MAGMA Forum

MAGMA is thread-safe when different routines are called from the same process (e.g., through different threads). Using MAGMA from different processes should have even less potential problems - at least non that we have heard of. If there is a problem it is usually the GPU setting and you may have to...

Hello, Just checking if you found a solution to this proble. MAGMA uses a default initial guess of 0, i.e., the RHS. I checked this particular matrix with matlab and looks like the rank of the Krylov space starting with ones remain 2, so the space ramains limited and non of the Krylov methods conver...

Thanks for reporting on this so that we know what to expect!
That's too bad interfaces were changed in CUDA 11 that brake the back compatibility.
We will have to put some more ifs in the next release.

Yes, one solution is to move back to CUDA 10.3 until we update support for CUDA 11. Alternatively, to solve the last problem, you can specify the GPU in your make.inc file. If you specify GPU_TARGET = Turing the makefile will not add other options. In this case Kepler. Looks like CUDA 11 is deprecia...

The LAPACK used has underscores added to the names, like sgerqf_, so when referenced by MAGMA it also has to be with underscores added, otherwise you get that sgerqf (and the other LAPACK routines used) is not defined. To fix this, you can add options -DADD_ to a few places in the make.inc file, nam...

The magma_d_spmv is a wrapper that is pre-tuned to call the "fastest" implementation (based on architecture, sizes, formats, etc.). You can take a look at it in file magma_d_blaswrapper.cpp. So, if the matrix is in CSR format for example, we currently call cuSparse.

We haven't considered adding it. MAGMA has CSR5 storage and kernel optimized for it.

The main focus of MAGMA is dense linear algebra, providing LAPACK for GPUs. MAGMA has also a sparse iterative solvers component but current support there is for single GPU only. There is a hipMAGMA port of MAGMA for AMD GPUs that ports CUDA to HIP.

You may also want to try the 2-stage reduction algorithms, e.g. ./testing_dsyevdx_2stage -JV --niter 2 -n 7000 These are much faster especially for the large sizes that you target. Maybe also using multiple GPUs would help (adding "--ngpu 4" option). Also, you can try with ESSL. There is make.inc ex...

Some of these errors seem to be large and inconsistent. This is what I get on one of our systems with V100 and Intel CPU. [tomov@a04 testing]$ ./testing_dsyevd -JV --niter 5 -c -l -n 7000 % MAGMA 2.5.2 svn compiled for CUDA capability >= 7.0, 32-bit magma_int_t, 64-bit pointer. % CUDA runtime 9020, ...