On Mon, Apr 25, 2011 at 9:06 AM, Eugene Kozlov <[email protected]>wrote:

Hello,

To answer any kind of question about performance, we need the full output of -log_summary. Matt

I am trying to test PETSc capability of solving sparse linear systems in parallel with GPU.

I compiled and tried to run example src/ksp/ksp/examples/tutorials/ex2.c, which can be executed in parallel.

In this example matrix and vectors created using VecSetFromOptions() and MatSetFromOptions().

According to the page http://www.mcs.anl.gov/petsc/petsc-2/features/gpus.html , I execute the program with keys

-vec_type mpicusp -mat_type mpiaijcusp

in parallel on the different number of GPUs. Full command:

cleo-submit -np 1 ex2 -ksp_type cg -vec_type mpicusp -mat_type mpiaijcusp -m 800 -n 800

Where 'cleo-submit' is a batch manager utility.

I tested the program on 1, 2 and 3 GPUs. As a result I have output as (for 3 GPUs):

Norm of error 0.00012322 iterations 448 Norm of error 0.00012322 iterations 448 Norm of error 0.00012322 iterations 448

and run times: 30, 40 and 46 seconds respectively.

What can be a cause of these results?

-- Best regards, Eugene

-- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener