Re: [AMBER] Fwd: AMBER 22 installation with CUDA support

Date: Mon, 5 Sep 2022 22:23:50 +0530

 Dear Prof. Valentin,

Thank you for your valuable suggestions. Although the default CUDA version
(CUDA 8.0) would suffice my need, It is mainly the GCC version that I am
worried about. As AMBER 22 requires GCC 6 or above and CUDA 8.0 supports
GCC 5.3 at max (as far as I know).

So, I think I have to ask the system administrator about installing an
updated version of the CUDA toolkit. Let's see if that solves the problem.
Thanks again.

With best regards,
Satyajit Khatua

*From:* Rybenkov, Valentin V. <>
*Sent:* Monday, September 5, 2022 10:03:40 PM
*To:* SATYAJIT KHATUA <>; AMBER Mailing List <>
*Subject:* RE: [AMBER] Fwd: AMBER 22 installation with CUDA support


I had similar problems installing Amber on a managed computer cluster.
There were constant conflicts between preinstalled packages and what I was
downloading. What worked for me was to use the preinstalled Cmake, OpenMPI,
etc and forgo the installation of my own versions. Just use the module load
xxx command. If you need a newer version of a soft than you have on your
cluster, talk to your system administrator.


Valentin V. Rybenkov
University of Oklahoma
101 Stephenson Parkway
Norman, OK 73019

*Sent: *Monday, September 5, 2022 11:03 AM
*To: *AMBER Mailing List <>
*Subject: *[AMBER] Fwd: AMBER 22 installation with CUDA support

As I have previously mailed only to Prof. Case, I am also floating the mail
to the amber mailing list as well. Any help in this regard would be greatly

---------- Forwarded message ---------
Date: Sun, Sep 4, 2022 at 9:57 PM
Subject: Re: [AMBER] AMBER 22 installation with CUDA support
To: David A Case <>

Dear Dr. Case,

As you have suggested, using a newly installed CMake compiler I have
successfully installed the GPU accelerated serial AMBER 22. But, the
test.cuda.serial has encountered the attached error. However inspecting the
log, I have seen that it is searching for a file. But the
installed CUDA version is 9.2. Thus there is file in my
LD_LIBRARY_PATH. Do I have to unlink the root installed

My .bashrc already contains the CUDA_HOME and its related LD_LIBRARY_PATH
variables. What am I doing wrong here? Please advise.

With regards,
Satyajit Khatua,
Research Scholar,
Department of Chemistry,
IIT Kharagpur, Kharagpur-721302

On Sun, Sep 4, 2022 at 6:23 PM SATYAJIT KHATUA <>

> Hi,
> Thanks for the quick response. I will try to install some recent releases
> on the Let's see if that works. I shall get back to you.
> Thanking you,
> Satyajit Khatua
> ------------------------------
> *From:* David A Case <>
> *Sent:* Sunday, September 4, 2022 6:02:27 PM
> *To:* SATYAJIT KHATUA <>; AMBER Mailing List <
> *Subject:* Re: [AMBER] AMBER 22 installation with CUDA support
> On Sat, Sep 03, 2022, SATYAJIT KHATUA via AMBER wrote:
> >
> >I am installing AMBER22 in our departmental CentOS 7 cluster,...
> >
> >(message): Error generating
> >***
> >Error 1make[1]: ***
> >Error 2make: *** [all] Error 2*
> Our Centos installation notes (
> suggests
> installing cmake from the website. Above is a cmake-error, but
> I
> don't remember seeing it reported by others; perhaps they were using a
> different version of cmake.
> One workaround: during the CUDA build phase, add this to your run_cmake
> script:
> -DDISABLE_TOOLS="cpptraj"
> ...good luck....dac
AMBER mailing list
Received on Mon Sep 05 2022 - 10:00:02 PDT
Custom Search