What version of B40 that has stable consecutive reduction

[dev]

Hello,

I do this:


reduction_enactor.template Reduce(*d_problem_storage,
totalRigidParticles, p_num_compacted, gpu_reduce_num_compacted,
reduction_op, equality_op, 0);
mRigidBodyBuffers->Get(BufferTotalForce)-
>SetPtr<float_vec>(d_problem_storage->d_values[1]); //Set the pointer
to the result
cutilCheckMsgAndSync("Rigid Body total force reduce error.");//crash
here and get CUDA out of memory error

And when I run this I get CUDA out of memory error. I used
cudaMemGetInfo and I am sure I only allocated few megabytes out of my
1GB card, so I believe it is some kind of invalid memory location read/
write.
Weirdly enough the crash does not appear in release build, but I get
sometimes weird results. Also when running a test console application
(no graphics) the problem does not appear even in debug mode.

I checked out the latest version (problem in console and graphics
application) and also the r728 version (problem only in graphics
application).

Which is the most stable version for the consecutive reduction?

Thanks for advance.