Unable to do double precision calcs

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi all,
I am new to CUDA. I have a 4xTESLA machine with each tesla being major revision 1 and minor revision 3 ( I checked by compiling and running the example which lists all your devices {devicequery?}).

Anyway I cannot use double numbers at all. The device is converting them to floats silently. I tried using the following nvcc command line options:

-sm_13
–gpu=sm_13
-arch=sm_13

none of which have any effect. I really need double precision for the work I am doing, anyone please?

Many Thanks!

#2

Please post the output from deviceQuery, along with the driver that you’re using.

#3

Thanks for the reply!

My driver is 181.20 on 64-bit windows XP, see attached file for screenshot of the devicequery.

That brings up another question: how do you pipe this output to a text file when there is no exe file?
devices.rtf (3.92 MB)

#4

Why do you think the device is converting them to floats?

#5

I figured it out myself. For others knowledge, I am using Visual Studio 2005 with a third party custom rules setup I got of the net (CUDA_VS_x64_Wizard.1.0) and I was trying to add the -arch sm_13 to the additional options in the Cuda command line section of the property pages inside VS. Instead I had to change my custom build rules file (cuda.rules) which was added by the aforementioned rules wizard and add it there (as the first command line option). Save cuda.rules and restart VS and away we go!

Cheers