Computer for machine learning on eeg data

Support & Discussions
1

I hope this is mne-adjacent enough, since many of you are doing the same work. Apologies if not.

Specing a computer for machine learning on eeg data. Datasets are in MiBs for now. <$5K Watched endless YouTube videos, read sites, :thinking:
Advice? :pray::pray::pray:
I9-13900K or eg. Single/Dual Xeon ?

RTX 30xx (maybe dual) or RTX A4500?

Concerns:

Xeon on Geekbench benchmarks, e.g. single core is same or worse than my i7-6700

No usb-c for Xeon mobos

Lack of PCI slots on i9 mobos, (TB3/4 would be nice)

RTX A4500 seems to be sweet spot for GPUs(?), with 7168 cuda cores, 224 tensor cores, 20GiB GDDR6, 320 bit bus

2

I spec’d one a couple years ago and ended up not going with Xeon but I can’t recall exactly why. Since you’re working with EEG data (no source imaging right?) the GPU will only matter for GPU-accelerated computations (not for actual 3D rendering) so look at your pipeline and see which steps can actually be done on GPU; it may turn out that simply having more RAM and CPU cores will make a bigger difference to your run times.

1 Like