-
Thanks for the reply Clint,
I am afraid I still don't see how these figures fit! If the Floating point operations are not delaying us, then the FLOPS are limited by how fast we can stream matrix A in the processor (You have stated this as well in your reply).
** If we are talking about single floating point data (4bytes/element), then 3GB/s is equivalent to 3*10^9/4 elements per second...
2007-07-23 07:52:00 UTC in Automatically Tuned Linear Algebra Soft.
-
I am running atlas on a AMD Athlon Dual core 3800_ running @ 2GHz. The system has 2x 1GB PC3200 DDR DIMMs.
I am running a series of matrix-vector products on single precision elements and I am getting around 2.9 GFLOPS. When I run the STREAM benchmark I get around 3GB/s memory bandwidth.
The problem is that these 2 figures don't fit together. I believe that for a matrix-vector product The...
2007-07-13 07:50:44 UTC in Automatically Tuned Linear Algebra Soft.
-
elzein committed revision 56 to the My Book List SVN repository, changing 2 files.
2007-05-27 01:16:49 UTC in My Book List
-
elzein committed revision 55 to the My Book List SVN repository, changing 4 files.
2007-05-26 12:18:22 UTC in My Book List
-
elzein committed revision 54 to the My Book List SVN repository, changing 7 files.
2007-01-07 09:46:55 UTC in My Book List
-
elzein committed revision 53 to the My Book List SVN repository, changing 1 files.
2007-01-04 21:18:34 UTC in My Book List
-
elzein committed revision 52 to the My Book List SVN repository, changing 17 files.
2007-01-04 21:09:26 UTC in My Book List
-
elzein committed revision 51 to the My Book List SVN repository, changing 3 files.
2007-01-04 01:21:11 UTC in My Book List
-
elzein committed revision 50 to the My Book List SVN repository, changing 1 files.
2007-01-04 00:03:40 UTC in My Book List
-
elzein committed revision 49 to the My Book List SVN repository, changing 1 files.
2007-01-03 23:51:29 UTC in My Book List