4

I am a researcher working on my first deep learning project, which consists of using a CNN (pre-trained VGG16+2 densely connected layers) to classify drone imagery of vegetation.

In trying to hack computing times for both training and prediction of new images, I am considering asking my employer for money to buy a (cheap) NVIDIA GPU.

Being a biologist and not a computer scientist, I do not have any sense of the upgrade this would get me. Searching online I found contradicting opinions.

I am currently working with Keras+Tensorflow on a desktop PC with i7, 3.6 ghz, 32 Gb RAM.

Question: how good a GPU would I need to get a sensible performance increase?

Thanks a lot!

Ethan
  • 1,625
  • 8
  • 23
  • 39
Levasco
  • 41
  • 4

1 Answers1

2

According to this guy, he got a 15x increase from Intel i7 to GeForce 1070.

You also may consider using AWS. You can use a machine 100x as powerful (as a single 1070) and your employer may find it attractive because the upfront sunk cost is zero.

B Seven
  • 292
  • 1
  • 9
  • 2
    Thanks for the suggestions. For completeness, I tried with a colleague's old nvidia graphic card, a Quadro K620. Despite it being low-end and not optimized for deep learning, I got a 5x speed increase in training the same model (once I got around the installation of tensorflow-gpu). Thus, **cheap GPU beats powerful CPU**. – Levasco Feb 22 '19 at 18:01
  • For context, you may want to look into building a high end Gaming machine. You can spend as much on the video card(s) as you would on the rest of the components. – B Seven Feb 22 '19 at 18:06
  • @Levasco - You've probably seen this, but its another good reference: https://datascience.stackexchange.com/questions/14941/after-the-training-phase-is-it-better-to-run-neural-networks-on-a-gpu-or-cpu?rq=1 – B Seven Feb 22 '19 at 18:09