Issue with Large Vertex Buffers in linux...

Hi all, I did a quick search and did not see this issue yet so I thought I would ask.

The setup:
I have a large scalar dataset (500x500x100 cells or 25 million linear elements). I run two simple kernels, the first performs calculations over this data, and the second takes the output from the first kernel and transforms the results into three space via writing spatial data (25x3 = 75 million floats) to a 300MB VBO.

Under Windows these kernels run fine, with no issues reserving and registering the VBO. Rendering the VBO also occurs w/o incidence or issues.

Under linux, however, I get an “out of memory error” in the very beginning of the program initialization stage when I attempt to register the large VBO with CUDA (via a call to cudaGLRegisterBufferObject). Note that this occurs before any device memory has been cudaMalloc’d yet so the card should have no problem.

So the quick (and perhaps self answering) question is, are there known issues with large, contiguous, VBO under linux?

HW: 8800GTX 768MB memory
SW: Windows XP latest SP, Linux Kubuntu 7.04

0.8.1 for both OSs
Linux Driver:

Thanks for your thoughts,

Use Linux driver 1.0-9751 that is required to run CUDA.


Thanks Peter, I will give that a shot.