Added GPU-Access to systemd-Service Definition (opensearch-project#10694)

TheBigK02 · kolchfa-aws · Konstantin Nebel · sumobrian · commit 42bbb6729411 · 2025-08-20T22:09:25.000-05:00
* Update parent title for install and configure benchmark pages (opensearch-project#10686) Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> Signed-off-by: Konstantin Nebel <konnebel@gmx.de> * Delete unused images from the site and correct image formatting (opensearch-project#10691) * Remove unused images Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> * Delete icons file and correct image formatting Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> * Heading formatting Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> --------- Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> Signed-off-by: Konstantin Nebel <konnebel@gmx.de> * Added GPU-Access for systemd --------- Signed-off-by: Konstantin Nebel <konnebel@gmx.de> * Update _ml-commons-plugin/gpu-acceleration.md Co-authored-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com> Signed-off-by: TheBigK02 <50010925+TheBigK02@users.noreply.github.com> * Update _ml-commons-plugin/gpu-acceleration.md Co-authored-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com> Signed-off-by: TheBigK02 <50010925+TheBigK02@users.noreply.github.com> --------- Signed-off-by: Fanit Kolchina <kolchfa@amazon.com> Signed-off-by: Konstantin Nebel <konnebel@gmx.de> Signed-off-by: TheBigK02 <50010925+TheBigK02@users.noreply.github.com> Co-authored-by: kolchfa-aws <105444904+kolchfa-aws@users.noreply.github.com> Co-authored-by: Konstantin Nebel <konnebel@gmx.de>
diff --git a/_ml-commons-plugin/gpu-acceleration.md b/_ml-commons-plugin/gpu-acceleration.md
@@ -67,6 +67,18 @@ else
 fi
 ```
 
+If you run OpenSearch natively (without Docker) using the packaged version of OpenSearch, `systemd` may block OpenSearch from accessing your GPU. To accelerate models, you need a working [CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit) installation and access to the NVIDIA device under `/dev`.
+
+To allow OpenSearch to use the GPU, update the `systemd` service by adding the following configuration:
+
+```ini
+systemctl edit opensearch.service
+
+[Service]
+DevicePolicy=auto
+``` 
+
+
 After verifying that `nvidia-uvm` exists under `/dev`, you can start OpenSearch inside your cluster. 
 
 ### Preparing AWS Inferentia ML node
