Bug in NVidia drivers of Geforce FX59xx series in pBuffer depthtest

After asking for help (pBuffer pixel problems) in this forum two times I found the reason: it’s an NVidia driver bug and it took me a lot of hours to detect it. To spare you from this ordeal I’m posting some related infos for interested guys (sample application with VisualStudio project available):

Bug appearance:
I detected a bug in my application while rendering in a Pbuffer and then reading the Pbuffer content for further rendering. The Pbuffer containts some “missing pixels” in the drawn GL_QUAD primitive, always near the lower right or upper left corner. At these pixels the clearColor is visible. This effect is reproducable in the simple_pixel sample application of the NVidia SDK 7.0 (with some modifications).

Bug conditions:
These obscure pixels occur when:

  • GL_DEPTH_TEST is on
  • pixelformat supports multisampling
  • viewport, orthovolume and GL_QUAD primitive are sized identically
  • special sizes (e.g.400230, 200230, …)
    …but not all sizes: For 720*576 everything seems to be okay.
  • the vertex order of the GL_QUAD primitive is counterclockwise and
    starts with upper left vertex (when starting with lower left vertex it’s okay. ).

I know, its hard to believe, but it’s a fact!

Hardware / driver:
The bug occurres on:
GeForce FX5900XT, driver 5303
GeForce FX5900XT, driver 5672
GeForce FX5950Ultra, driver 5303

The bug does NOT occurre on:
GeForce FX5700, driver 5672
GeForce FX5700Ultra, driver 5672
GeForce 4 Ti4200, driver 5303

In other applications the bug also occurred with tesselated polygons.

I’m still looking for a workaround. Any ideas are highly welcome!


If you upload your program somewhere i can test it on the latest registered developer drivers.

It appears that the problem is still there and i would recommend that you email the problem to nvidia.

It’s already posted to NVidia - without answer until now.