lammps/gpu on cilindrical Mac Pro, openCL vs. cuda

Hello people,

Has anyone run gpu-accelerated lammps calculations, in particular with the eam interaction model, on the cylindrical Mac Pro?

Our group needs to buy some hardware for visualisation purposes and eam/gpu number crunching. For the former a Mac Pro would be a good though somewhat expensive machine. If the number crunching can be done on the same hardware (Mac Pros have two fairly good AMD cards in them), that would make it a very good solution that kills two birds with one stone. Would anyone have some data on lammps eam/gpu performance on the cilindrical Mac Pro? The system we'd like to run would be fairly large, millions to lower dozens of millions of atoms.

One drawback of the Mac Pro would be that the AMD cards in it won't do cuda. Would anyone have some performance data on how eam/gpu stacks up against eam/cuda for large systems?


I like OpenCL and want to see it succeed. However, the way things are
currently in early 2015, I would not recommend people purchase new
equipment for general scientific computation if it does not support

Unfortunately that includes all current Apple products.

If you have decided that you want invest in computational hardware,I
would suggest a PC running a mainstream desktop linux OS, 8 core
intell CPU, and an Nvidia TitanX(s).


More ranting:
   If you must have a Mac for running Keynote, or other mac-only
software, purchase a macbook air, or an iOS device for that purpose.
(Furthermore, installing open-source libraries and software on a Mac
can be harder than it is on a windows PC. I never liked the Macports,
although perhaps "homebrew" is better. I doubt either system is as
good as "apt-get".)
    Perhaps LAMMPS supports AMD graphics cards, however I have seen a
lot of other scientific and visualization software that (for a long
time) only supports CUDA/Nvidia. And even for LAMMPS, my own
experience was that the OpenCL-compatible "GPU" package for LAMMPS ran
faster on Nvidia cards than AMD cards (but this was probably because
the Nvidia cards I was using at the time were more expensive models.
As Axel and others said, check the benchmarks.)
    I have no experience regarding the new Intel "Phi" coprocessors.
(But I am concerned by the fact that Intel has not made the compiler
you need to take advantage of the Phi freely available.)
   For relatively short simulations, it might be cheaper (or faster)
to rent computer time from a cloud-service (EG: google, amazon/ec2,
penguin, rescale, ...), and avoid purchasing a new computer. (I might
also consider applying for time on XSEDE.)