Offload figure graphic computations from NVIDIA to local intel GPU

I have an application where resource-intensive computations are being performed on an NVIDIA GPU, and many figures containing status information, data plots and 3D graphics are being updated as processed data is retrieved from the GPU. Currently, Matlab is using the installed NVIDIA GPU for both the computations and figure displays. Is it possible to perform the figure graphics updates on the local Intel graphics GPU rather than on the NVIDIA GPU to prevent resource competition?

Answer 1

[Walter Roberson]

https://superuser.com/questions/1657952/disabling-nvidia-gpu-for-routine-tasks shows how to change your graphics card preference to the Intel GPU

Beyond that: some models of NVIDIA cards (higher-end ones only!! ) can be configured in TCC (Tesla Compute Cluster) mode on Windows, which disables use as a graphics card and can produce higher efficiency. The models I find listed as generally supporting TCC are some Tesla class, a number of Quadro class, and some Titan models. GeForce models do not typically support TCC.

I am having difficulty finding any list of current TCC support.

[Tim]

Strangely, this does not seem to work; after opening up the NVIDIA control panel and switching Matlab's preferred graphics processor to "Integrated graphics" (on this laptop, an Intel UHD 630 GPU), I still see the NVIDIA 3D load being the only responder when I push the graphics with a ~5million patch 3D figure. The integrated GPU processor doesn't respond at all. This is still true after a program reset and computer restart. Re-opening NVIDIA's control panel verifies that the preferred graphics processor is still "Integrated graphics."

Unfortunately, I am just using an RTX2070 so I don't have TCC mode.

[Tim]

Also, "opengl info" still shows the renderer as the RTX2070