Cublas_getError() retrieved unknown error!

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi,

I’m new at CUDA programming and I’m trying to make a simple M*N matrix multiplication.
But after calling the function “cublasDgemm” , Cublas_getError retrieved me the error number equal to 8. I don’t know what 8 stands for so I can’t find why the matrix multiplication failed.

See here for the errors number:

http://javagl.de/jcuda/jcublas/doc/constant-values.html

Thanks for your help :)

#2

Lol it works perfectly with matrix of “float”.

Do I need to put more stuff if I want the matrix multiplication to work with matrix of “double”? (conversion?..)

#3

Are you running on 1.3 hardware? (GTX2x0) or Tesla C1060/S1070 Quadro FX4800/5800?

Older versions cannot do double precision (so you could convert your doubles to floats)

#4

I run on a GeForce 8800 GTS …
An old card so it certainly doesn’t support doubles…thx for the answer.
Are they other limitations for old cards? Where I can find them?

#5

There is a list in the programming guide about the abilities of the different compute capabilites.

You can find out which compute capability your card is by running devicequery from the SDK