different results numpy.solve Vs. scikits.cuda.cula.culaDeviceDgesv() about scikit-cuda

[llvll0hsen]

any guess?

[lebedov]

I'm investigating.

[llvll0hsen]

Thanks for helping. I just realized for single precision case (culaDeviceSgesv) this error accrues on more 2000 samples.

[lebedov]

Not sure why this is occurring; it doesn't seem to be due to the Python bindings. I suggest that you inquire on the culatools forum.

[llvll0hsen]

I managed to implement it with pycula "gpu_gesv" which using the same cull function as scikits.cuda. I tried it with 7000 samples and it worked fine! so it's more likely to be my code fault. right?

[lebedov]

I would have to investigate further to determine whether that is the case.

[llvll0hsen]

Have you find anything?! are you getting the same error?

[llvll0hsen]

Have you found anything?Any idea?

[lebedov]

There is a bug in your code; for a system A*x = b, the nrhs parameter must be set to the number of columns in b. See http://pastebin.com/beFQNSNs for the corrected code.

[llvll0hsen]

Great.I wil check it tuesday morning when I come back to the lab. Thanks a lot