malexander, on 07 May 2012 - 12:12 PM, said:
One thing I've discovered about the Nvidia drivers for OpenGL/CL is that OpenCL buffers do not seem to evict OpenGL objects from VRAM. So, if you've filled up your VRAM with OpenGL data by using lots of heavy geometry, lots of large textures, and/or lots of framebuffer (high AA levels, dual monitors, more than one Houdini or 3D app session running), then OpenCL allocations will fail. This is where a second graphics card would come in handy, as you could run your sim on the secondary card and not have its VRAM cluttered up with graphics data. Or you could just have 1 card with oodles of RAM (3-6GB).
Note that this doesn't seem to be an issue on AMD cards.
Hi Mark, this is really enlightening. I was under the impression that the benefits of having 2 graphics cards is only realized with the Tesla + Quadro 6000 Maximus setup (thank you Nvidia marketing).
Are you saying that if I have say, a Quadro 4000 and a Quadro 2000 installed, this setup will allow me to dedicate one these cards just for OpenCL and the other for the OpenGL viewport?
Are there environment variables or the like to allow the user to pick which card for OCL/OGL?
EDIT: Nevermind, I just found your other post about HOUDINI_OCL_DEVICENUMBER
I'll give it a try as soon as I get my hands on the hardware.
Edited by melazoma, 04 July 2012 - 06:02 PM.