Accelerate Machine Learning with the cuDNN Deep Neural Network Library

Technical Blogs & Events Technical Blog
1

Originally published at: https://developer.nvidia.com/blog/accelerate-machine-learning-cudnn-deep-neural-network-library/

Machine Learning (ML) has its origins in the field of Artificial Intelligence, which started out decades ago with the lofty goals of creating a computer that could do any work a human can do. While attaining that goal still appears to be in the distant future, many useful tools have been developed and successfully applied…

2

Very exciting! Thanks Larry and nVidia! It will be interesting to see if there are any architectural changes that can be made to support deep learning and other new AI architectures.

3

Awesome!

4

Hi, all, there is a similar library that can be found from http://libnn.com. It is totally free.

5

There is a caffe version, optimized for CPU ("openmp" branch). Imagenet training on this CPU-version (MKL + openmp, dual-socket E5-2680 ) is < 2x ( not 11x! ) slower than caffe-GPU (cuBLAS, K40).

6

This is awesome!

7

This is great

8

Dr. Niu, why cuDNN is awesome?

9

that's great!

10

Hi Larry ,

I want to register and download cuDNN but I could not be able to download. When I pressed on register then submit no thing is happen and when I pressed the downlaod I recieved this message "n Error message You do not have permission to view this form.". So please any help.

11

Dear Larry and others

I want to register and download cuDNN but I could not be able to download or register. When I had pressed on register then submit no thing was happen and when I pressed the download I received this message "n Error message You do not have permission to view this form.". So please any help.

Best regards,

Salem

12

It turns out that cuDNN comes with windows version. My windows porting of Caffe can be hopefully accelerated as well.

13

Given that cuDNN seems to be about adding DNN primitives, what exactly can be expect of the "support for splitting computation across multiple GPUs on the same node"?

What sort of computations will be split?

14

It would be great if the library worked with the Jetson TK1 board. Are there any plans to provide binaries for ARM?

15

is there ANY working example of this thing ? ANY documentation (besides the PDF file bundled with the library) ?

16

It would be great if the library worked.

=)

17

Stay tuned...all I can say is you won't have to wait too long...

18

What are you trying to accomplish? You can post questions on the NVIDIA Developer Forums and we will do our best to answer and help. cuDNN is integrated with development branch of CAFFE right now, and you should be able to post on the CAFFE forums to get help with that. Once CAFFE v1.0 is officially launched, there will be easy to follow instructions on how to enable cuDNN. cuDNN is also rapidly being incorporated into other frameworks as well. The cuDNN User Guide and the article are what exists at the moment, but that seems to be enough for many folks to do successful integration.

19

Those folks are much smarter then me. I am just a humble developer trying to see if this library is of any use for me. I am trying to have Boltzmann machine running on a GPU cluster, but convolutional network is also great.

20

ANY working examples anyone ?