Hello, I have have a working installation of TF2.1 under Ubuntu (details below)
$ nvidia-smi # secondary monitor connected to my notebook
Mon Jan 20 15:40:48 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.33.01 Driver Version: 440.33.01 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce MX150 On | 00000000:01:00.0 Off | N/A |
| N/A 63C P0 N/A / N/A | 1883MiB / 2002MiB | 8% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1109 G /usr/lib/xorg/Xorg 28MiB |
| 0 1309 G /usr/bin/gnome-shell 46MiB |
| 0 1547 G /usr/lib/xorg/Xorg 196MiB |
| 0 1746 G /usr/bin/gnome-shell 193MiB |
| 0 4042 G /usr/lib/firefox/firefox 3MiB |
| 0 7253 G ...uest-channel-token=10215247264308201648 21MiB |
| 0 29634 C ...nv/versions/3.7.6/envs/dl/bin/python3.7 1381MiB |
+-----------------------------------------------------------------------------+
(the last 2 lines refer to computing: jupyter and python, and take only 70% of the GPU)
[b]I would like to be able to reserve the nVIDIA GPU (GeForce 150 MX) to tensorflow, and leave graphics (xorg, gnome, etc.) to the Intel integrated device, and ideally to be able to switch the behaviour without rebooting or relogging.
I expect that I can create a suitable application profile in nvidia-settings, but I am afraid to mess things up. Is there a link to specific tutorial or doc with examples? I prefer to avoid gross mistakes that require reinstalling everything.[/b]
NB. Any other advice to solve the need, even not using nvidia-settings, is welcome
Update: I found https://devtalk.nvidia.com/default/topic/991849/-solved-run-cuda-on-dedicated-nvidia-gpu-while-connecting-monitors-to-intel-hd-graphics-is-this-possible-/ but it seems to deal with desktop hw, rather than notebooks, and is now >2 years old
BTW, I might have already spoiled something, as I get the following error:
$nvidia-settings
(nvidia-settings:30724): GLib-GObject-CRITICAL **: 15:46:46.524: g_object_unref: assertion 'G_IS_OBJECT (object)' failed
GPU at BusId 0x1 doesn't have a supported video decoder
If it matters, some configuration details and comments:
Asus Vivobook S510UN (secureboot disabled in bios)
Ubuntu 18.04.3 (fresh install)
nvidia-driver-440 (from CUDA nvidia-repo)
CUDA10.1, CUDNN7.6.5, TENSORRT6.0.1.5 (from nvidia web)
python3.7.6 (with pyenv)
tensorflow2.1.0, etc… (with pip in dedicated virtualenv)