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


Postby dni » Mon Oct 03, 2011 6:03 am

I tested function sgesvd using program testing_sgesvd.cpp. The result seems to indicate there is not much improvement for GPU over the CPU version. I also tried the same function from CULA library, the speedup is very dramatic. I wonder if there is something wrong with this function in Magma?
Posts: 5
Joined: Tue Aug 09, 2011 3:42 am


Postby Stan Tomov » Wed Oct 05, 2011 4:27 pm

Thanks for this note. There are a few components of the gesvd that we have accelerated in MAGMA but don't use yet in the magma gesvd. Right now we only use the bi-diagonalization that is about 10x faster than MKL on multicore CPUs (depending on machine and matrix sizes). The other GPU-accelerated components will be added in MAGMA 1.1 that will be released for SC11.
Stan Tomov
Posts: 258
Joined: Fri Aug 21, 2009 10:39 pm


Postby fletchjp » Wed Oct 12, 2011 12:41 am

For those (like myself) not familiar with SC11 here is a link. I am looking forward to having MAGMA 1.1

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

Return to User discussion

Who is online

Users browsing this forum: Google [Bot] and 2 guests