[AMBER] Building Amber16 CPU with Intel Compiler and GPU with GNU Compiler vs. MPI

From: Ryan Novosielski <novosirj.rutgers.edu>
Date: Thu, 12 Apr 2018 17:14:20 +0000

Good afternoon,

I have read that the suggestion is to build the GPU version with GNU compilers and the CPU version with whatever compilers give you the best performance (with some suggestion that that would be the Intel Parallel Stuidio), since on the GPU version, optimization isn’t as important/handled by NVCC, and because that’s what the developers do. I had previously compiled the whole thing with the Intel compilers, and that seemed to work, but I’m reluctant to do something not recommended by the developers. That said, I more recently tried to compile the whole thing with the Intel compilers and ran into a version problem — it was saying that the version of NVCC I had couldn’t work with newer Intel compilers (though I am quite sure that one of the combinations I was trying was a repeat of one I’d used earlier — perhaps the AmberTools17 changes or that -lcuda that slipped in were the cause — I have to investigate further).

My question is what is the recommendation for what to do about MPI, and subsequently, the modules system (which is out of scope here, but someone may have solved the problem). What I’d done so far is compiled everything with the same Intel compiler, including the MVAPICH2 stack, and this had worked. If I use the GNU compiler for the GPU portions, I suppose I’ll have to use MVAPICH2 compiled with the GNU compiler as well, which I suppose them complicates hierarchical modules/prevents one from using the MPI-enbabled CPU and GPU versions together, if that’s something anyone ever needs to do.

At present, it seems I have a build that works that is using Intel 18.0.2 for the CPU portions, Intel MPI 18.0.2 for the MPI stack, GNU 4.8.5 for the GPU portions, and CUDA 8.0.

Does anyone who’s messed with this much have a recommendation — other than just do the whole thing with GNU compilers, which I’d think would reduce performance at least somewhat, but more importantly take away the CPU auto-dispatching that the Intel compiler is doing to support our heterogenous environment.

Thanks in advance.

--
____
|| \\UTGERS,  	 |---------------------------*O*---------------------------
||_// the State	 |         Ryan Novosielski - novosirj.rutgers.edu
|| \\ University | Sr. Technologist - 973/972.0922 (2x0922) ~*~ RBHS Campus
||  \\    of NJ	 | Office of Advanced Research Computing - MSB C630, Newark
     `'



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber

Received on Thu Apr 12 2018 - 10:30:02 PDT
Custom Search