Is it a terrible bug of cula spare?

[dreamseeker]

Hi,
I want to solve a linear equation like A*x=b where A is symmetrical but not positive definite matrix stored in a spare matrix format. B is a vector that just has one element b(300)=2.3 and x is the vector that I want. Data format is double precision. So I choose the solver Gmres to solve my equation and I get the result I expect. when I only change b(300) to 17.8 , the solution x would be the same as before but multiple by 17.8/2.3. However , the actual result I get from the function Gmres is very different with what I expect. The version is cula S6 and cuda 6.0 and my code is based on example code iterativeSystemSolveFortran.f90. Dose anyone can explain how could this happen ?