GPU choice for Convolutional Neural Networks


I am new on the forum and I would like some advice about what GPU is the best option for my research. I am working on Computer Vision, more specifically, on Deep Convolutional Neural Networks. I use Caffe library and large datasets of images.

Because of that, I would need a GPU with enough memory to deal with high resolution images. My first option was the Titan X Pascal, however, I found that Tesla K40 and Tesla K80 are also good at deep learning.

I am sure about choosing Titan X opposite Tesla K40. However, it would be interesting for me to try the multi-GPU usage of Caffe. So my question is, what would be better? Two Titan X or one Tesla K80? I don’t know much about the hardware details of connecting 2 GPUs, so I ask.

Thank you for your help.

The 12 GB Pascal (or Maxwell) Titan X will be far better for deep learning than either the Tesla K40 or Tesla K80. You most likely to not need double precision and the GFLOPS are better for the Titan X.

The GTX 1080 with 8GB is a good choice as well.