Which Xeon to choose for a workstation for machine learning?

Question

Iron

Good afternoon. I'm thinking about assembling a workstation for tasks from the field of machine learning. Based on the price-quality ratio, it seems reasonable to use a dual-processor assembly on the Xeon. Something like E5-2683 v4 on socket 2011-3. Has anyone implemented something like this? If so, what processors and boards would you recommend? Only processors and motherboards are of interest, there is no need to write about the video card.

Answer 1

And how will Xeon help you in this regard? In my opinion, it’s better to save money and eventually save money for some GPGPU - a computer, like NVIDIA TESLA or whatever you can get cheaper there now (possibly used) And choose by the amount of memory, the more the better. I'm not a very big specialist in this and maybe they will correct me, but in my opinion everything slows down very much when the model for training ceases to fit in the memory of the accelerator. The processor is only indirectly involved in this matter. On the other hand, I don’t know what models you will have, maybe they won’t be driven to the GPU at all ...

Answer 2

the old xeon is not suitable for this task at all ...
expensive server processors for neural networks, with a large cache size, a large number of cores (by the way, not necessarily) and multi-channel memory,
if we are talking about small amounts, it is better not to find desktop ones - amd rysen cezanne 5700g (
5600x /5600g) or intel rocket like 1170k (or 1160kf) see different models with or without integrated video ram) is about 20-100 times slower than on a good gpu, look through the topic about the Sberbank Neroset