Now that sounds like they arbitrarily changed the MPI configuration between
Ubuntu 10.x and 11.x. I've stuck with 10.04 because of the abomination
called Unity... This likely has nothing to do with the GPU and everything
to do with the default location of mpi.h and how to correctly map to it.
On Mon, Feb 27, 2012 at 11:24 AM, Mettu, Ramgopal <rmettu.tulane.edu> wrote:
> Thanks for the clarification about the CUDA versions. To summarize, I can
> successfully compile and test the serial and parallel code for AmberTools
> and Amber. It is only the CUDA version that is failing. Moreover, from the
> errors (see earlier in this thread) I inferred that there was a problem
> with code having to with the GPU. Perhaps it is a more basic issue than
> that? Have any other Ubuntu (11.1) users experienced similar issues while
> trying to compile the parallel (CUDA) version?
>
> Ram
>
> On Feb 27, 2012, at 10:48 AM, Ross Walker wrote:
>
> > AMBER works fine with the latest drivers and should also work fine with
> the 4.1 compilers although it is about 10% slower than the 4.0 compiler. So
> probably best to stick to the 4.0 compiler for now. Note the driver /
> compiler version has no bearing on parallel for now so I am confused about
> your serial vs parallel comment.
> >
> > As for driver version. There is no such thing as the "version 4.1
> driver" so I have no idea where people are getting this confusing
> terminology from.
> >
> > All the best
> > Ross
> >
> >
> >
> > On Feb 27, 2012, at 7:37, "Mettu, Ramgopal" <rmettu.tulane.edu> wrote:
> >
> >> All:
> >> I just wanted to follow up on my initial question about CUDA
> compatibility. Is it the case that AMBER does not yet work with the latest
> CUDA drivers? While I can still use the serial version, one of my main
> goals is to leverage the GPUs I have available, so I'd greatly appreciate
> any information about compatibility from the developers. Thanks!
> >>
> >> Ram
> >>
> >> On Feb 24, 2012, at 11:16 AM, Scott Le Grand wrote:
> >>
> >>> CUDA 4.1 and AMBER should be compatible, but there's a 10% slowdown
> >>> currently with it due to a bug in nvcc.
> >>>
> >>> Stick to 4.0 if you can for now...
> >>>
> >>>
> >>> On Fri, Feb 24, 2012 at 8:36 AM, Mettu, Ramgopal <rmettu.tulane.edu>
> wrote:
> >>>
> >>>> To be on the safe side, I downloaded Amber and AmberTools again and
> >>>> rebuilt everything, incorporating all bug fixes, as noted below. I
> get the
> >>>> same errors as I had mentioned initially - can the developers comment
> on
> >>>> CUDA 4.1 compatibility? Best,
> >>>>
> >>>> Ram
> >>>>
> >>>> On Feb 24, 2012, at 8:24 AM, Ismail, Mohd F. wrote:
> >>>>
> >>>>> I don't think CUDA 4.1 is supported yet.
> >>>>>
> >>>>>
> >>>>> *******************************
> >>>>> Mohd Farid Ismail
> >>>>> Graduate Student
> >>>>> Dept. of Chemistry/Biochemistry
> >>>>> University of Oklahoma
> >>>>> Norman 73019
> >>>>>
> >>>>> ________________________________________
> >>>>> From: Mettu, Ramgopal [rmettu.tulane.edu]
> >>>>> Sent: Thursday, February 23, 2012 10:45 PM
> >>>>> To: AMBER Mailing List
> >>>>> Subject: Re: [AMBER] CUDA version for Amber 11?
> >>>>>
> >>>>> Hi Ross,
> >>>>> Yes, I forgot to mention that I applied all the bug fixes you mention
> >>>> below -- any thoughts on what's happening?
> >>>>>
> >>>>> On Feb 23, 2012, at 10:37 PM, Ross Walker wrote:
> >>>>>
> >>>>>> Hi Ram,
> >>>>>>
> >>>>>>> As for the bug fixes, I decided to start from scratch and install a
> >>>>>>> fresh copy of AmberTools and Amber. I was able again to correctly
> >>>>>>> install the serial and parallel versions without errors (except
> for a
> >>>>>>> few discrepancies in the test scripts). Unfortunately I get a new
> set
> >>>>>>> of errors when trying to compile the CUDA binaries, that look like
> >>>>>>> this:
> >>>>>>
> >>>>>> But did you take that fresh copy and apply all the bugfixes for
> AMBER
> >>>> and
> >>>>>> AMBERTools from the website? - The bugfixes are required for
> >>>> compatibility
> >>>>>> beyond cuda 3.2. The procedure is:
> >>>>>>
> >>>>>> 1) tar xvjf Amber11.tar.bz2
> >>>>>> 2) tar xvjf AmberTools-1.5.tar.bz2
> >>>>>> 3) export AMBERHOME=/path_to/amber11
> >>>>>> 4) cd $AMBERHOME
> >>>>>> 5) wget http://ambermd.org/bugfixes/AmberTools/1.5/bugfix.all
> >>>>>> 6) patch -p0 < bugfix.all
> >>>>>> 7) rm -f bugfix.all
> >>>>>> 8) wget http://ambermd.org/bugfixes/11.0/bugfix.all.tar.bz2
> >>>>>> 9) wget http://ambermd.org/bugfixes/11.0/apply_bugfix.x
> >>>>>> 10) chmod 700 apply_bugfix.x
> >>>>>> 11) ./apply_bugfix.x $AMBERHOME/bugfix.all
> >>>>>> 12) cd $AMBERHOME/AmberTools/src/
> >>>>>> 13) ./configure gnu
> >>>>>> 14) make
> >>>>>> 15) cd ../../
> >>>>>> 16) cd src
> >>>>>> 17) make serial
> >>>>>>
> >>>>>> At this point you have the serial (non-GPU) versions of AMBERTools
> and
> >>>> AMBER
> >>>>>>
> >>>>>> built. Best to test them at this point.
> >>>>>>
> >>>>>> 18) cd $AMBERHOME/AmberTools/test/
> >>>>>> 19) make test
> >>>>>>
> >>>>>> Check the test logs to see if everything worked.
> >>>>>>
> >>>>>> 20) cd $AMBERHOME/test
> >>>>>> 21) make test
> >>>>>>
> >>>>>> Again check the test logs to see if everything worked.
> >>>>>>
> >>>>>> Now we can build the serial GPU code.
> >>>>>>
> >>>>>> 22) cd $AMBERHOME/src/
> >>>>>> 23) make clean
> >>>>>> 24) cd $AMBERHOME/AmberTools/src/
> >>>>>> 25) make clean
> >>>>>> 26) ./configure -cuda gnu
> >>>>>> 27) cd ../../
> >>>>>> 28) cd src
> >>>>>> 29) make cuda
> >>>>>>
> >>>>>> Hopefully this time it should build correctly.
> >>>>>>
> >>>>>> 30) cd ../test
> >>>>>> 31) make test.cuda
> >>>>>>
> >>>>>> All the best
> >>>>>> Ross
> >>>>>>
> >>>>>> /\
> >>>>>> \/
> >>>>>> |\oss Walker
> >>>>>>
> >>>>>> ---------------------------------------------------------
> >>>>>> | Assistant Research Professor |
> >>>>>> | San Diego Supercomputer Center |
> >>>>>> | Adjunct Assistant Professor |
> >>>>>> | Dept. of Chemistry and Biochemistry |
> >>>>>> | University of California San Diego |
> >>>>>> | NVIDIA Fellow |
> >>>>>> | http://www.rosswalker.co.uk | http://www.wmd-lab.org/ |
> >>>>>> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> >>>>>> ---------------------------------------------------------
> >>>>>>
> >>>>>> Note: Electronic Mail is not secure, has no guarantee of delivery,
> may
> >>>> not
> >>>>>> be read every day, and should not be used for urgent or sensitive
> >>>> issues.
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>> _______________________________________________
> >>>>>> 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
> >>>>>
> >>>>> _______________________________________________
> >>>>> 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
> >>>>
> >>> _______________________________________________
> >>> 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
> >
> > _______________________________________________
> > 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
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Feb 27 2012 - 12:00:02 PST
