[AMBER] amber18 tests failure

From: Antonio Bauzá <antonio.bauza.uib.es>
Date: Thu, 20 Feb 2020 10:37:39 +0100

Message: 14
Date: Wed, 19 Feb 2020 16:46:51 +0100 (CET)
From: Antonio Bauz? Riera<antonio.bauza.uib.es>
Subject: [AMBER] amber18 tests failures
Content-Type: text/plain; charset="utf-8"

Dear amber community, I have installed amber18 on my computer to perform gpu
accelerated MD calculations. I am using openSUSE Tumbleweed. I am running the
calculations on a Geforce RTX2080ti. While running the tests after installation
I have seen that some of them didn't pass, however I do not understand the
output given. I am attaching both log and diff files, I would appreciate if you
could shed some light on this.
Thank you in advance,
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 2020-02-19_16-19-35.diff
Type: text/x-patch
Size: 85479 bytes
Desc: not available
Url :http://lists.ambermd.org/mailman/private/amber/attachments/20200219/a21b0dbb/attachment-0002.bin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 2020-02-19_16-19-35.log
Type: text/x-log
Size: 60228 bytes
Desc: not available
Url :http://lists.ambermd.org/mailman/private/amber/attachments/20200219/a21b0dbb/attachment-0003.bin


Message: 15
Date: Wed, 19 Feb 2020 11:02:41 -0500
From: David Cerutti<dscerutti.gmail.com>
Subject: Re: [AMBER] amber18 tests failures
To: AMBER Mailing List<amber.ambermd.org>
Content-Type: text/plain; charset="UTF-8"

I don't see "upload failed" very often. If I had to take a guess, I'd say
your GPU may be broken... but there are plenty of cases that do pass,
including standard PME-MD and many cases of standard GB-MD even though the
first failures are versions of the GB Ala3 case. The differences in the
calculated energies are quite large for cases that should be strictly
deterministic. Do you have a colleague with a GPU machine you might try
installing Amber on?

Perhaps also try an SPFP installation and see what those tests look like
(failures are expected due to floating point precision conventions, with
relative errors in the range of 1.0e-4 to 1.0e-5).

One other thing, could you perhaps re-run the test suite to see if you get
the same errors?

Is anything else running on the card when you do the tests?


Hi Dave, I've repeated the tests today and obtained the same results, there are no other things runnning when I do the tests. The GPU is on a desktop computer, not a server. When installing AMBER I just followed the instructions from the AMBER webpage. I have cuda 9.2 installed on my system.

./configure -cuda gnu
make install

How can I specify which precision model I want to install?

Best regards,


AMBER mailing list
Received on Thu Feb 20 2020 - 02:00:01 PST
Custom Search