Issues with multisample FBOs on nVidia cards.

I have implemented multisample frame buffer objects via the EXT_framebuffer_multisample and EXT_framebuffer_blit extensions. On my dev machine containing a nVidia geForce 8800GTX it works great, no error messages and I can antialias up to 16x.

On our target machine containing a Quadro 5600 however, I get no errors, but am hitting an apparent software path as the frame rate is about 0.1 fps, regardless of how many multisamples I am requesting ranging from 1 (min supported) to 16 (max reported support).

On another dev machine with a 8800gts, it runs fast at 2x, and slow at anything above. I do get MAX_MULTISAMPLES, and clamp to it, so the driver seems to be reporting support for 16x, but then kicking to software.

Any advice from you nVidia guys??

I’ve been working on an 8800GTS, and I haven’t seen what you’re describing. I can think of three possible issues, though:

  1. Try glHint(GL_MULTISAMPLE_FILTER_HINT_NV, GL_FASTEST). The driver might default to GL_NICEST.
  2. You’re using to much memory, and it’s being forced to swap to main RAM with higher sample counts
  3. You’re using an out-of-date driver on one or more of the systems.
  1. I’ll give that a try, thanks.
  2. There is more memory on the quadro than the geforce cards.
  3. No, I’m not, same drivers for the geforce cards, and it’s the latest, and the quadro driver is also as new as possible.

As of 8/8 there are new Quadro drivers that work just fine. So apparently it was a driver related issue. Thanks to all who read this post.