OpenGL multisampling performance

Dear Lazyweb,

If you turn on multisampling in xscreensaver on OSX, like this:

defaults -currentHost write org.jwz.xscreensaver.Geodesic multiSample -bool true

then one or two things happen. The first is, everything looks much prettier because the polygon edges are anti-aliased. Yay. But the second is, sometimes, the frame rate drops from 30fps to like 5fps, and CPU utilization goes up to 95%. Boo.

I don't understand under what circumstances multisampling kills performance. I assume it's exhausting some resource and falling back to software rendering, though I don't know what that resource is or what else is contending for it. Also, I haven't seen this happen on my current machine. I last saw it on a Mac with a GeForce 8800 GS in it, which is something like a 2008-vintage graphics card.

So... I'd like to turn on multisampling by default, except on machines where that's likely to suck.

Perhaps the question I want to ask is, "does this machine's video card have gobs and gobs of RAM?" Or maybe it's just, "was this machine's video card manufactured after 2013?"

Do any of you know how to ask questions like that of an NSOpenGLContext, or using some other MacOS API?

Tags: , , , ,

One Response:

  1. cobbal says:

    Getting the VRAM seems to be fairly straightforward:

    If you want to check if you're getting the software renderer, it's

    [pixfmt getValues:&val forAttribute:NSOpenGLPFAAccelerated forVirtualScreen:...];

    My OpenGL is rusty though, and I don't have an old machine to test these on either.