2

I need to put together a proposal for buying a better computer for machine learning. Is there any good way to estimate the general training speeds of computer hardware? Basicly, I want to be able to say that if we purchase a certain computer it in trains x times faster than our current computers. Currently I have a i5 63000U 2.4ghz 2 core cpu, with no gpu. I know switching to a computer with a gpu will speed things up greatly. I just need a way to quantify the improvement to justify the investment. Any ideas will be greatly appreciated. Thanks in advance.

Matthew L
  • 21
  • 1
  • What is the problem that you are facing with current config ? E.g. which model (and how much data) is taking too long to train? – Shamit Verma Oct 03 '19 at 17:32
  • We mostly do deep learning cnn's and its taking weeks to finish. I was hoping there was some method to find out roughly what kind of performance increases I could expect. – Matthew L Oct 03 '19 at 17:50

1 Answers1

1

Is there any good way to estimate the general training speeds of computer hardware?

Probably not in general, because "training speed" means something completely different if you train a kNN model (virtually no training) or a complex DL model, what kind of data and what kind of task.

However estimations exist for fixed settings, for instance I found this MLPerf Benhchmark or this paper.

Btw desktop computers are not ideal for very computer-intensive ML, cluster or cloud computing offer better options.

Erwan
  • 24,823
  • 3
  • 13
  • 34