hughperkins / distro-cl

OpenCL Torch
147 stars 17 forks source link

Cannot require CUDNN in Torch! Please help! #26

Open Hidden-dreamz opened 7 years ago

Hidden-dreamz commented 7 years ago

When i require 'cudnn' i get this error on ubuntu 16.04, nvidia driver 378.13, Cuda 8.0 (full newest release), Cudnn 6 (same problem with 5.1) using torch-cl and installed all cutorch cunn cudnn with laurocks. Torch works with my gpu i just cant get CUDNN to work. Please help! thanks! (and im using 1070 gpu)

/home/josh/torch-cl/install/share/lua/5.1/cudnn/init.lua:98: attempt to call field 'hasFastHalfInstructions' (a nil value) stack traceback: /home/josh/torch-cl/install/share/lua/5.1/cudnn/init.lua:98: in function 'fasterHalfMathTypeForCurrentDevice' /home/josh/torch-cl/install/share/lua/5.1/cudnn/init.lua:112: in function 'configureMath' /home/josh/torch-cl/install/share/lua/5.1/cudnn/init.lua:131: in main chunk [C]: in function 'require' stdin:1: in main chunk [C]: at 0x00405e90

hughperkins commented 7 years ago

distro-cl is for opencl. cunn, cudnn are the cuda versions. Please use require clnn instead.

Hidden-dreamz commented 7 years ago

i want to use the cudnn version how to go back to using the standard torch distro if everything is linked to the opencl one? thanks for the help

hughperkins commented 7 years ago

You'll wnat to install into two separate folder, one for distro-cl, one for standard torch, eg, you could install distro-cl into ~/torch-cl, and the standard distro into ~/torch

To choose the one you want, simply call activate from the one you want to use, something like:

source ~/torch/bin/torch-activate
NerminSalem commented 7 years ago

Hi can you help, i m using cpu mode as my laptop doesn't support NIVIDIA and i installed cltorch and clnn and set gpu= -1 but still i m getting libthclnn_searchpath /home/nermin/torch-cl/install/lib/lua/5.1/libTHCLNN.so
terminate called after throwing an instance of 'std::runtime_error' can u help?

hughperkins commented 7 years ago

Can you check the result of running clinfo please? You need at least one GPU available, visible from clinfo.