MAGMA Forum

Is magma_zstedx even accelerated? It looks like a complete pass through to the host LAPACK.

Thanks Keitat for pointing out the reasons why MAGMA is not thread safe. It's worth noting that the MAGMA was probably "more thread safe" before the CUDA 4.0 changes to GPU binding. In an unrelated project, I was required to completely overhaul my multi-threaded code to account for the multi-GPU cha...

A developer can correct me if I'm wrong, but I don't think MAGMA is thread safe.

mgates3 wrote:This is because a tridiagonal solve has O(n) operations on O(n) memory, so is memory bandwidth limited.

This isn't entirely true. Check out NVIDIA's paper on cyclic-reduction algorithms (not in MAGMA).

I've found that almost every MAGMA BLAS kernel takes liberties with reading out of bounds GPU memory when reading into shared memory buffers. On some systems this is OK. However, on others (like yours) it causes errors. Typically this memory won't be accessed so it's not an obvious error, but an err...

MATLAB uses Intel's MKL which has a special algorithm called successive band reduction for the symmetric eigenvalue factorization step.

There was a lot of changes to multiGPU in CUDA 4.0 when it came out a few months ago... perhaps MAGMA hasn't updated to support that?

Is it possible to compile MAGMA for both Tesla and Fermi support in one library? Right now you must specify the generation in the Makefile. I have a pre-compiled application where I don't know what generation the end user will have. They might have a Fermi, Tesla, or even a G80. I know this is possi...

OK, thanks. That explains why the GPU performance is so low. Do you think these routines will be in RC4?

Is it also safe to assume the numbers from the paper are from only single-threaded LAPACK? That seems to be the conclusion we have found here.

So MAGMA's SSYMV and SGEMV aren't the ones from the paper?