Re: [AMBER] Amber11: CUDA-enabled version does not work

From: Mohd Farid Ismail <>
Date: Fri, 16 Jan 2015 10:53:24 +0800

   Did you try CUDA toolkit 4.x?

   16.01.2015, 10:23, "Khuong Truong Gia" <>:

     Hi Jason,
     Thank you very much for your sharing, you're very kind.
     About installation, I've already applied all bug fixes. But it still
     doesn't work. So, it's time to retire Amber11.
     Thanks again,
     On Thu, Jan 15, 2015 at 9:11 PM, Jason Swails <[1]>

     On Thu, 2015-01-15 at 17:24 +0900, Khuong Truong Gia wrote:

     Hi David,
     This is result when I run "":
     Using default GPU_ID = -1
     Using default PREC_MODEL = SPDP
     make[1]: Entering directory `/opt/amber11/test'
     cd cuda && make -k test.pmemd.cuda GPU_ID=-1 PREC_MODEL=SPDP
     make[2]: Entering directory `/opt/amber11/test/cuda'
     Running CUDA Implicit solvent tests.
     Precision Model = SPDP
     GPU_ID = -1
     cd trpcage/ && ./Run_md_trpcage -1 SPDP netcdf.mod
     diffing trpcage_md.out.GPU_SPDP with trpcage_md.out
     cd myoglobin/ && ./Run_md_myoglobin -1 SPDP netcdf.mod
     diffing myoglobin_md.out.GPU_SPDP with myoglobin_md.out
     *.....(all Implicit solvent tests PASSED),Then, it stood still (no update
     for long time) at line "cd 4096wat/ && ./Run.pure_wat -1 SPDP


     cd gb_ala3/ && ./Run.irest1_ntt1_igb1_ntc2 -1 SPDP netcdf.mod
     diffing irest1_ntt1_igb1_ntc2.out.GPU_SPDP with irest1_ntt1_igb1_ntc2.out
     Running CUDA Explicit solvent tests.
     Precision Model = SPDP
     GPU_ID = -1
     cd 4096wat/ && ./Run.pure_wat -1 SPDP netcdf.mod
     Is that my installation's mistake?

     It is hard to tell. Your symptoms look a lot like problems that were
     addressed by bugfix.12 for Amber 11 (see
     [2] If you have not already applied all
     of the bug fixes, you need to. If you have and it still doesn't work,
     you probably need to upgrade to Amber 14 to use pmemd.cuda.
     As Dave pointed out, Amber 11 is 5 years old at this point and
     represents pmemd.cuda in its infancy. Amber 11 was the first release
     that featured a production-ready GPU code. Support for that version
     ended not too long after Amber 12 was released (almost 3 years ago now).
     Since that time, GPU hardware has changed dramatically (Amber 11 was
     released alongside the Tesla architecture, like the Tesla 1060 and 1070
     cards and the GTX 2xx series). NVidia has gone through several
     different generations of architectures (and CUDA toolkits) since then --
     through Fermi, Kepler, and now on to Maxwell. Each architecture
     necessitated changes to pmemd.cuda's internals in order to take better
     advantage of the new hardware. As a result, Amber 11 won't run (or
     won't run well) on Kepler or Maxwell cards. Furthermore, numerous bugs
     have been found and fixed since Amber 11 has been released, and those
     fixes will most certainly not be ported back to Amber 11. Performance
     has also improved *dramatically* since the days of Amber 11.
     Long story short, it is probably time to retire Amber 11 if you are
     having problems with it. You are unlikely to find anybody on this list
     that can help out that has used Amber 11 within the last 3 to 5 years. I
     don't think Amber 11 is something we actively support anymore.
     Hope this helps,
     Jason M. Swails
     Rutgers University
     Postdoctoral Researcher
     AMBER mailing list

     AMBER mailing list


   Mohd Farid Ismail



AMBER mailing list
Received on Thu Jan 15 2015 - 19:00:02 PST
Custom Search