CUDA 5.0 + CULA R15 + Kepler = slow!!!

[itzik]

I have tried to build an application using CULA dense R15 and CUDA 5.0 - it runs fine and produces the correct results on a Kepler board (GeForce 690) but the performance is about 10x SLOWER than the same code linked with CULA R14 and CUDA 4.2 running on a Fermi board (GeForce 580) - any ideas?

[john]

Hello, we'll need to have you provide more details, such as the routine(s) being used, sizes of matrices, etc.