MultiGPU and Projection Mapping Problem

Hi all,
I am making a project with 9 projectors output to do some projection mapping with 4K resolution and hope FPS will up to above 45.

My hardware list:

  • Single PC system
  • CPU : AMD Ryzen TR PRO 3955WX (16 cores/32 threads】3.9G/280W/72M/7nm/PCIe4.0
  • Mother board : ASUS Pro WS WRX80E-SAGE SE WIFI
  • RAM : KLEVV CRAS XR RGB DDR4 3600 32GB(16Gx2)
  • GPU : NVIDIA RTX A4000(16G GDDR6 256bit/CUDA:6144/24.2cm) X 3
    • Also use NVIDIA Quadro Sync card to connect 3 GPU.

Now I try to use 3 TD toe file to process 4,3,2 projectors output separately.
For communication, I use Touch In(TOP) / Touch Out(TOP) operator to send my video content between each toe file. And also use this technique below to specify TD process that is connected to the particular GPU, but it seems only one GPU to do all of those things and FPS drop down to 15-20…

I am just stuck now, and don’t have any idea to figure out this problem…
Does anyone have some experiences with this problem?
How multiple GPU would work in single system?

Thank you very much! I will be vey grateful for your help!

Katherine

Hi @DPTW,

Are you starting your toe files with a batch script, or manually?

One important consideration here is that the bounds of an output can’t cross onto a display that’s connected to another card - if they do, then that data ends up being computed on the other GPU and your efficiency is lost. This is done by both using the batch script to start TouchDesigner, and by setting your output window COMP in your toe files to correctly map to the correct output(s).

I usually start in these situations by setting up a simple test file to make sure everything is working as expected, move into getting the actual files up and running correctly.

When you run your batch file does touch start on all of the correct displays?