2666?
Most of the AI server guides i can find say "at least 128GB ram" this board would be either 192Gb or 384Gb.
What makes sense?
Dual cascade lake cpus, 6230 20/40 core. 4 V100 gpus.
The budget is still a little limited. it has 6 channel ram, so i intend to fill all 24 slots. It'll be either 8GB or 16GB dimms. Should i bias towards more total ram, but slower, 2133, or less total ram, but faster, 2933 or 2666?
Most of the AI server guides i can find say "at least 128GB ram" this board would be either 192Gb or 384Gb.
What makes sense?
I'm still learning myself so don't really know but my impression is that if you don't plan to offload to CPU and just stick to GPUs then I think it doesn't really matter and don't need much RAM. So I think get less faster RAM and as fast as your budget allows.