CODEWITHSUNDEEP
ubuntutensorflowopenclgpu
sebap123
sebap123

Reputation: 2685

TensorFlow cannot find integrated GPU

I'm trying to run TensorFlow on my integrated Intel GPU on machine that has both integrated and discrete graphic card.

user@host:~$ lscpu | grep "Model name"
Model name:          Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz

When I'm running TensorFlow's script to check list of devices

from tensorflow.python.client import device_lib

def get_available_gpus():
    local_device_protos = device_lib.list_local_devices()
    return [x.name for x in local_device_protos]
    print(get_available_gpus())

I'm getting following result

18:30:21.335442: E tensorflow/stream_executor/cuda/cuda_driver.cc:397] failed call to cuInit: CUDA_ERROR_UNKNOWN
18:30:21.335465: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:145] kernel driver does not appear to be running on this host (host): /proc/driver/nvidia/version does not exist
['/device:CPU:0', '/device:XLA_CPU:0', '/device:XLA_GPU:0']

I suppose this might be related to some problem with OpenCL, because when I'm running clinfo I'm getting

user@host:~$ clinfo
Number of platforms                               0
user@host:~$ ls /etc/OpenCL/vendors/
intel.icd  nvidia.icd

Another thing is that when I run some sample testing application I found online

#include "CL/cl.h"
#include <stdio.h>

int main()
{
    cl_platform_id pid;
    cl_uint num;
    cl_uint n=1;
    cl_int error=clGetPlatformIDs(1,&pid,&num);
    printf("Error code= %d\nNo. of platforms= %d",error,num);
    getchar();
}

I'm getting error code -1001 and 0 platforms. As I found out this might mean I need to install drivers, but as I understand looking at /etc/OpenCL/vendors/ I already have them installed.

In PRIME Profiles I have

user@host:~$ sudo prime-select query
[sudo] password for user:
intel

Switching to nvidia clinfo is working properly as well TensorFlow.

I'm running Ubuntu 18.04 LTS

user@host:~$ uname -a
Linux host 4.15.0-30-generic #32-Ubuntu SMP Thu Jul 26 17:42:43 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

Additional info I think might be helpful in debugging is that I'm working on this machine remotely without any display connected directly.

I'd really appreciate any suggestion what I'm missing here to fix TensorFlow and clinfo issue.

Upvotes: 3

Views: 5893

Answers (1)

Biren Mer
Biren Mer

Reputation: 9

First of all please check the CUDA and cuDNN version you are using is correct or not, FROM:- https://www.tensorflow.org/install/source for your corresponding OS and TF version.

IF that's correct then do check if you are using tensorflow-gpu not tensorflow-cpu.

Also I suggest you to work with the discrete GPU that you have on your machine instead of the integrated one.

Here's a link that might help Issue as mentioned above

Upvotes: 1

Related Questions