Reverting to Reference Platform

[Jack Shultz]

I recently installed Zephyr from the binary distribution and I cannot get the GPU platform working. I get a message saying "Reverting to Reference platform. You asked for GPU OpenMM nVIdia. Inspect log for details. About a year ago I had it working on my Windows 7 preview release, same machine. That was 32 bit. This one is 64 bit. I'm not sure which log file I am supposed to inspect for further details.

[Gary Skinner]

Make sure you have the correct version of the CUDA sdk installed, you need version 2.2 or later. Also, make certain your graphics card is one of those that supports CUDA, this you can check on the nvidia website.

[Jack Shultz]

Installed Developer Drivers for WinVista and Win7 (257.21) 64-bit
Installed CUDA Toolkit 3.1 (June 2010)
Installed pre-compiled binary again
It installs Visual Studio 2008 Redistributable

Ran the example on both OpenCL and CUDA platforms
C:/Zephyr/testData/alanylalanine_capped.pdb

OpenCL runs ok but CUDA referts to reference

Center of mass motion removal mode is Linear
We have the following groups for center of mass motion removal:
0: rest
There are: 32 Atoms
agb parameter file line=<> is being skipped.
agb parameter file line=<> is being skipped.
OpenMM Platform: Reference

I'm having a suspicion there is a permission or file ownership issue. Windows 7 is kind of funny the way it administers permissions.

[Gary Skinner]

A bit like Vista probably. Have you tried running as administrator, or running in XP compatibility mode? I know I had this problem myself, and I think I resolved it by making sure I had the right CUDA version. I see you have 3.1, but Zephyr specifies that you need 2.2. I know it says "or later", but have you tried just using the older 2.2, maybe that solves it?

Gary

[Maxim Kondratiev]

This problem present in my system too!
(Win7-64 CoreDuo T6600 + Nvidia GT320M)
I think Win 64 cause is it. Dear developers! Can you solve this problem with Zephyr? Admin mode and XP simulation cannot help.

[Joy Ku]

Maxim,

We currently only support Zephyr on 32-bit Windows and don't have any experience with it on 64-bit. However, here are some things you might try to help pinpoint what's going on.

1) Check if the acceleration works with OpenCL, CUDA, or neither. Use the pull-down menu in Zephyr to choose the specific platform.

2) Determine if the problem is with the Zephyr interface or the underlying OpenMM libraries and how they interact with the NVIDIA libraries.

Do this by downloading the OpenMM pre-compiled binary for Windows and examples from here: https://simtk.org/project/xml/downloads ... oup_id=161. Run the HelloArgon program and see if you get acceleration. If not, the problem is with OpenMM, and you should post a question to the OpenMM forum about this. If the HelloArgon program does run, the problem is with Zephyr.

We'd be very interested in hearing how things go.

Best,
Joy





Joy

[Maxim Kondratiev]

What you mean?
I've run TestReferenceHarmonicBondForce program from precompiled package - success with message: "Running test - PASS". Where is HelloArgon program?

[Joy Ku]

HelloArgon is a program that comes with the OpenMM examples located at http://simtk.org/home/openmm (click on Downloads). The suggestion is to separately install the OpenMM binary there and run the example program to diagnose if the problem is with the Zephyr GUI layer or the underlying OpenMM library (or neither).

Joy

[Maxim Kondratiev]

acceleration with OpenCL and normal MD (using CPU) works fine.

[Maxim Kondratiev]

Success!
The HelloArgon on WinXP-32bit and Win7-64bit generate following list (on the long-scrolled screen)
MODEL 252
ATOM 1 AR AR 1 0.201 0.000 0.000 1.00 0.00
ATOM 2 AR AR 1 5.000 0.000 0.000 1.00 0.00
ATOM 3 AR AR 1 9.799 0.000 0.000 1.00 0.00
ENDMDL

What that mean?
Problem with Reverting to Reference Platform is with Zephyr GUI? Because two computers with openMM and CUDA generate identical output.