Mix of Discrete and Internal GPU: Performance? Best Practice?

i always thought that when i have a machine with 2 Graphics Cards; for example a strong dedicated one and a weaker onboard-card that when i use both, the performance will go down? at least this is what i experienced in the past.

but lately i tried with a new machine (just a quick test) and i didn’t see any performance loss. can you shine some light into this? what is the truth?

This is not the case on laptops, for the most part you do not have to worry about it and the system takes care of it. Laptops do not run the program on both GPUs at the same time, the system switches to one GPU or another. This can introduce all sorts of other issues, just search for ‘Optimus’ in this forum and you’ll see the problems some machines run into, like selecting the wrong GPU, V-sync issues between monitors, or the Screen Grab TOP not working because it tries to access the wrong GPU etc. There are also issues unique to systems using AMD iGPUs and Nvidia discrete GPUs (for which the solution is elsewhere in the forum) but it has to do with startup or crashing, not performance once working.

When looking at Windows Task Manager performance, you will sometimes see both GPUs ‘working’ hard. This is usually because many laptop designs have the integrated GPU directly connected to the laptop monitor, and therefore must pipe the graphics output from the discrete GPU through the integrated GPU to display it on screen. The application is only running on the discrete GPU in these cases and the integrated is simply handing it off.

Thx Ben, but it still does’t fully answer my question.

Imagine I have RTX 3090 and i use all 4 4K outputs and I would like to connect my control monitor on the onboard card. How does ist behave? Bottleneck or not? May feeling says yes, but as I wrote I am not sure anymore.

Oh you are not using a laptop? I didn’t think it was possible to run the output from the CPU’s integrated graphics when the discrete card it in use on a workstation. I have never seen this configuration in use and we haven’t tested for it. Its certainly not a ‘best practice’ configuration as you might get random behavior, each TouchDesigner process will be assigned to one of the GPUs and multi-GPU setups are only recommended with Quadros and using GPUaffinity for separate processes to be assigned to separate GPUs.

Perhaps someone else has tried and can comment. If you try it and have some results to share, please post your full specs and findings.

that’s exactly what i thought too … thx