On Thu, Jun 13, 2019, Francesca Lønstad Bleken wrote:
>
>I am trying to install amber18 (with ambertools19) with CUDA as local
>user on a supercomputer with intel compiler.
>
>When I try to install the CUDA version, however I run into the following
>problem:
Let me add a bit to what Ross already said:
First, your errors are coming from compilation of the cuda version of
pbsa, not pmemd. In the simple case that you don't intend to run
cuda-accelerated Poisson-Boltzmann calculations, edit your
AMBERHOME/AmberTools/src/Makefile: look for the lines:
cuda_serial: math_libraries
(cd pbsa && $(MAKE) clean && $(MAKE) -j1 install )
And comment out the second line, that is, skip building pbsa.cuda.
[Aside: if the only cuda program you want is pmemd.cuda, I think things
are even simpler. After running configure, cd to $AMBERHOME/src and
type "make install".]
[Partial explanation: the cuda code for pbsa is relatively recent, and
may never have been tested on old versions of the cuda SDK. And/or, it
may not have been tested with the Intel compilers.]
If you *are* interested in getting Poisson-Bolztmann calculations to run
on your system, I'm no expert, but I'd suggest three things that are worth
trying:
1. use GNU compilers rather than Intel. There is generally no
need for Intel compilers, since the nvcc compiler is doing the
real work.
2. As Ross said, consider upgrading to a more current version of
CUDA.
3. Try sending an email directly to rluo.uci.edu : that group does
follow the Amber mailing list, but maybe not all the time (since
there is a very low volume of PBSA-related posts here. But the
UC Irvine group would be the ones to know best what might be
going on.
Thanks for the report....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Jun 13 2019 - 09:30:02 PDT
