Does this work with the Nvidia Tesla K80 GPU accelerator?
Or does it only work with typical Nvidia video cards? Thanks
Hi robmausser,
redinferno is right in saying that CuDNN supports multiple GPU's, however it requires the program to address the available GPU's individually.
Dain does not currently support the use of multiple GPU's in one instance as it would require the program to decide which GPU is going to do what and that requires extra programming which is outside the scope of the original Dain project which was for an academic research paper.
However, the answer to your specific question is no. Dain lets you target whatever single GPU you want to use and DainApp gives you a nice dropdown list :) so you could use whatever Nvidia GPU you want as long as it supports compute 5.0, 5.2, 6.0 and 6.1 or higher (if manually configured). The Tesla K80 uses compute 3.7 which is not naively supported by the Dain project.
Edit: You can run multiple instances of DainApp. I've not tried this myself as I only have one GPU in this PC right now but you might be able to use multiple instances of DainApp each using a different GPU to interpolate multiple files at once. Can someone test this or do we already know? End of Edit
If you have a K80 lying around then go for it, but if you are looking for a GPU to buy, have a look at CUDA benchmarks and compute capability before you purchase as the K80 has less CUDA performance than even the GeForce GTX 1050 which has compute 6.1. The K80 does have a HUGE amount of RAM which means you probably wont ever have to split frames even if they were 4K, but the actual processing time will be excruciating even if it works which it might not (without manual configuration and maybe not even then).
I hope this helps,
Theo