Help w/ Machine Learning / AI Build?

HanClinto

I'm a computer programmer who is working in machine learning, and I'd like to have a personal workstation that can do local training of machine-learning models.

I have acquired an NVIDIA GeForce RTX 3090 Ti video card, but the machine I was hoping to put it into is far too small. Even if I could physically fit it into the case (which I can't), the power supply is far too weak to support it.

So I'm trying to see what it would take to build a workhorse PC to run this beast of a card and let me crunch some numbers. Could I get some feedback on this build please?

www.microcenter.com/site/content/custom-pc-builder-intel.aspx?load=93fd49fd-08ec-4360-ac41-91754fae70ba

I'm not entirely clear on what sorts of things I want to keep in mind when training models -- for instance, is it better to have 64gb of DDR5-5200, or 32gb of DDR5-5600? That's the sort of thing that I'm very inexperienced about. Does anyone have suggestions for me? Thank you!

Ian

Greetings. I'm not too familiar with machine learning myself, do you happen to know what software you would be using for this? Typically the manufacturer's site for software would have recommended specs available.

HanClinto

I'll be using a variety of ML frameworks and development tools -- none of them really have recommended hardware specs, since it's designed to run on everything from small devices to large supercomputers. For instance, Tensorflow has a list of CUDA-enabled GPUs that it supports, but there aren't specific hardware requirements beyond that:

www.tensorflow.org/install/pip#hardware_requirements

gqrbqge

Honestly Just looking at the spec sheet you should be just fine with what you have. I personally have used Revit and tons of other Autodesk software that won't work on the average office pc. You maybe might want 64gb of ram but that's just my personal opinion.