In case this helps, here is a thumbnail sketch of what I have done.
I am trying to build a system to do md calculations. I picked up a tesla
(1060) card and decided to install it in a vanilla machine and (i7-870)12
GB RAM. I ran into challenging issues.
1 - accidentally install the 32-bit system instead of the 64bit
2 - Clean install of 64-bit system. before installing the nvidia drivers
and packages necessary for amber i auto-updated. This automatically
installed some nvidia drivers of a previous version (295.40). when i
installed the developers drivers (295.41)… there was a version conflict and
I lost all x-terminal access (GUI) after a few hours of trying to
resuscitate I gave up.
3 - So I did another clean 64-bit install. started with the necessary
amber packages and the did the nvidia stuff.
I found that using make -k was really helpful in terms of getting the sdk
to install the trouble shooting app that amber guru's ask that you have
installed (deviceQuery).
I successfully Finished serial and parallel installation and tests. No
problems.
cuda install - as I posted before. No good.
I finally decided to update the packages on the system since I seemed to
have hit a wall. updates via the package manager, crushed my system.
nvidia cards war no longer found. i installed synaptic package manager and
searched for nvidia. I found what looked to be the most recent dev version
and installed it. that rescued the display and the cards were once again
recognized.
- this updated the driver version to 295.49.
cuda install again failed at the same place.
I tried david case's suggestion but… no dice.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 30 2012 - 12:00:02 PDT
