faq.md

@@ -194,10 +194,14 @@ myjob:
 
 The shared GitLab runners do not support GPU workloads. If your project requires GPU (e.g., for training ML models), you must configure your own runner on a machine with GPU access, cf [Installing your own specific runners](https://gitlab.inria.fr/siteadmin/doc/-/wikis/faq#installing-your-own-specific-runners).
 
-We recommend that you use light installations, for example with this command tried and tested by Pascal Carrivain (without the default installation of the profiler and NVIDIA Nsight Graphics):
+We recommend that you use light installations, for example with this command tried and tested by Pascal Carrivain :
 ```
 apt-get -y install --no-install-recommends cuda-compiler-12-8 cuda-runtime-12-8 cuda-profiler-api-12-8 libcurand-12-8 libcurand-dev-12-8
 ```
+instead of complete installation, which includes the profiler and NVIDIA Nsight Graphics :
+```
+apt-get -y install cuda-toolkit-12-8
+```
 
 🔧 Options for GPU-Enabled Runners
 1. Grid'5000 (Public Infrastructure)