Microsoft CNTK

Anybody try it? It requires 7.5 for one thing, perhaps 7.0 would work. I’ll take a look at it this weekend regardless.

In the process of installing CNTK. Surprisingly the BSP basic install doesn’t seem to have much, not even OpenCV. But after doing a few updates and a bunch of installs its running along.

However the problem with running this appears to be (interestingly enough given this is an NVIDIA board) CUDA. CNTK requires CUDA 7.0 (not 7.5), on installing it complained about being an unsupported platform. Going ahead and attempting a CUDA sample I get

"/usr/local/cuda-7.0"/bin/nvcc -ccbin g++ -I../../common/inc  -m64    -dc -gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_52,code=sm_52 -gencode arch=compute_52,code=compute_52 -o cdpSimpleQuicksort.o -c cdpSimpleQuicksort.cu
/usr/local/cuda-7.0/bin/nvcc: 1: /usr/local/cuda-7.0/bin/nvcc: cannot open 6L▒u▒C▒▒▒T▒5Q▒: No such file
/usr/local/cuda-7.0/bin/nvcc: 1: /usr/local/cuda-7.0/bin/nvcc: /usr/local/cuda-7.0/bin/nvcc: 1: /usr/local/cuda-7.0/bin/nvcc: ⌂ELF: not found}▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒C▒y▒?▒▒▒#oF▒▒vN▒▒▒Z▒=_♥?▒▒►WP►k▒▒MZh: not found

/usr/local/cuda-7.0/bin/nvcc: 2: /usr/local/cuda-7.0/bin/nvcc: Syntax error: word unexpected (expecting ")")
make[1]: *** [cdpSimpleQuicksort.o] Error 2
make[1]: Leaving directory `/home/ubuntu/NVIDIA_CUDA-7.0_Samples/0_Simple/cdpSimpleQuicksort'
make: *** [0_Simple/cdpSimpleQuicksort/Makefile.ph_build] Error 2

During the install I also got the following error

Missing recommended library: libGLU.so
Missing recommended library: libX11.so
Missing recommended library: libXi.so
Missing recommended library: libXmu.so

How do we get CUDA working, is (I hope!) 7.5 supported?

Hi shallownet,

TX1 is locked down with CUDA 7.0, you may refer to below link to get CUDA working on TX1.
http://developer.download.nvidia.com/compute/cuda/7_0/Prod/doc/CUDA_Getting_Started_Linux.pdf

Thanks

Currently, JetPack 3.1 contains CUDA 8.0, and Microsoft announced that CNTK moved to CUDA 8 on 04/03/2017.
https://docs.microsoft.com/en-us/cognitive-toolkit/archive/cntk-move-to-cuda8

So is CUDA version not an issue now for running CNTK on TX1?

Hi,

Besides CUDA version, another critical issue is if CNTK supports ARM platform.

Currently, Jetson is 64-bit ARM-based Linux system.
From the following comment, CNTK team plans to support 64-bit ARM system at the end of this year.

For the latest progress, please check with CNTK team here.

Thanks.

Hi,

I’ve been trying to build CNTK on Jetson TX1.

The README.md page on the following GitHub repository describes the build steps. It builds CNTK 2.5 GPU version.