I implemented VBO buffers (both ELEMENT_ARRAY and ARRAY) for my terrain system, and was shocked that on the GeForce FX 5900 and GeForce FX 5200 systems that we have – there was no speedup. I even had to put a breakpoint to make sure my VBO render path was getting called, and not the fallback path .
Taking it over to a RADEON 9600 system however, showed that the performance was greatly increased, to the point where it was about 4x faster than the nVidia systems.
Now… does anyone know if there’s a problem with the nVidia implementation of the VBO extension? I have the latest Detonator drivers. The terrain system is extremely vertex hungry (which my belief in was boosted, based on the results I’m seeing on the RADEON boards), but the nVidia cards are beyond slow (some patches are xferring about 7MB of vertex data per frame without the VBO support).
What does VTune tell you about where your time is spent? What is the code doing where the time is spent? (Yes, this may require disassembling the driver DLLs)
I implemented VBO buffers (both ELEMENT_ARRAY and ARRAY)
There is a known issue with nVidia cards in terms of accessing element array data. You should make sure that your element array buffer is a different buffer object from your actual vertex data.
I haven’t plugged VTune in yet, since it seems (so far) a nVidia vs. ATi problem. Some other demo programs on the net that use VBO seem to have the same performance issues as my code, so it seems to be a VBO implementation problem. I’ll get a newer copy of VTune (the one we have is kinda old) and give it a shot.
And I do have two separate buffer objects [didn’t strike me that they could be the same object, that’s interesting] BTW. Seems strange that the nVidia interface is so much slower, but it seems like that’s what everyone’s experiencing with them.
I don’t like the idea of going and using the nVidia VAR extension instead, since VBOs seem so much cleaner. Darn.
VBO are correctly implemented on nvidia drivers 5x, adn are not on older versions. I had a perf boost from 90 fps to 240 fps on a 150K polys scene, simply by upgrading drivers 44 to 52.
Yes indeed, the latest Detonator drivers are much faster, but on my NV35 (FX 5900) it still runs quite a bit slower than the RADEON 9600. On the order of half the speed.
Seems like a driver issue more than anything else (I hope).
.