I believe that the GTX 285 card is the currently best available on the Mac
platforms, at least in regards to CUDA calculations:
http://eshop.macsales.com/item/EGVA/01GP31080TR/
Cheers,
Thomas
On Fri, May 14, 2010 at 2:58 PM, Jason Swails <jason.swails.gmail.com>wrote:
> Hello,
>
> On Fri, May 14, 2010 at 4:16 PM, Steve Constable
> <sjconstable.uwaterloo.ca>wrote:
>
> > Hello,
> >
> > I am currently trying to install Amber 11 with nVidia CUDA support on my
> > Mac OS X 10.6 workstation. When it comes time to build CUDA support, the
> > linking stage fails because nVidia only provides i386 libraries, and
> amber
> > is built for the x86_64 architecture in 10.6. I tried to hack my own
> > configure script that would build Amber for the i386 architecture, but I
> > again experienced linker problems (netCDF refused to export symbols
> > correctly, and pmemd.cuda could not load the CUDA dynamic libraries
> > correctly.) I'm including my customized configure script for reference,
> and
> > I am using the GCC 4.5 suite from hpc.sourceforge.net<
> > http://hpc.sourceforge.net/>.
> >
>
> I spent a couple weeks trying to get it to work awhile ago (prior to
> release). It won't. I got as far as forcing the .cpp code to be compiled
> with g++ (got rid of some errors), and adding the -m32 flag to the gnu
> compilers to force a 32-bit build (since you're right, mac nvcc is 32-bit
> only), but I never got it working.
>
> Another thing worth trying is to install GCC 4.3 or earlier using macports
> or fink. GCC 4.5 won't work with the current nvcc (check google, this will
> confirm). However, I gave up on this since I had errors building GCC 4.3
> with macports (probably because I made macports gcc 4.4 my default
> compilers
> and macports was expecting the xcode version).
>
> Macs also don't ship with the best cards for doing calculations on (last I
> checked you couldn't customize a mac with a video card having more than 512
> MB of dedicated video RAM).
>
> If you do have any success with my above-mentioned tips, please post back
> since it would be worth knowing
>
> (If I recall, I think I got a non-working binary compiled once, but I'm not
> sure :)).
>
> All the best,
> Jason
>
>
> >
> > Thank You
> > -------------------------------------
> > Stephen J. Constable
> > University of Waterloo
> > Waterloo, Ontario, N2L 3G1
> > sjconstable.uwaterloo.ca<mailto:sjconstable.uwaterloo.ca>
> > (519) 888-4567 x34099
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> >
>
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Graduate Student
> 352-392-4032
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri May 14 2010 - 15:30:05 PDT
