diff options
Diffstat (limited to 'devdocs/docker/compose%2Fgpu-support%2Findex.html')
| -rw-r--r-- | devdocs/docker/compose%2Fgpu-support%2Findex.html | 119 |
1 files changed, 119 insertions, 0 deletions
diff --git a/devdocs/docker/compose%2Fgpu-support%2Findex.html b/devdocs/docker/compose%2Fgpu-support%2Findex.html new file mode 100644 index 00000000..02d3394f --- /dev/null +++ b/devdocs/docker/compose%2Fgpu-support%2Findex.html @@ -0,0 +1,119 @@ +<h1>Enabling GPU access with Compose</h1> + +<p>Compose services can define GPU device reservations if the Docker host contains such devices and the Docker Daemon is set accordingly. For this, make sure to install the <a href="https://docs.docker.com/config/containers/resource_constraints/#gpu">prerequisites</a> if you have not already done so.</p> <p>The examples in the following sections focus specifically on providing service containers access to GPU devices with Docker Compose. You can use either <code class="language-plaintext highlighter-rouge">docker-compose</code> or <code class="language-plaintext highlighter-rouge">docker compose</code> commands.</p> <h3 id="use-of-service-runtime-property-from-compose-v23-format-legacy">Use of service <code class="language-plaintext highlighter-rouge">runtime</code> property from Compose v2.3 format (legacy)</h3> <p>Docker Compose v1.27.0+ switched to using the Compose Specification schema which is a combination of all properties from 2.x and 3.x versions. This re-enabled the use of service properties as <a href="../compose-file/compose-file-v2/index#runtime">runtime</a> to provide GPU access to service containers. However, this does not allow to have control over specific properties of the GPU devices.</p> <div class="highlight"><pre class="highlight" data-language="">services: + test: + image: nvidia/cuda:10.2-base + command: nvidia-smi + runtime: nvidia + +</pre></div> <h3 id="enabling-gpu-access-to-service-containers">Enabling GPU access to service containers</h3> <p>Docker Compose v1.28.0+ allows to define GPU reservations using the <a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#devices">device</a> structure defined in the Compose Specification. This provides more granular control over a GPU reservation as custom values can be set for the following device properties:</p> <ul> <li> +<a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#capabilities" target="_blank" rel="noopener" class="_">capabilities</a> - value specifies as a list of strings (eg. <code class="language-plaintext highlighter-rouge">capabilities: [gpu]</code>). You must set this field in the Compose file. Otherwise, it returns an error on service deployment.</li> <li> +<a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#count" target="_blank" rel="noopener" class="_">count</a> - value specified as an int or the value <code class="language-plaintext highlighter-rouge">all</code> representing the number of GPU devices that should be reserved ( providing the host holds that number of GPUs).</li> <li> +<a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#device_ids" target="_blank" rel="noopener" class="_">device_ids</a> - value specified as a list of strings representing GPU device IDs from the host. You can find the device ID in the output of <code class="language-plaintext highlighter-rouge">nvidia-smi</code> on the host.</li> <li> +<a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#driver" target="_blank" rel="noopener" class="_">driver</a> - value specified as a string (eg. <code class="language-plaintext highlighter-rouge">driver: 'nvidia'</code>)</li> <li> +<a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#options" target="_blank" rel="noopener" class="_">options</a> - key-value pairs representing driver specific options.</li> </ul> <blockquote> <p><strong>Note</strong></p> <p>You must set the <code class="language-plaintext highlighter-rouge">capabilities</code> field. Otherwise, it returns an error on service deployment.</p> <p><code class="language-plaintext highlighter-rouge">count</code> and <code class="language-plaintext highlighter-rouge">device_ids</code> are mutually exclusive. You must only define one field at a time.</p> </blockquote> <p>For more information on these properties, see the <code class="language-plaintext highlighter-rouge">deploy</code> section in the <a href="https://github.com/compose-spec/compose-spec/blob/master/deploy/#devices" target="_blank" rel="noopener" class="_">Compose Specification</a>.</p> <p>Example of a Compose file for running a service with access to 1 GPU device:</p> <div class="highlight"><pre class="highlight" data-language="">services: + test: + image: nvidia/cuda:10.2-base + command: nvidia-smi + deploy: + resources: + reservations: + devices: + - driver: nvidia + count: 1 + capabilities: [gpu] +</pre></div> <p>Run with Docker Compose:</p> <div class="highlight"><pre class="highlight" data-language="">$ docker-compose up +Creating network "gpu_default" with the default driver +Creating gpu_test_1 ... done +Attaching to gpu_test_1 +test_1 | +-----------------------------------------------------------------------------+ +test_1 | | NVIDIA-SMI 450.80.02 Driver Version: 450.80.02 CUDA Version: 11.1 | +test_1 | |-------------------------------+----------------------+----------------------+ +test_1 | | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | +test_1 | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | +test_1 | | | | MIG M. | +test_1 | |===============================+======================+======================| +test_1 | | 0 Tesla T4 On | 00000000:00:1E.0 Off | 0 | +test_1 | | N/A 23C P8 9W / 70W | 0MiB / 15109MiB | 0% Default | +test_1 | | | | N/A | +test_1 | +-------------------------------+----------------------+----------------------+ +test_1 | +test_1 | +-----------------------------------------------------------------------------+ +test_1 | | Processes: | +test_1 | | GPU GI CI PID Type Process name GPU Memory | +test_1 | | ID ID Usage | +test_1 | |=============================================================================| +test_1 | | No running processes found | +test_1 | +-----------------------------------------------------------------------------+ +gpu_test_1 exited with code 0 + +</pre></div> <p>If no <code class="language-plaintext highlighter-rouge">count</code> or <code class="language-plaintext highlighter-rouge">device_ids</code> are set, all GPUs available on the host are going to be used by default.</p> <div class="highlight"><pre class="highlight" data-language="">services: + test: + image: tensorflow/tensorflow:latest-gpu + command: python -c "import tensorflow as tf;tf.test.gpu_device_name()" + deploy: + resources: + reservations: + devices: + - capabilities: [gpu] +</pre></div> <div class="highlight"><pre class="highlight" data-language="">$ docker-compose up +Creating network "gpu_default" with the default driver +Creating gpu_test_1 ... done +Attaching to gpu_test_1 +test_1 | I tensorflow/stream_executor/platform/default/dso_loader.cc:48] Successfully opened dynamic library libcudart.so.10.1 +..... +test_1 | I tensorflow/core/common_runtime/gpu/gpu_device.cc:1402] +Created TensorFlow device (/device:GPU:0 with 13970 MB memory) -> physical GPU (device: 0, name: Tesla T4, pci bus id: 0000:00:1e.0, compute capability: 7.5) +test_1 | /device:GPU:0 +gpu_test_1 exited with code 0 +</pre></div> <p>On machines hosting multiple GPUs, <code class="language-plaintext highlighter-rouge">device_ids</code> field can be set to target specific GPU devices and <code class="language-plaintext highlighter-rouge">count</code> can be used to limit the number of GPU devices assigned to a service container. If <code class="language-plaintext highlighter-rouge">count</code> exceeds the number of available GPUs on the host, the deployment will error out.</p> <div class="highlight"><pre class="highlight" data-language="">$ nvidia-smi ++-----------------------------------------------------------------------------+ +| NVIDIA-SMI 450.80.02 Driver Version: 450.80.02 CUDA Version: 11.0 | +|-------------------------------+----------------------+----------------------+ +| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | +| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | +| | | MIG M. | +|===============================+======================+======================| +| 0 Tesla T4 On | 00000000:00:1B.0 Off | 0 | +| N/A 72C P8 12W / 70W | 0MiB / 15109MiB | 0% Default | +| | | N/A | ++-------------------------------+----------------------+----------------------+ +| 1 Tesla T4 On | 00000000:00:1C.0 Off | 0 | +| N/A 67C P8 11W / 70W | 0MiB / 15109MiB | 0% Default | +| | | N/A | ++-------------------------------+----------------------+----------------------+ +| 2 Tesla T4 On | 00000000:00:1D.0 Off | 0 | +| N/A 74C P8 12W / 70W | 0MiB / 15109MiB | 0% Default | +| | | N/A | ++-------------------------------+----------------------+----------------------+ +| 3 Tesla T4 On | 00000000:00:1E.0 Off | 0 | +| N/A 62C P8 11W / 70W | 0MiB / 15109MiB | 0% Default | +| | | N/A | ++-------------------------------+----------------------+----------------------+ +</pre></div> <p>To enable access only to GPU-0 and GPU-3 devices:</p> <div class="highlight"><pre class="highlight" data-language="">services: + test: + image: tensorflow/tensorflow:latest-gpu + command: python -c "import tensorflow as tf;tf.test.gpu_device_name()" + deploy: + resources: + reservations: + devices: + - driver: nvidia + device_ids: ['0', '3'] + capabilities: [gpu] + +</pre></div> <div class="highlight"><pre class="highlight" data-language="">$ docker-compose up +... +Created TensorFlow device (/device:GPU:0 with 13970 MB memory -> physical GPU (device: 0, name: Tesla T4, pci bus id: 0000:00:1b.0, compute capability: 7.5) +... +Created TensorFlow device (/device:GPU:1 with 13970 MB memory) -> physical GPU (device: 1, name: Tesla T4, pci bus id: 0000:00:1e.0, compute capability: 7.5) +... +gpu_test_1 exited with code 0 +</pre></div> +<p><a href="https://docs.docker.com/search/?q=documentation">documentation</a>, <a href="https://docs.docker.com/search/?q=docs">docs</a>, <a href="https://docs.docker.com/search/?q=docker">docker</a>, <a href="https://docs.docker.com/search/?q=compose">compose</a>, <a href="https://docs.docker.com/search/?q=GPU%20access">GPU access</a>, <a href="https://docs.docker.com/search/?q=NVIDIA">NVIDIA</a>, <a href="https://docs.docker.com/search/?q=samples">samples</a></p> +<div class="_attribution"> + <p class="_attribution-p"> + © 2019 Docker, Inc.<br>Licensed under the Apache License, Version 2.0.<br>Docker and the Docker logo are trademarks or registered trademarks of Docker, Inc. in the United States and/or other countries.<br>Docker, Inc. and other parties may also have trademark rights in other terms used herein.<br> + <a href="https://docs.docker.com/compose/gpu-support/" class="_attribution-link">https://docs.docker.com/compose/gpu-support/</a> + </p> +</div> |