From: Robert Wohlhueter (bobwohlhueter_at_earthlink.net)
Date: Mon Sep 22 2014 - 13:39:19 CDT
I am now running Ubuntu 14.04 on an AMD64 machine with a GTX 660Ti
graphics board. The out-of-the-box instalation of 14.04 includes CUDA
version 5.5. After my upgrade to 14.04 my previous installation of
namd2 (NAMD_2.9_Linux-x86_64-CUDA) seems to be broken.
My first impulse was to install a newer namd build, namely
NAMD_CVS-2014-09-20_Linux-x86_64-multicore-CUDA. Running that version
presents cuda errors -- which is doubtless explained from the note in
the namd-user-manual that v. 2.10b1 (and newer?) "supports" CUDA 6.0
(probably implying that it does not support CUDA 5.5). Following the
manual's advice to make sure the libcudart.so delivered with
NAMD_CVS-2014-09-20_Linux-x86_64-multicore-CUDA is out-front in
LD_LIBRRY_PATH is of no avail.
My next idea was thus to upgrade my cuda support. NVIDIA's website has
a recipe for that:
ran "sudo -i " on that deb package, then "sudo apt-get update", then
"sudo apt-get install cuda" and got complains, most saliently that
dependencies were not met. Exploring this approach a little more with
synaptic, I find synaptic lists "cuda" and claims a latest version as
6,5-14. But when I mark it for install, it tells me that it will
uninstall a bunch of things, including ocl-icd libs and wine! Those
removals would cause me a lot of grief.
So it seems the path of least resistance would be to try to revert to a
previous version of namd-cuda which would be compatible with v. 5.5 of
cuda. Which is that?
Or perhaps one can suggest a slick way of installing getting to CUDA 6.5
without destroying WINE, etc.
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:52 CST