Therefore this module is much faster than the wrappers around nvidia-smi. TensorFlow relies on a technology called CUDA which is developed by NVIDIA. Install synaptic and atom from Ubuntu's package manager search and install nvidia-390 from synaptic Download CUDA 9. This can be downloaded here. You should see the details of your GPU printed on the terminal. John Kirkham (Conda Forge, NVIDIA, Dask) is taking a look at this along with the UCX developers from Mellanox. sudo apt-get update && sudo apt-get --assume-yes upgrade sudo apt-get --assume-yes install tmux build-essential gcc g++ make binutils sudo apt-get --assume-yes. I'd like to know whether this situation is normal. This is going to be a tutorial on how to install tensorflow 1. 次に GPU に対応させるため CUDA と cuDNN が必要となる. conda create -n envname python=2. If you want accurate usage statistics use nvidia-smi or modify the GPU setting in Task Manager to "CUDA usage". 04 搭建深度學習開發環境 RTX 2080 + CUDA 10. To check the installation, type: $ docker run --runtime=nvidia --rm nvidia/cuda nvidia-smi. 更新package conda update -n py27 numpy. 0 which requires NVIDIA Drivers 384. CUDA 5 added a powerful new tool to the CUDA Toolkit: nvprof. For me, nvidia-smi showed ERR as the device name, so I installed the latest Nvidia drivers (as of October 2018) to fix it:. 111 Run the following command to check whether the driver has installed successfully by running NVIDIA’s System Management Interface (nvidia-smi). Solution : Reinstall your NVIDIA GPU driver using this instruction. 0和内核驱动一起安装上,而且确实安装了名为nvidia-kmod的包,但重启后执行nvidia-smi. Again restart the PC and after that run the below command. Getting started. C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9. sudo apt install nvidia-driver-410 nvidia-settings ## sudo ubuntu-drivers autoinstall; reboot system sudo reboot; check nvidia-smi; Part 2. Pay attention that in the above method, we install the latest Nvidia drivers. 04 So I installed tensorflow and the CPU Version works fine but I can't seem to get the GPU to work. I want to use graphics card for my tensorflow and I have installed and re-installed again but tensorflow is not using GPU and I have also installed my Nvidia drivers but when I run nvidi-smi then a command is not found. If you have a zombie process using it you can try this. Install Conda. When I import pytorch in python, and it shows. conda create -n py36 python=3. 0を指定しているので、tensorflow自体やkerasのバックエンドでtensorflowを使う場合は9. And Keras will utilize the GPU since it's using TensorFlow as the backend. POSTS Installing Nvidia, Cuda, CuDNN, Conda, Pytorch, Gym, Tensorflow in Ubuntu October 11, 2019. This is graphics reinvented. Do you really want to understand creating your own deep learning network? Have you tried and didn’t succeed yet? Let me take you to the journey of creating your own deep learning network using Tensorflow 2. html > conda install Flask > python web. I want to use graphics card for my tensorflow and I have installed and re-installed again but tensorflow is not using GPU and I have also installed my Nvidia drivers but when I run nvidi-smi then a command is not found. Before we install NVIDIA Drivers, we need to remove any existing version with the following command. I had exactly the same issue. 04 update apt-get Install apt-get deps inst. Photo by Fernand De Canne on Unsplash. 次に, cudaとNVIDIA driverの対応を確認します. Deep learning and AI frameworks for the Azure Data Science VM. Experimenting with OpenMP 4 / OpenACC (code annotation of Fortran, C or C++) on GCC 7. terminal 1: tmux new -s train conda activate keras time python train_alexnet. import tensorflow. conda activate env_name. watch -n -1 nvidia-smi to monitor the GPU usage. nvidia-smi NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Activate the correct conda environment,. I already explained the benefits of CUDA and even showed a simple code example. 1 could be installed on it. nvidia-smi nvcc --version conda install cudnn==7. nvidia-smi # Install development and runtime libraries (~4GB). 我更喜欢使用nvidia-smi来监视GPU的使用情况。如果您在开始编程时显著增加,则表明您的tensorflow正在使用GPU。 第6种办法. Install nvidia driver. Use discourse. Releases all unoccupied cached memory currently held by the caching allocator so that those can be used in other GPU application and visible in nvidia-smi. x can be installed with either conda or pip package managers and also from source. Set the dockerRuntime value to be nvidia. I work with GPUs a lot and have seen them fail in a variety of ways: too much (factory) overclocked memory/cores, unstable when hot, unstable when cold (not kidding), memory partially unreliable, and so on. source activate dlenv Install baisc packages. the Nvidia card of my system:enter image description here. Occasionally it showed that the Python process is running, but otherwise it was not useful to me. conda install pytorch torchvision cuda80 -c soumith. Forum rules Read the FAQs and search the forum before posting a new topic. nvidia-smi 명령어로 현재 설치된 그래픽 드라이버를 확인해본다. conda install tensorflow-gpu. Click on the green buttons that describe your target platform. deb sudo apt-get update sudo apt-get install cuda-drivers sudo reboot 检查驱动是否安装成功 #安装成功 输出gpu信息 nvidia-smi. I installed Nvidia 396) Then you check the other link that I posted for guidance on installing tensorflow with conda on linux. These nodes contain dual E5-2680 processors (24 total cores), and 256GB of RAM. Used by thousands of students and professionals from top tech companies and research institutions. x can be installed with either conda or pip package managers and also from source. However, to collect GPU stats on systems with one or more GPUs, ensure that nvidia‑smi is installed. cz - 16 cores / 32 threads, 256GB RAM, 500GB SSD, 8 x NVIDIA GTX 1080Ti The servers are running continuously and supports simultaneous work of several users, but each user should work only with 1 GPU card at same time. Nvidia driver version mismatch (which cause tensorflow gpu not work) Then the "nvidia-smi" should work. We have discussed about GPU computing as minimally needed theoretical background. However after installing the vib using the Grid quick. conda create -n envTF113 python=3. 0 -c pytorch. This should have installed the display driver as well. Behind the scenes, we are zipping the current working directory, creating a Docker container, and running the command you provided. NVIDIA 드라이버를 설치하고, 아나콘다를 기반으로 CUDA, cuDNN 라이브러리 등과 GPU를 지원하는 텐서플로우 버전을 설치하는 방법에 대해서 설명드립니다. 1 -c pytorch instead of installing it with the installtion snippet generated for me by pytorch website: conda install pytorch torchvision cudatoolkit=10. If you are using conda, you might have installed the cpu version of the tensorflow. keras Create Keras json configuration file. nvidia-smi と入力してドライバの状態を確認する*1. We will also be installing CUDA 10. However, after reinstalling my operating system and the drivers, every time I run nvidia-smi it prints the system time and date and then prints nothing else. cd C:\Program Files\NVIDIA Corporation\NVSMI nvidia-smi. Hi, The GPUs are exposed to Jupyter, and TensorFlow should be able to find them. At the moment you can't just run install , since you first need to get the correct pytorch version installed - thus to get fastai-1. They will be shown as 4 separate devices to CUDA code. 发现内存泄露问题,即没有进程时,内存被占用,0号GPU上运行的python程序已经退出,但是显存没有被释放. nvidia-smi nvcc --version conda install cudnn==7. total,driver_version,temperature. 2 如果nvidia-smi下没有显示任何进程. [[email protected] ~]$ conda create -n cuda python=3. I want to use graphics card for my tensorflow and I have installed and re-installed again but tensorflow is not using GPU and I have also installed my Nvidia drivers but when I run nvidi-smi then a command is not found. 12 GPU version. Copy the following into a cell and run the cell. 04 is not listed in the. 参考:Tensorflow GPU Installation Made Easy: Use conda instead of pip. Before doing these any command make sure that you uninstalled the normal tensorflow. in your terminal, issue the following command: $ watch -n 1 nvidia-smi. See the main installation article for details on other available options (e. I completed the process successfully once, and was able to run nvidia-smi to see my graphics card usage. After this, I installed TensorFlow 1. Turn off Secure Boot (necessary to load NVIDIA driver in Ubuntu kernels 4. まずは, NVIDIA driverのversionを確認します. nvidia-smi And you should see something like this:. the Nvidia card of my system:enter image description here. 04 machine with NVIDIA's new GTX 1080 Ti graphics card for use with CUDA-enabled machine learning libraries, e. conda install -c pytorch pytorch cuda100 Below are the instructions for installing CUDA using the. 04 LST 버전이다. 加权随机采样算法--Weighted Sampling Algorithm; svn命令行常用操作. This tool is optional and Guild will run without it. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. Feel free to spin up an instance of your own and follow along. Linux Driver Linux Tutorial. NVidia 1060 sudo add-apt-repository ppa:graphics-drivers sudo apt-get update sudo apt-get install nvidia-384 sudo apt-get install nvidia-367 sudo apt-get install nvidia-smi [email protected]:~$ lsmod | grep -i nvidia nvidia_uvm 671744 0 nvidia_drm 45056 1 nvidia_modeset 843776 5 nvidia_drm nvidia 13119488 268 nvidia_modeset,nvidia_uvm drm_kms_helper 151552 2 i915,nvidia_drm drm 352256 5 i915,nvidia_drm. The workstation you are going to use TensorFlow has CentOS 7 or Red Hat Enterprise Linux Workstation 7; Ubuntu is often the most often used examples on the Internet, at Brown, we tend to use Red Hat and CentOS for most work, especially if you require support from the CIS or the. 04 with CUDA GPU acceleration support for TensorFlow then this guide will hopefully help you get your machine learning environment up and running without a lot of trouble. For this example, we install miniconda to Windows and use the python. Check to make sure the driver was installed correctly with: $ nvidia-smi. At the time of writing this document, the latest stable driver version is 384. 48 as seen below, and the cards are two Tesla K40m. The nvidia-settings should be installed by default, incase you still need to manually install it $ sudo apt install nvidia-settings. After complete anaconda setup, create a new conda environment. 04 Server With Nvidia GPU. # activate your conda environment. In this post we will explain how to prepare Machine Learning / Deep Learning / Reinforcement Learning environment in Ubuntu (16. This is graphics reinvented. I want to use graphics card for my tensorflow and I have installed and re-installed again but tensorflow is not using GPU and I have also installed my Nvidia drivers but when I run nvidi-smi then a command is not found. If you want accurate usage statistics use nvidia-smi or modify the GPU setting in Task Manager to "CUDA usage". Nvidia Smi displays me my graphics card. This tutorial shows how to use TensorFlow with Horovod on a Deep Learning AMI with Conda. py > ls templates/ template. 加权随机采样算法--Weighted Sampling Algorithm; svn命令行常用操作. I've only tested this on Linux and Mac computers. tensorflow-gpuでGPUが認識されない問題 conda create -n ml tensorflow-gpu keras としたところ、GPUが使用できませんでした。 パッケージを確認してみます。 conda list -n ml | grep-e python -e cud -e tensorflow # python 3. conda create -n fastai source activate fastai. fuser -v /dev/nvidia* kill掉所有(连号的)僵尸进程. The nvidia-settings should be installed by default, incase you still need to manually install it $ sudo apt install nvidia-settings. それで、ちゃんとNVIDIAのホームページからvoltaに対応しているnvidia-384. 結果がこちら 少なくともこちらの記事(DeepLabCutのインストールについての備忘録)で最初にconda createで作成. Make sure that the latest NVIDIA driver is installed and running. TensorFlow is a very important Machine/Deep Learning framework and Ubuntu Linux is a great workstation platform for this type of work. 查看package信息 conda search numpy. See Selecting the Instance Type for DLAMI for more info. 04 machine with NVIDIA's new GTX 1080 Ti graphics card for use with CUDA-enabled machine learning libraries, e. I, for example, commanded Conda to install pytorch when the package is in fact named torch. 4 をインストールしていった。anacondaで仮想環境を作ってその上にpython3. nvidia-smi returns failed result on Cisco b200-m4. 04 with CUDA GPU acceleration support for TensorFlow then this guide will hopefully help you get your machine learning environment up and running without a lot of trouble. Stack Exchange network consists of 175 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers. Installing TensorFlow with GPU on Windows 10. If you have not done so already, download the Caffe2 source code from GitHub. 0 - Added new functions for NVML 3. , tensorflow-gpuを動かすようにする。 conda create -n tf python =3. If that's the case, wipe any remnants of NVIDIA drivers on your system, and install one NVIDIA driver of your choice. nvidia-smi. 1 && conda install -c fragcolor cuda10. I, for example, commanded Conda to install pytorch when the package is in fact named torch. 1 and CUDNN and added the new variables to the path. Is there a way to keep the file at local but still get the output from !nvidia-smi?. 26 CUDA Version: 10. In this blog post I will discuss how to get TensorFlow working on the AWS p2 instances, along with some tips about configurations and optimizations. Conda: it is an open source package management system and environment management system that runs on Windows, \user_path\Anaconda3\envs\fastai_v1\lib\site-packages no nvidia-smi is found. Are the GPUs available when you run nvidia-smi at a terminal? [DEPP LEARING VM AZUR] unable to Access to GPU through jupyter notebook. 0 Initialization (After all prerequisites have been installed). 删除package conda remove -n py27 numpy 安装cuda. I can't find nvidia-smi. In order to use TensorFlow on your workstation, there are a few assumptions and requirements. Use pip to install tensorflow-gpu. terminal 1: tmux new -s train conda activate keras time python train_alexnet. Posts about anaconda written by wolfchimneyrock. exe deviceQuery. sudo dpkg -i nvidia-driver-local-repo-ubuntu1604-387. sudo apt-get install nvidia-driver-415 ppa源极其慢,没有镜像,只能等。由于要很久,防止显卡过热,这一步我是把显卡拆下来的(记得关机断电操作)。 安装完成后重启就可以正常识别显卡了。 终端输入. Reboot and Check if Drivers are installed correctly by running: nvidia-smi. 내가 사용한 버전은 Ubuntu 데스크탑 16. Conda install will automatically decide the compatible version of CUDA and cuDNN to be installed. CUDA is NVIDIA's relatively mature API for data parallel GPU computing. GPU+ Machine. 이전에 설치 과정에서 로그인 무한반복의 쓴 맛을 본 관계로 그냥 겁이 났었다. Configuring a deep learning rig is half the battle when getting started with computer vision and deep learning. cd C:\Program Files\NVIDIA Corporation\NVSMI. conda update --all. 7 with pytorch and everything it needs. conda create -n theano. If you have not done so already, download the Caffe2 source code from GitHub. Tensorflow currently supports CUDA versions 9. Second, run nvidia-smi in your terminal window to determine the number of available GPUs on your DLAMI. Build and run a sample code like vectorAdd or bandwidthTest. One key benefit of installing TensorFlow using conda rather than pip is a result of the conda package management system. exe Starting. nvidia-smi. exe Starting. the Nvidia card of my system:enter image description here. Nvidia RTX 2080 TI; NVIDIA-SMI 430. We have discussed about GPU computing as minimally needed theoretical background. I want use pytorch with gpu. You can set the environment according to your preference. 54も選べそうだが、しばらく様子見とする. 1 python -c "import tensorflow as tf; print(tf. conda install -q -y --force -c richli fftw=3. 0 - Added new functions for NVML 5. Your scratch directory has a quota capping the total size and number of files you may store in it. Best practices when writing playbooks will follow the trend of using command unless the shell module is explicitly required. 1 -c pytorch instead of installing it with the installtion snippet generated for me by pytorch website: conda install pytorch torchvision cudatoolkit=10. cfg alexnet. 04 搭建深度學習開發環境 RTX 2080 + CUDA 10. I installed Cuda by downloading the. 0 Existing code can be reused. This can be downloaded here. I'd like to know whether this situation is normal. This script is based one `nvidia-smi`, but it can show complete process names and commands. - Added nvidia_smi. curl https://conda. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. Install OmniSci just use conda install XXX where XXX is the name of the package you want to install. 深度学习(deep learning)是机器学习的分支,是一种试图使用包含复杂结构或由多重非线性变换构成的多个处理层对数据进行高层抽象的算法。. Releases all unoccupied cached memory currently held by the caching allocator so that those can be used in other GPU application and visible in nvidia-smi. watch -n 1 nvidia-smi top tips about training large model. Activate the correct conda environment,. Check by typing \nvidia-smi" and \nvcc {version" in command line terminal. In addition, your CPU is used to prepare the data for the GPU. x virtual environment and vice versa. 参考:Tensorflow GPU Installation Made Easy: Use conda instead of pip. Open a terminal (Anaconda Prompt in Windows) and activate the Conda environment where you installed idtracker. We will also be installing CUDA 10. python 패키지는 전부그런듯. GPU Support of Tensorflow on Linux Mint/Ubuntu 14. see cuda-toolkit for cuda driver version. 0-base nvidia-smi. ARC3 and GPUs Tool Trade off Program ming NVIDIA CUDA (C dialect, nvcc compiler) Existing codes need to be rewritten. cz - 16 cores / 32 threads, 256GB RAM, 500GB SSD, 8 x NVIDIA GTX 1080Ti taylor. And "nvidia-smi" can help you check availble GPU device. 12 GPU version. 更新package conda update -n py27 numpy. conda install -c anaconda tensorflow-gpu(anacondaリポジトリから) conda install jupyter matplotlib scipy h5py scikit-learn keras. 0 (depending on the Tensorflow version being used). 这里也可以通过添加apt或者yum sourcelist的方式进行安装,但是我没有root权限,而且update容易引起docker重启,如果不是实验室的个人环境不推荐这么做,防止破坏别人正在运行的程序(之前公司一个小伙子就是在阿里云上进行了yum update,结果导致公司部分业务停了一个上午)。. Things change in Ubuntu 18. NVIDIA System Management Interface The NVIDIA System Management Interface (nvidia-smi) is a command line utility, based on top of the NVIDIA Management Library (NVML) , intended to aid in the management and monitoring of NVIDIA GPU devices. Assumptions. This is a machines I’ve dedicated for experimentation. We used out of the box TensorFlow with conda; Investigating further with NVIDIA's System Management Interface (SMI), we found out that GPU did run at a its highest turbo speed: 1. conda create -n fastai source activate fastai. 启动安装程序,一直按q,输入accept接受条款 输入n不安装nvidia图像驱动,之前已经. 56) is now complete. Windows support is at an experimental stage: it should work fine but we haven't thoroughly tested it. nvidia-smi NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. pip install tensorflow-gpu Open python and try to run a simple tensorflow function. Question Idea network consists of 175 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers. • In-order to launch Jupyter Notebook use the port number provided to you along with your login credentials. Install OmniSci just use conda install XXX where XXX is the name of the package you want to install. I'm using Ubuntu 18. 创建conda虚拟环境+安装python3. Due to technical limitations, the conda package does not support GPUs at the moment. 0 or above with an up-to-data Nvidia driver. Instructions drawn from this guide to installing CUDA and CUDNN. Use it only if you know for sure the problem is not a general graphics board issue (in that case use the tag: graphics). This guide will walk through building and installing TensorFlow in a Ubuntu 16. 04 LTS with CUDA 8 and a NVIDIA TITAN X (Pascal) GPU, but it should work for Ubuntu Desktop 16. [[email protected] ~]$ conda create -n cuda python=3. The library is based on research into deep learning best practices undertaken at fast. 以下の場所に、deviceQuery. 深度学习(deep learning)是机器学习的分支,是一种试图使用包含复杂结构或由多重非线性变换构成的多个处理层对数据进行高层抽象的算法。. $ nvidia-smi. As a portal member with the Advanced Notebooks privilege, open a new notebook. 1 -c pytorch instead of installing it with the installtion snippet generated for me by pytorch website: conda install pytorch torchvision cudatoolkit=10. You can also configure docker to use the nvidia runtime by default by using the following. 0\bin\win64\Debug コマンドプロンプトから、上記を実行すると、CUDAが認識しているGPUデバイスの情報が出力される > deviceQuery. The fastai library simplifies training fast and accurate neural nets using modern best practices. See Selecting the Instance Type for DLAMI for more info. speed,utilization. Toggle navigation. In this note, I detail a step-by-step instruction I followed to setup software on NVIDIA-based "Deep Learning Box". TechPowerUp makes a pretty popular GPU monitoring tool called GPU-Z which is a bit more friendly to use. # This script outputs relevant system environment info # Run it with `python collect_env. However, it is wise to use GPU with compute capability 3. 9 GHz, which. 6) conda create -n fastai python=3. 90はv100をサポートしてない。 NVIDIA DRIVERS Linux x64 (AMD64/EM64T) Display Driver. 1 Accessing GPU Resources To access a GPU. Actually, in the official repository, a build script named build_windows. Stack Exchange network consists of 175 Q&A communities including Stack Overflow, the largest, most trusted online community for developers to learn, share their knowledge, and build their careers. The command tries to infer this information from parsing the output so it may not be exact. exe utility even after installing Tesla C1060 WHQL driver (file size 106 MB) for Windows 7. Provides a Python interface to GPU management and monitoring functions. Please mark any answers that fixed your problems so others can find the solutions. During usage, both CPU and GPU usage spike somewhat linearly. Nvidia Docker: While the Nvidia drivers for the GPUs make sure that the GPUs can be used when logged into the machine, Nvidia-docker makes sure that the GPUs are visible from inside the containers. Solution using your image, but without Docker. Assumptions. conda installation, installing development versions, etc. See the fastai website to get started. $ nvidia-smi. $ nvidia-smi. If you are wanting to setup a workstation using Ubuntu 18. deb from Nvidia. conda install numpy scipy mkl Test that it was loaded correctly after the reboot, executing the command nvidia-smi from the command line. We only support the installation of the requirements through conda. This includes getting the images into a format that the GPU can work with easily and quickly as well as any augmentation that you have selected. 7 with pytorch and everything it needs. conda installation, installing development versions, etc. The command nvidia-smi will write to the standard out and then the container will simply linger around forever because of the sleep infinity command. You can browse for and follow blogs, read recent entries, see what others are viewing or recommending, and request your own blog. Nó có thể downgrade các gói đã cài (chỉ trong môi trường của Conda) để đảm bảo cài đặt thành công. 4 をインストールしていった。anacondaで仮想環境を作ってその上にpython3. Installing TensorFlow with GPU on Windows 10. And I would verify the CUDA install using the instructions in the linux install guide provided by NVIDIA. Install CUDA toolkit and cuDNN library by following the official instructions from Nvidia. 5 with Cuda 9. 0, man that was fun, lots of googling with multiple visits to ubuntu and nvidia forums and reading up on several blog posts and stackoverflow articles and almost at the end of the long day am running cuda 9. 0和内核驱动一起安装上,而且确实安装了名为nvidia-kmod的包,但重启后执行nvidia-smi. 1 along with the GPU version of tensorflow 1. sh on the Tegra device. The idea is to package all the necessary tools for image processing. Or follow these steps to solve the issue: nvidia-smi. I had exactly the same issue. 深度学习(deep learning)是机器学习的分支,是一种试图使用包含复杂结构或由多重非线性变换构成的多个处理层对数据进行高层抽象的算法。. conda create -n envname python=2. If running nvidia-smi shows no GPU, then you need to spin up another DLAMI using an instance with one or more GPUs. I'm using Ubuntu 18. I completed the process successfully once, and was able to run nvidia-smi to see my graphics card usage. In this post we will explain how to prepare Machine Learning / Deep Learning / Reinforcement Learning environment in Ubuntu (16. CUDA is NVIDIA’s relatively mature API for data parallel GPU computing. Here is another my documentation you may refer to Medium – 17 May 19 Deeplabcut GPU env using Conda. source activate tf でそれをアクティベート. 0 (depending on the Tensorflow version being used). If running nvidia-smi shows no GPU, then you need to spin up another DLAMI using an instance with one or more GPUs. Run the resulting container, and it will execute the %runscript $ singularity run example. However, these functions can be used to allow profiling to be performed selectively on specific portions of the code. command prompt automatically closes after open in my pc. Installing and Updating GTX 1080 Ti Drivers / CUDA on Ubuntu April 29, 2017 machine learning, python, nvidia, CUDA, drivers, tensorflow. This is graphics reinvented. The driver version is 367. The NVIDIA CUDA® Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. Guild uses NVIDIA System Management Interface (nvidia‑smi) on GPU accelerated systems to collect GPU metrics. If CNTK lists a GPU, make sure no other CNTK process is using it (check nvidia-smi, under C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.