Hello everyone!
I’ve asked this question in Maya general forum but didn’t get answer. So I decided to ask it here.
I’m wondering if Maya GPU cache has memory limitations. I have Quadro K2000 on board with 2GB of memory but when I try to import “heavy” scenes(15-20 million faces) I get only bounding box in the viewport.
It is meant to stream from the video memory, so, yes, it is VRAM bound.
Your VRAM will be limited by whatever needs drawing + aero or equivalent and all that jazz, so depending on resolution you will only have so much of it to spare.
Just a few frames at 20M faces (likely a lot more triangles than that) plus whatever attributes might be trying to travel it’s pretty easy to exhaust it.
Thank you, Raffaele!
What about second K2000 in SLI mode? Will this solve the problem? I mean 4GB VRAM, will be sufficient for this number of faces/triangles?
Nope.
In first place SLI isn’t supported by any mainstream DCC app, and even if it was, SLI doesn’t double the memory as data is mirrored across videocards, therefore you can get (potentially) close to twice the number crunching, but memory equivalent to just one card.
Your only option, assuming you absolutely need such monster polycount, is to upgrade the videocard to something with more RAM.
As for how much ram would be needed, I don’t know the footprint of the implementation nor how much past just the geo you are trying to move across, and finally what resolution you are working at on how many monitors.
Ok, understood. It is the bifrost mesh I’m trying to export. The polycount oscillates between 10-30m. Double monitor 1920x1080.
Thank you very much for such detailed explanation.
Double 1080p + Maya and a few apps around might take away some half a gig or so at the most.
10-30M polys though, about double that in tris if the mesher works out to mostly quads, well… that’s a fair chunk of space.
If you’re on windows you can download GPU-Observer (a free sidebar gadget) to get a pretty accurate real time report on what is going on with you vdieocard, then progressively pump stuff into the GPU cache to find the boundaries of your current hardware.