Tesla c2070 card with production bundle of CUDA 6.5 on ubuntu 14.04 yields a strange error message when I try to run sample code. I used the RUN package to install drivers, cuda and sample files. The sample code compiled, but exits during runtime with an error. Below is an example:
richman@Copacabana:~/CUDA/NVIDIA_CUDA-6.5_Samples/0_Simple/simpleAtomicIntrinsics$ ./simpleAtomicIntrinsics
simpleAtomicIntrinsics starting…
modprobe: ERROR: …/libkmod/libkmod-module.c:809 kmod_module_insert_module() could not find module by name=‘nvidia_331_uvm’
modprobe: ERROR: could not insert ‘nvidia_331_uvm’: Function not implemented
CUDA error at …/…/common/inc/helper_cuda.h:1032 code=30(cudaErrorUnknown) “cudaGetDeviceCount(&device_count)”
I even tested to install the module with the result that first error message dissapeared.
Either nvidia_uvm module did not get installed (and is required for CUDA programs to run), or it has failed compilation (most likely given you used the run file to install drivers). You’ll have to go over the installer logs and see why it’s failing to compile… it’s possible it might be due to a patch that is now unnecessary that the installer attempts to apply.
Ok, I didn’t look into this for a long while so sorry for not keeping anyone interested posted.
Anywho:
The run and repo for 14.04 of ubuntu did not work. I used the latest drivers from the run or the releases for ubuntu. I tried both 5.5 and 6.5 versions of CUDA and it was simply not working and I was very meticulous about having cleared the entire system of nvidia software before reinstalling a new combination of software products.