about VASP compiling on opteron

hakuna · Post by **hakuna** » Fri Aug 11, 2006 3:27 am

hi, andri
I know from Dr. Greame that you have a opteron cluster, perhaps you have the experinece compiling VASP on this platforms.
I have a problem in using VASP on opteron cluster when compiling the code using INTEL compiler.

the compiled vasp can go through small works(short time and small memory), but the big jobs always gradually use up the memory and then the SWAP and at last suicide. this occures when using LIBGOTO or mkl.

perhaps you know how to deal with it.
and a MAKEFILE is surely appreciated.

thanks

andri · Post by **andri** » Fri Aug 11, 2006 12:22 pm

Actually the Opteron cluster is Graeme's personal cluster (and his research group's) and I have not used it that much. I think the only things (VASP included) that I have complied there have been using Graeme's makefiles and pgf90. Maybe he will know some more about using ifort with amd processors. But it sounds like you have a memory leak somewhere in your executable.

Post by **graeme** » Fri Aug 11, 2006 3:12 pm

That's very puzzling. I am using the Portland Group compilers on the Opteron cluster, but I did try the intel compiler at some point as well. It turned out to run slower, so I didn't use it. I also can't seem to find the makefile that I used. I don't remember any memory issues, but it also could be that I just didn't notice in my timing runs.

I remember that Andri has seen many problems with the intel compiler, and that they seem to get fixed (or introduced) as new versions come out. Try the latest compiler, and start with no optimization. If that works, it is likely a problem with the compiler.

Also, is this a problem with any vasp run, or only when you use our code? If we are introducing a leak, or a problem with the intel compiler, I would definitely like to track it down.

hakuna · Post by **hakuna** » Sat Aug 12, 2006 2:18 am

[quote="andri"]Actually the Opteron cluster is Graeme's personal cluster (and his research group's) and I have not used it that much. I think the only things (VASP included) that I have complied there have been using Graeme's makefiles and pgf90. Maybe he will know some more about using ifort with amd processors. But it sounds like you have a memory leak somewhere in your executable.[/quote]

thanks
and
sorry for my misunderstanding that you are in Greame's group.

hakuna · Post by **hakuna** » Sat Aug 12, 2006 2:30 am

[quote="graeme"]That's very puzzling. I am using the Portland Group compilers on the Opteron cluster, but I did try the intel compiler at some point as well. It turned out to run slower, so I didn't use it. I also can't seem to find the makefile that I used. I don't remember any memory issues, but it also could be that I just didn't notice in my timing runs.

I remember that Andri has seen many problems with the intel compiler, and that they seem to get fixed (or introduced) as new versions come out. Try the latest compiler, and start with no optimization. If that works, it is likely a problem with the compiler.

Also, is this a problem with any vasp run, or only when you use our code? If we are introducing a leak, or a problem with the intel compiler, I would definitely like to track it down.[/quote]

thanks, Greame
the problem is not invoked from your code, i think it may be the problem of ifort or mkl or libgoto when using on opteon cluster. there is no probelm when compiling vasp on intel machine(PC) using ifort and libgoto.

it seems that some report says the ifor have problem when using memroy lager than 2G, i am not sure.

Post by **graeme** » Sat Aug 12, 2006 3:24 am

Ok, good luck with it. I've linked to both the mkl and goto libraries without problems on the Opterons. I don't think that the compiler could cause a problem with these, assuming you are linking to prebuilt binary libraries.