none
GPU Cuda not working? RRS feed

  • Question

  • Hello!

    I am doing the `test drive` with the data science linux machine before trying it out further.

    I primarily want to use this DSVM for the GPU, and the fact that cuda should be preconfigured,

    but it doesn't seem to work for me.

    I can see that CUDA is installed.

    <style type="text/css">p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Menlo; color: #ffffff; background-color: #272822} span.s1 {font-variant-ligatures: no-common-ligatures} </style>

    nvcc --version

    nvcc: NVIDIA (R) Cuda compiler driver

    Copyright (c) 2005-2016 NVIDIA Corporation

    Built on Tue_Jan_10_13:22:03_CST_2017

    Cuda compilation tools, release 8.0, V8.0.61

    However, I can't detect the GPU.

    <style type="text/css">p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Menlo; color: #ffffff; background-color: #272822} span.s1 {font-variant-ligatures: no-common-ligatures} </style>

    $ nvidia-smi

    NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

    Furthermore, when I use `pytorch` it does not connect/detect cuda/gpu.

    Any advice? 

    Friday, January 19, 2018 6:10 AM

Answers

  • Hi,

    The test drive does not provide GPU instances. The Ubuntu DSVM supports deep learning workloads on the CPU as well, so you are welcome to run samples there - perhaps with smaller datasets or fewer epochs. Many of our customers take advantage of this feature to complete their initial development and testing of deep learning code on a CPU instance, then resize the VM to have one or more GPUs when they are ready to scale up.

    Friday, January 19, 2018 7:20 PM
    Owner