Page 1 of 1

SVD accuracy (single vs double precision)

PostPosted: Fri Aug 05, 2011 2:32 pm
by felipe
Hi,

i'm a noob regarding to CUDA/cula and I'm trying to test the gesvd function with a 12x9 matrix. Is it normal to have a little error using single precision? The orthogonal and reconstruction tests are close but they are not exact. If I use double precision, the error would be less? or am I not considering something else?

Thanks in advance.
Regards.

Re: SVD accuracy (single vs double precision)

PostPosted: Mon Aug 08, 2011 8:18 am
by john
I'll need a little more information from you here. Can you quantify how inexact and your method for measuring the error? Floating point arithmetic is basically never "exact" and results between different platforms will basically always show some sort of difference. The question is how much.

Re: SVD accuracy (single vs double precision)

PostPosted: Mon Aug 08, 2011 12:48 pm
by felipe
Hi john,

I realized that I've not considered something. The program that printed the matrices of the SVD was truncating the decimals after the comma (I did not write the program), so obviously the result was not exact when the original matrix was reconstructed via A = U * S * VT.

Sorry and thanks for your time.
Regards.

Re: SVD accuracy (single vs double precision)

PostPosted: Thu Oct 02, 2014 12:46 pm
by softbird
I prefer to use 215/35R17 Hankook Tires for my cruze as this seems to give the best riding comfort and the great advantage is that it suits varied sizes of alloys. This is indeed one of the best option for cruze as discussed with the pete's team.