I'm currently training CNNs using Tensorflow (Python) on my GTX 970 (specs here). I recently took a look at the new pascal based Titan Xs and I'm wondering what an estimated performance/speed gain would be if I upgraded?

The memory increase is an obvious benefit for larger models, but I'm mostly wondering about speed.

Does the doubling of core count and bump in memory speed offer that much of a performance gain? Anyone have any experience with the new (or old) Titan X cards?


1 Answer 1


There are a lot of parameters which matter when using GPU's for machine learning, some of them are:

  1. CUDA core count
  2. Memory bandwidth (GB/s)
  3. Memory per core (MB)
  4. Raw Speed (MHz)
  5. Total Memory available (GB)
  6. Performance on 16-bit, 32-bit floating ops/sec

Tim Dettmers has an excellent (frequently updated) blog where he's compared different cards, near the end he's also given a simple speed-up comparison. Using that as a guide, it can be estimated that Titan X pascal would be upto 5 times faster than a GTX 970 for Deep Learning.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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