Was it seriously intended to have all CUDA Toolkit 3.0 binaries under ‘bin’ as i386 (which is good, considering a lot of OSX machines are still 32bit) and open64 binaries as x86_64 (and thus unable to run on anything but 64bit macs)? or was this just a massive oversight?
I could understand if they were universal binaries, but mixing and matching 32bit/64bit binaries in what is supposed to be a 32bit release of the CUDA Toolkit is … just bad.
Any ETA on when this will be fixed? We’re unable to upgrade to 3.0 until we can compile on 32bit and 64bit macs (due to a number of our osx build servers being 32bit).