Saurabh_29

5 ms·

1. 1.▲
   Fast approximate knn using random projections(github.com)

   3 points by Saurabh_29 5y ago|1 comments
2. 2.▲
   by Saurabh_29 5y ago

   Hi, C++ implementation of approximate nearest knn which doesn't suffer from problem of curse of dimentionality whilst giving theoretical bounds on approximation.
3. 3.▲
   Text to Speech CUDA Programming(github.com)

   3 points by Saurabh_29 7y ago|0 comments
4. 4.▲
   by Saurabh_29 7y ago

   +1, it is the biggest bottleneck at times
5. 5.▲
   by Saurabh_29 7y ago

   Also, in some cases like small RNN/LSTMs, CPU's can be faster.
6. 6.▲
   by Saurabh_29 7y ago

   I will agree with you. It is a combination of both. Plus I would like to add some extra points, pytorch/tensorflow essentially use the same CUDA/Cudnn libraries, the thing they are developed with the motive to catering wide corner
7. 7.▲
   by Saurabh_29 7y ago

   The main bottleneck is the time spent in adding bias after Conv/dense. A well-optimized code can remove that bottleneck. I haven't done it in my code as it makes it less readable.
8. 8.▲
   by Saurabh_29 7y ago

   The main bottleneck is the data transfer speed between the GPU and the SMs. Also, using tensor core doesn't necessarily apply using half-precision as now NVIDIA supports single-precision operation in Tensorcore too.
9. 9.▲
   Waveglow Inference in CUDA C++(github.com)

   72 points by Saurabh_29 7y ago|10 comments