How to install for emulation only? I don't have an nvidia graphics card

Greetings,

I’m trying to set up a build environment where I cannot guarantee that all the machines will have the nvidia card.

The nvidia driver seems to provide the libcuda.so. But I cannot install the nvidia graphics driver on machines which do not have the nvidia card.

libcuda.so is needed by at least the NVIDIA_CUDA_SDK/projects/matrixMulDrv project.

How can I get this library onto those systems which do not have nvidia card?

Thanks,
Bill

Applications like matrixMulDrv use the driver API and cannot run in emulation mode.

You can build all of the SDK apps that do not require the CUDA driver by adding the -k option to your ‘make’ command.

I think I have the same issue as described in the first post of this thread. I need to install the driver kit on a machine with no GPU so that users can do builds. Runs will then be made on other machines with GPUs.

There appear to be no advanced option to the driver install to cleanly install the libraries etc. Have I missed an option?

Thanks

You can extract the driver package’s contents with the -x option, and then grab libcuda.so* from within $DRIVER_DIR/usr/lib

True. And that’s what I’ve done but I was hoping for an sure fire, automated, supported way :)