TX2 ERROR: Check failed: error == cudaSuccess (8 vs. 0) invalid device function

Autonomous Machines Jetson & Embedded Systems Jetson TX2
#1

Hi,

I am following jetson-inference example on github for the detectnet example and encountered the following error during training the model.

ERROR: Check failed: error == cudaSuccess (8 vs. 0) invalid device function
conv1/7x7_s2 → conv1/7x7_s2
Check failed: error == cudaSuccess (8 vs. 0) invalid device function

The training stop essentially in one second after commencing the job on Digit web interface. I am using Ubuntu 14.4 and Quadro K4000 on the host machine.

Appreciate the help in advance.

Thanks.

#2

Hi,

Error code=8 is the invalid device.

Suppose you install NvCaffe as framework; please add the correct GPU architecture to the configure file.
[url]caffe/Makefile.config.example at caffe-0.16 · NVIDIA/caffe · GitHub

Please find GPU architecture information on this page:
[url]https://developer.nvidia.com/cuda-gpus[/url]

Thanks.

#3

Hi AastaLLL,

Thanks for the reply. Am I suppose to look up the compute capability for for card which is 3.0 for Quadro K4000 and replace line 37-41? So this is the original block of code for CUDA_ARCH

CUDA architecture setting: going with all of them.

CUDA_ARCH := -gencode arch=compute_50,code=sm_50
-gencode arch=compute_52,code=sm_52
-gencode arch=compute_60,code=sm_60
-gencode arch=compute_61,code=sm_61
-gencode arch=compute_61,code=compute_61

CUDA architecture setting: going with all of them.

CUDA_ARCH := -gencode arch=compute_30,code=sm_30

Thanks.

#4

Hi,

Could you advise where is the configuration file is located in a Window10 environment?
My NVidia card is RTX 3070 … and I am trying to run an application call OpenPose and encounter same error when trying to run this application. I understand by setting should … compute_86, code=sm_86 based on this thread of emails but no idea where is this configuration file it is referring to.

This card is working fine in Blender and other application that I have that uses GPU Cuda.

#5