Different behaviour for contact target velocity when using CPU/GPU solvers

Using PhysX 3.4 beta, I have found that modifying the target velocity for contact points yields different behaviour. Below is the behaviour with PhysX set to use the GPU processor:

Note that I am setting the target velocity to be orthogonal to both the contact position and the contact normal. The box is positioned at the origin. The behaviour shown above is what I would expect. However, when switching to use the CPU processor, the behaviour is different:

It’s as if all the contact points are receiving the same target velocity.

Thanks for reporting the issue. The fix should be included in the final 3.4 release


Just to clarify… Will the behaviour using the CPU solver be as in the first image (rather than the other way round)?

Yes, the CPU will work the same as the video you posted of the GPU behavior.

I can see that this has been fixed in the 3.4 release candidate. Thank you!