01-25-2013, 03:04 PM
I remember similar situation with NV.
When they released CUDA 3.2(which required 260.xx), there was a HUGE speed loss, but oclhc was using OpenCL for NV then.
I was using the old version of oclhc and old drivers(25x.xx).
Everything was running fast(no goddamn 30% speed drops), but in the end, I had to upgrade, since so many features have been implemented, yet I was unable to use them.
When they released CUDA 3.2(which required 260.xx), there was a HUGE speed loss, but oclhc was using OpenCL for NV then.
I was using the old version of oclhc and old drivers(25x.xx).
Everything was running fast(no goddamn 30% speed drops), but in the end, I had to upgrade, since so many features have been implemented, yet I was unable to use them.