eGPU + Keras (tensorflow) + NVIDIA

Question

ajay OP

Created Jun ’17

Replies 1

Boosts 0

Views 2.9k

Participants 2

Hi,

I am interested in getting an external GPU to connect to my macbook pro so that I can train my keras models faster.

I am using tensorflow as a backend to keras, so I know that it requires a CUDA enabled GPU (so I would have to use a NVIDIA gpu).

Cuda compatible GPUs listed here: https://developer.nvidia.com/cuda-gpus

Has anyone tried to use their Nvidia (cuda enabled gpu) with High Sierra? Does it work with out any messing around?

Any recommendations on a starter (not too expensive) eGPU for this use case? Any employees at Apple in the ML team have any advice?

I am pretty new to machine learning and have never used an eGPU so would appreciate any advice, recommendations or corrections to my statements above.

I know that there are rental options offered by Amazon and Google, but would prefer to buy my own eGPU.

Thanks.

Sincerely,

Ajay

Boost

Answer 1

igerard OP

Jun ’17

Hi,

I have a setup with Sierra and an eGPU with a GTX 980 Ti, webdriver installed and CUDA 8 working, but I didn’t have yet the time to test TensorFlow accelerated on it.

something to know is that TensorFlow stop supporting GPU on macOS, bad !

not sure that there is any hope to see a Webdriver supporting Metal 2 in the near future, then High Sierra seems not the version to use.

I Will try to test Tensorflow gpu accelerated on my config this week-end and I will give an update.

the site to,visit for all eGPU related information : egpu.io Great community 🙂

0