Soumith Chintala

@benanne http://arxiv.org/abs/1412.7580

@apark263 the numbers were very arbitrarily chosen to what I thought might be a "typical" L2. I've moved to more realistic and real-world "imagenet-winners" benchmarks for exactly this reason, that...

Hey @dadaism. I fixed it in my cuda-convnet2.torch fork by rewriting some of the kernels without texture loading and adding them to the dispatcher for the case when the tensor...

It could be done. And the corner cases can be loaded via a regular load. It would also remove this 512mb limit in ccn2. No one has done it, that's...

Some of the libs don't support in-place relu (without which you won't fit minibatch 64), and I was going to try moving the benchmarks to gtx980, so I made this...

@andravin I think you make a good point. I will just make it mini-batch size 64. Waiting for a new card for that to happen.

I just finished up benchmarking on Titan-X. What's extremely exciting is how fast GPUs are going, and even more exciting is the fact that people are pushing the limits of...

@skelleher right on the front-page readme: https://github.com/soumith/convnet-benchmarks/blob/master/README.md It's interesting to compare just the GPUs themselves. For example: Titan-X vs Titan Black (which is slightly faster than 780, as fast as...

@f0k The repo I benchmarked is: https://github.com/NervanaSystems/nervanagpu . It is private and you need to request access from the Nervana guys. In this new repo they have ops that take...

Completely forgot, @hughperkins 's DeepCL numbers added. DeepCL is unique in the sense that it's the only real contender at this point for OpenCL based deep learning, Huge props to...