Timing PLASMA 2.4.5

Open forum for general discussions relating to PLASMA.

Timing PLASMA 2.4.5

Postby eyguacel » Wed Feb 22, 2012 2:42 pm

Hi all
I have installed PLASMA in a "true" 16 core machine (2xAMD6128HE). All tests are ok but timing results are a bit strange

PLASMA dgemm
./time_dgemm_tile --n_range=30000:30000:1000 --atun
# N NRHS threads seconds Gflop/s Deviation
30000 1 16 0.437 4.12 0.00

Scalapack and hpl results are more consistent

Scalapack LU (xdlu test)
TIME M N NB NRHS NBRHS P Q LU Time Sol Time MFLOPS CHECK
---- ----- ----- --- ---- ----- ---- ---- -------- -------- -------- ------
WALL 30000 30000 128 1 1 4 4 220.94 0.36 81345.8 PASSED

HPL
================================================================================
T/V N NB P Q Time Gflops
--------------------------------------------------------------------------------
WR11C2R4 30000 128 4 4 180.82 9.956e+01
--------------------------------------------------------------------------------
||Ax-b||_oo/(eps*(||A||_oo*||x||_oo+||b||_oo)*N)= 0.0038872 ...... PASSED
================================================================================

¿How should I interpret PLASMA timings?
TIA
eyguacel
 
Posts: 1
Joined: Fri Feb 17, 2012 7:12 am

Re: Timing PLASMA 2.4.5

Postby mateo70 » Fri Mar 02, 2012 1:09 pm

If you run the dgemm timing, you need to specify K which is set by default to NRHS, meaning to 1.
If you add the option --nrhs=30000, you should have better results.

Mathieu
mateo70
 
Posts: 94
Joined: Fri May 07, 2010 3:48 pm


Return to User discussion

Who is online

Users browsing this forum: No registered users and 0 guests

cron