AMD vs Intel, diffrent collision detction results.

omaremad · Post by **omaremad** » Sat Jul 07, 2007 11:00 am

I just started using bullet, i compiled bullet using gcc on windows using both intel and amd cpus. I even forced it to genrate intel specific instrucions.

In my game on amd, collision detection on rigid bodies is normal with sliding being smooth.
On intel cpus the physics simulation is fine, but during a collision the frame rate drops significantly and the object just passes through the other mesh.

In both cases i have used simple shape vs trimesh or concave meshes. Amd behaves normally while intel has no colldetect (there doesnt seem to be a collision response).

The values all seem to be tweaked properly since the physics is perfectly fine on amds. I tested diffrent gravities from 1 to 90.

Anyone experienced these descrepancies before?

bone · Post by **bone** » Mon Jul 09, 2007 2:45 pm

What happens if you copy the executable made on one machine and run it on the other?

I recently encountered a difference in behavior (not related to Bullet) on two Intel machines, most likely due to some compiler option. On one machine the results appeared to be calculated in single-precision, vs. the other machine doing it in double-precision. I know, for one, that enabling SSE in the compiler options can affect this (that's using VC8, not gcc).

Dirk Gregorius · Post by **Dirk Gregorius** » Mon Jul 09, 2007 5:20 pm

Yeap, I also had issues with SSE and VS2005. The SP1 resolved those issues though...

Erwin Coumans · Post by **Erwin Coumans** » Wed Jul 11, 2007 3:29 am

I'm very suprised on the issue, perhaps some compiler preprocessor definition is missing?

Bullet source code should be working cross platform using all compilers, so if there is an issue it would be good to find out how to reproduce. Can you help with this?

What build system are you using? Which version of gcc? Mingw?

Thanks,
Erwin