Problem with dsytrd and dsyevd

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)

Problem with dsytrd and dsyevd

Postby Abdelfattah » Wed Dec 14, 2011 4:31 am


I'm trying to run the MAGMA test for tridiagonalization (SYTRD) and symmetric eigen values (SYEVD) on double precision. I was running to get results up to 15000 matrix dimension. However, I had a seg fault after 8000 on dsyevd, and after 10000 on dsytrd. It seems this is a memory issue. But I do really extend the runs beyond 10000.

I'm using MAGMA 1.0.0-rc5 and running on Fermi C2070.

You comments are appreciated.

Posts: 1
Joined: Wed Dec 14, 2011 4:16 am

Re: Problem with dsytrd and dsyevd

Postby jeremiahpalmer » Thu Dec 15, 2011 1:29 pm

One remedy for the memory issue would be to use a multi-GPU sytrd. Does anyone on the magma team have an idea of when this type of feature will be ready for use?

Posts: 58
Joined: Fri Jan 28, 2011 12:46 pm

Re: Problem with dsytrd and dsyevd

Postby Stan Tomov » Wed Feb 08, 2012 3:15 am

For larger matrices you would have to use algorithms supporting multiGPUs. We developed them for the symmetric and general matrices cases and will release them in MAGMA 1.2.
Stan Tomov
Posts: 258
Joined: Fri Aug 21, 2009 10:39 pm

Re: Problem with dsytrd and dsyevd

Postby fletchjp » Fri Feb 10, 2012 12:27 pm

Do you have a features list and estimate of release date for 1.2?


Posts: 203
Joined: Mon Dec 27, 2010 7:29 pm

Return to User discussion

Who is online

Users browsing this forum: No registered users and 1 guest