3 ms·
This is incredible work. Predicting, sharing, asynchronous compilation, and reverse-engineering the pipeline are all very creative solutions to a really difficu
by peterburkimsher 9y ago
This is incredible work. Predicting, sharing, asynchronous compilation, and reverse-engineering the pipeline are all very creative solutions to a really difficult problem. As I understand, deep learning basically runs graphics cards backwards to generate text from images.
How can we apply these excellent algorithms to machine learning?
- 21 9y agoMachine learning doesn't run the graphics card backwards. What they did is not really useful for ML. As said by them, their ubershader is massively inefficient.
- partingshots 9y agoThis is incredibly false. How did you even manage to get this picture of what deep learning is? I'm genuinely curious as to what articles/papers you've read that made you think deep learning is basically "running a graphics card backwards".