Jetson TK1: Cannot install cuDNN

Hello,

I just installed JetPack to my newly bought Jetson TK1, and installed Torch on it. Torch is running correctly, but when I want to install cuDNN using

luarocks install cudnn

it gives strange errors as follow:

Building on 4 cores
-- The C compiler identification is GNU 4.8.4
-- The CXX compiler identification is GNU 4.8.4
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Found Torch7 in /home/ubuntu/trunk/torch/install
-- Found CUDA: /usr/local/cuda-6.5 (found suitable version "6.5", minimum required is "6.5") 
-- Removing -DNDEBUG from compile flags
-- TH_LIBRARIES: TH
-- MAGMA not found. Compiling without MAGMA support
-- Autodetected CUDA architecture(s): 3.2
-- got cuda version 6.5
-- Could not find CUDA with FP16 support, compiling without torch.CudaHalfTensor
-- CUDA_NVCC_FLAGS: -gencode;arch=compute_32,code=sm_32
-- THC_SO_VERSION: 0
-- Performing Test HAS_LUAL_SETFUNCS
-- Performing Test HAS_LUAL_SETFUNCS - Failed
-- Configuring done
-- Generating done
-- Build files have been written to: /tmp/luarocks_cutorch-scm-1-9108/cutorch/build
[  1%] [  2%] [  3%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/generated/./THC_generated_THCTensorMaskedDouble.cu.o
[  4%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir//./THC_generated_THCReduceApplyUtils.cu.o
Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir//./THC_generated_THCBlas.cu.o
Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir//./THC_generated_THCSleep.cu.o
/tmp/luarocks_cutorch-scm-1-9108/cutorch/lib/THC/THCBlas.cu(405): error: identifier "cublasSgetrsBatched" is undefined

/tmp/luarocks_cutorch-scm-1-9108/cutorch/lib/THC/THCBlas.cu(422): error: identifier "cublasDgetrsBatched" is undefined

2 errors detected in the compilation of "/tmp/tmpxft_00000d2e_00000000-6_THCBlas.cpp1.ii".
CMake Error at THC_generated_THCBlas.cu.o.cmake:267 (message):
  Error generating file
  /tmp/luarocks_cutorch-scm-1-9108/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCBlas.cu.o

make[2]: *** [lib/THC/CMakeFiles/THC.dir/./THC_generated_THCBlas.cu.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [lib/THC/CMakeFiles/THC.dir/all] Error 2
make: *** [all] Error 2

Error: Failed installing dependency: https://raw.githubusercontent.com/torch/rocks/master/cutorch-scm-1.rockspec - Build error: Failed building.

I cannot find any information regarding this online, could someone help on it? Thank you in advance.

NVIDIA is not the provider of lua or luarocks or torch

If you want to install CUDNN on your Jetson, I believe there are instructions provided by NVIDIA that you could follow. Those are available on the web page that you download cudnn from. However, I don’t think your problem at the moment has anything to do with cudnn.

It appears to be related to cublas, and it appears that your torch setup is expecting a function in cublas that was not available until a CUDA version later than CUDA 6.5. It was introduced in CUDA 7.0

Since you have CUDA 6.5 installed, you don’t have it.

If you install the latest jetpack on your jetson, I believe you will get CUDA 7 or later.

https://developer.nvidia.com/embedded/jetpack

also note there are jetson-specific forums here.

The JetPack I installed is the latest version, but the point is, it only installed CUDA 6.5 on the device. Seems like TK1 only supports CUDA 6.5 version.
Btw, I’ve already done many internet search before posting here.

Yep, my mistake, Jetson TK1 is limited to CUDA 6.5:

https://devtalk.nvidia.com/default/topic/805540/no-cuda-7-0-support-for-jetson-tk1-board/

So you should probably get yourself a Jetson TX1, or else use something besides the current version of Torch on TK1.

My comment on the forums was to point out that you may get better answers on this Jetson TK1 question by posting it in the Jetson TK1 forum here:

https://devtalk.nvidia.com/default/board/162/jetson-tk1/