'compute_10' and 'sm_10' architectures are depreciated

Hello. I’m getting this warning with cuda/6.0 (e.g. with the MersenneTwister example from the cuda SDK):

nvcc -c ./MersenneTwister_kernel.cu

nvcc warning : The ‘compute_10’ and ‘sm_10’ architectures are deprecated, and may be removed in a future release.

What does this mean, and is it something I should be worried about?

Currently Loaded Modulefiles:

  1. cuda/6.0 2) pgi/14.4

Rob.

What does this mean, and is it something I should be worried about?

It means in with a future release of CUDA you will no longer be able to target NVIDIA devices with compute capability 1.0. My guess it wouldn’t effect you since 1.0 devices are quite old and pre-date double precision support which was introduced with CC 1.3.

By default, we compile for multiple CC devices in a single binary but have stopped including all CC 1.x targets. You can still specifically target CC 1.0 via the “-Mcuda=cc10”, or “-ta=tesla:cc10” flags. However, once CUDA stops support for CC 1.0, we will as well.

This Wikipedia article gives a nice list of devices at the various CC, including 1.0. Are you using any of the 1.0 devices? http://en.wikipedia.org/wiki/CUDA

Note that you can still use your older devices, but would need to build binaries with older tool sets and any 1.0 specific tool set or device issues would not be addressed.

  • Mat