Building an ML PC quote

Hello,

I am very new to ML and have set up the DeepSpeech project successfully on different machines (virtual/cloud).

My company have decided to invest in a physical machine. Now we’ve seen various recommendations on the internet but we intend to use DeepSpeech with the CommonVoice data set. We tried it on a p3.2xlarge AWS machine, but it was taking too long (and expensive!).

Any machine specs you guys recommend for ML with large data sets?

Pick two:

  • how large
  • budget
  • time for training

FTR, my development system for working on ~250h of french is a ThreadRipper 1950X based, with two RTX2080Ti. This allows to perform one epoch of training in ~17min.