unable to build Sample Mnist - TensorRT


I get the following error when I try to run sample_mnist

ERROR: cudnnLayerUtils.cpp (288) - Cuda Error in smVersion: 35
terminate called after throwing an instance of 'nvinfer1::CudaError'
 what():  std::exception

where, I have flashed L4T 28.1 from Jetpack 3.1 and CuDNN, Cuda, TensorRT, OpenCV 3.3 and Multimedia API from JetPack 3.2


Looks like your program is compiled with incorrect GPU architecture.
It is abnormal to find sm=35 on TX2 platform since it is sm=62 architecture.

Do you build TensorRT sample code right on the TX device?
Could you try to compile it with following command again?

cp -r /usr/src/tensorrt/ .
cd tensorrt/samples/sampleMNIST
cd ../../bin/


yea @AastaLLL, the error statement specified some architectural mismatch
thats why, I tried to find where sm=35 is set in make files, but could not find this

and yes I built it on TX2 from /usr/src/tensorrt

however I cannot check again as per your suggestion, since I am on L4T 28.2 now and error was appearing on 28.1


Okay, guess we find the problem.

Due to different driver version, you can’t install the libraries from JetPack3.2 but use the JetPack3.1 OS.
Please remember to use the OS and libraries from the same installer to avoid incompatibility.


yup thats the case