I’m experiencing problems when upgrading to the NVidia driver 260.24 that comes with Cuda3.2 under 64bit Fedora. A program that I have works fine with the previous drivers (256.40) yet it fails (unchanged) under the new drivers. I can’t possibly post my program for inspection but I’ve found something that doesn’t work with the new drivers. When compiling and running oclMatrixMul in the SDK everything works fine when BLOCK_SIZE is 16 (defined in matrixMul.h) but it fails if you change that to something else. I’ve tried BLOCK_SIZE=8 or BLOCK_SIZE=18 (both work fine under the previous driver) and it fails with the new driver. Is that something that people have noticed? I’m using cuda3.2 and the NVIDIA_GPU_Computing_SDK_3.2.
Thanks in advance for any help,