Previous | Next --- Slide 38 of 41
Back to Lecture Thumbnails
althalus

If DNN are more reliant on GPUs, does this mean that DNNs try to deal with large amount of computation rather than dealing with cache issues?

blairwaldorf

@althalus: what do you mean by "rather than dealing with cache issues"?

althalus

@blairwaldorf, I meant that when DNNs are designed, the designers choose to consider how to deal with large amounts of computations rather than to deal with cache issues since its on the GPU.

patrickbot

NVIDIA (on their pascal GPUs) have been working on DNNs.

https://developer.nvidia.com/cudnn

After a cursory glance of the webpage, it looks like they focus on tuning computation for the GPUs.

bdebebe

A recent CMU student, George Hotz, is making a self-driving car using DNN's. With this increased computing power, his research is becoming reality, as you can see in this video: https://www.youtube.com/watch?v=YjTnYBaQQpw