Running the sample ./mnistCUDNN failed on RTX 2070

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Installed CUDA 10.1.243(update2) from local deb package and cuDNN (7.6.5) and facing the following issue when checking ./mnistCUDNN

cudnnGetVersion() : 7605 , CUDNN_VERSION from cudnn.h : 7605 (7.6.5)
Host compiler version : GCC 7.4.0
There are 1 CUDA capable devices on your machine :
device 0 : sms 36  Capabilities 7.5, SmClock 1710.0 Mhz, MemSize (Mb) 7949, MemClock 7001.0 Mhz, Ecc=0, boardGroupID=0
Using device 0

Testing single precision
Cublas failure
Error code 1
mnistCUDNN.cpp:404
Aborting...

The same with CUDA 10.2 (and cuDNN 7.6.5.32), works just fine. Is it a known issue?

2

Hi,

It might be due to cuDNN-CUDA compatibility issue.
Please refer below link for cuDNN and CUDA/Driver version compatibility:
https://docs.nvidia.com/deeplearning/sdk/cudnn-support-matrix/index.html#cudnn-cuda-hardware-versions

Thanks

3

Nope, it’s due to the libcublas library error, there are some issues with symlinks in CUDA 10.1

Refer, https://devtalk.nvidia.com/default/topic/1047981/cuda-setup-and-installation/cublas-for-10-1-is-missing/

Basically, what I did was created a symlink to copy libcublas, and libcublasLt via

sudo ln -s /usr/lib/x86_64-linux-gnu/libcublasLt.so.10.2.1.243 /usr/local/cuda/lib64/libcublasLt.so
sudo ln -s /usr/lib/x86_64-linux-gnu/libcublas.so.10.2.1.243 /usr/local/cuda/lib64/libcublas.so
4

Experiencing the same issue with my 1080 Ti. Unfortunately, setting the symbolic links exactly like Cataluna did not solve the problem for me.

Output of ./mnistCUDNN

cudnnGetVersion() : 7605 , CUDNN_VERSION from cudnn.h : 7605 (7.6.5)
Host compiler version : GCC 7.4.0
There are 3 CUDA capable devices on your machine :
device 0 : sms 28  Capabilities 6.1, SmClock 1582.0 Mhz, MemSize (Mb) 11178, MemClock 5505.0 Mhz, Ecc=0, boardGroupID=0
device 1 : sms 28  Capabilities 6.1, SmClock 1582.0 Mhz, MemSize (Mb) 11178, MemClock 5505.0 Mhz, Ecc=0, boardGroupID=1
device 2 : sms 28  Capabilities 6.1, SmClock 1582.0 Mhz, MemSize (Mb) 11176, MemClock 5505.0 Mhz, Ecc=0, boardGroupID=2
Using device 0

Testing single precision
Cublas failure
Error code 1
mnistCUDNN.cpp:404
Aborting...

UPDATE: Solved the issue by downgrading the version to 10.2.1.243 like Cataluna proposed but additionally replacing the symlinks inside the /usr/lib/x86_64-linux-gnu/ folder:

sudo ln -s /usr/lib/x86_64-linux-gnu/libcublasLt.so.10.2.1.243 /usr/lib/x86_64-linux-gnu/libcublasLt.so.10
sudo ln -s /usr/lib/x86_64-linux-gnu/libcublas.so.10.2.1.243 /usr/lib/x86_64-linux-gnu/libcublas.so.10