Hi tdma~
I'm looking over the system details that you have posted, and I see that you are running clAmdBlas tuning on an Nvidia software stack on Nvidia hardware. Our OpenCL libraries should be platform agnostic, but I must confess that we don't have the same setup that you are running on in our testing lab.
clAmdBlasTune is fairly complicated code, stressing both the underlying software and hardware layers to find 'optimal' kernels. It does not suprise me greatly that you found issues on the Nvidia platform. At this point, I can only say that the errors that you see are probably real, but i can't guess at this point where in the software stack the problem is. We will take a look at increasing test coverage on Nvidia platforms for future releases.
Thank you for reporting this.