I'm looking for an upgrade from my i5 6600. Workload - ML prototyping in Keras and heavy IO, compression etc.

I'm considering Ryzen 9 3900X + MSI X570-A PRO. What architectures in this price area (around 1000$) would you recommend, having in mind mostly multicore perforformance per $?


Two assumptions first:

  • by architecture you actually mean platform. ARM is not up to standard with performance yet and Power9 seems over budget for you
  • your 1000$ budget is for CPU and motherboard (if it is just for the CPU please comment and I will adjust my answer)

In the current market and in your budget you will likely not be able to get a better CPU.

The Ryzen 9 3950X is supposed to come out soon but it is supposed to be 50% more expensive by MSRPs from the launch event. Counting by retail prices in my country (Poland) this will put the X570-A PRO + 3950X at around 1150 USD. Of course the price might be different in different countries.

Another option is to invest that money in the overall build quality, primarily the motherboard and cooling, and try to eke out more performance but the returns here are minimal.

Personally, if you are unable to stretch your budget or wait for the 3950X, I would spend maybe 50$ on a bit better motherboard and keep the rest for the next GPU upgrade.

Edit: I don't know much about the data transfer patterns in ML vs games, but ML uses much more memory so it might benefit greatly from PCIe 4.0 once next-gen GPUs implement it

Edit 2: new release from AMD states that we are two months off from the 3950X, not a week or two as originally planned

  • Any particular motherboard you had in mind for 50$ + ? I'm asking bcasue I thought the MSI is an overkill already for my needs Sep 22 '19 at 19:11
  • Yes, both assumptions are correct. Sep 22 '19 at 19:12
  • @PiotrFalkowski it's the cheapest X570 board in Poland right now. I would have to look up some reviews. Also: are there any single-core bottlenecks in your framework? Particularly in communicating with the GPU. Sep 22 '19 at 20:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.