CUDA DOUBLE PRECISION  [SOLVED]

option, parallelism,...

Moderators: fgoudreault, mcote

Forum rules
Please have a look at ~abinit/doc/config/build-config.ac in the source package for detailed and up-to-date information about the configuration of Abinit 8 builds.
For a video explanation on how to build Abinit 7.x for Linux, please go to: http://www.youtube.com/watch?v=DppLQ-KQA68.
IMPORTANT: when an answer solves your problem, please check the little green V-like button on its upper-right corner to accept it.
Locked
mancini
Posts: 40
Joined: Mon Nov 09, 2009 10:56 am

CUDA DOUBLE PRECISION

Post by mancini » Sat May 14, 2011 6:41 pm

Hello to all,
At this moment I'm doing some test on BUDA machine which have 4 tesla GPU (thanks Jean Michel and Xavier).
I was having some doubt about my cuda code because nothing works in DOUBLE PRECISION.
This was strange because, my last test in DP passed all RECURSION tests.

After many tests, trying to debug DP, I remember that my old tests DP was obtained with the compilation option 1.3:
(NVCCFLAGS+= --arch sm_13)
This option permits to use the "compute capability" 1.3 which includes double.
Otherwise all is demonted to float causing some problemes.

I added manually this option and all work fine: tests are successful with GPU

Yann, Could be possible to do that automatically when cuda DP is chosen?

Regards,
Marco
Marco Mancini

LUTH, Observatoire de Paris
5, place Jules Janssen
92190 MEUDON - FRANCE

User avatar
jbeuken
Posts: 365
Joined: Tue Aug 18, 2009 9:24 pm
Contact:

Re: CUDA DOUBLE PRECISION

Post by jbeuken » Sat May 14, 2011 8:15 pm

Hello Marco,

As suggested by Alain, it would be preferable to use in build system, the nomenclature "GPU compute capability" rather than accuracy ( SP or DP )
( http://www.geeks3d.com/20100606/gpu-com ... ive-table/ )

my 50 cents, 8-)

jmb
------
Jean-Michel Beuken
Computer Scientist

mancini
Posts: 40
Joined: Mon Nov 09, 2009 10:56 am

Re: CUDA DOUBLE PRECISION  [SOLVED]

Post by mancini » Sat May 14, 2011 11:27 pm

Jean-Michel,
I don't agree totaly with this point of view:
You can have SP with a compute capability 1.3,
but it seems that the inverse is not true.

Marco
Marco Mancini

LUTH, Observatoire de Paris
5, place Jules Janssen
92190 MEUDON - FRANCE

Locked