I am in the process of upgrading a compute node with a pair of K10 cards from CUDA 5.5 to 6.0. The problem is that, after the upgrade LAMMPS and Gromacs, two CUDA applications that I compiled under 5.5 stop working, and the error:
NVRM: API mismatch: the client has the version 319.37, but
NVRM: this kernel module has the version 331.62. Please
NVRM: make sure that this kernel module and all NVIDIA driver
NVRM: components have the same version.
NVRM: nvidia_frontend_ioctl: minor 255, module->ioctl failed, error -22
appears in /var/log/messages.
The nodes run RHEL 6.3, and the same binaries work well on another CUDA 6 system where I have access to (RHEL 6.4). I have tried the upgrade in several ways, including any possible combination of upgrading or fresh installation, using the .run CUDA package or through the .rpm method, installing manually the driver (./NVIDIA-Linux-x86_64-331.62.run)… Downgrading to 5.5 always makes the applications to work again.
Things that work well after the upgrade include Matlab, compiling and running any CUDA program that I have tested so far, and running the old 5.5 CUDA Samples.
Any help would be really welcome, I am completely stuck.
Thank you in advance,