Dear all,
I want to share my experience, especially with academic colleagues, of
running Amber18 in Amazon AWS, to give you a rough idea of the iter,
costs and performance of the system.
1. Customer activation and setup of EC2 machine
Activation of the account was straightforward, but after selecting the
server zone I had to wait a couple of days to get a P3.2xlarge instance
activated. The default limit is 0 and you need to send a customer
support request to activate it.
I selected an ubuntu 16.4 server image and 64GB of space in a EBS space.
You must select an EBS volume if you want to keep the virtual machine
once switched off. Installation of cuda and amber was straight forward
following the manual of Amber and CUDA 9.2 ubuntu installaation
istructions. I also installed intel MKL libraries.
2. Run
I did an aMD test on a system which has ~180K atoms, in TIP3P water.
AMber18 patch 9 on the Nvidia Volta V100 GPU (single card) performed an
average of 96.5 ns per day.
3. Costs
On this system 25 Hrs of usage and including data transfer accounted
with taxes for about 140 $ . The price does not include storage which is
however quite cheap.
2. Issues
Most of the issues I had to face were related to the too fast data
accumulation on the HD, with hindsight I would have selected a bigger
storage space. To overcome the problem I had to prepare a script which
rsyncs and deletes the trajectories as soon as they are done to keep the
hard drive free.
All the best
On 8/13/2018 5:37 PM, Luca Codutti wrote:
> Dear all,
>
> Sorry for the newbie email. I am planning to run MDs on Amazon AWS, I
> need to run 4 MDs for a largish system (~260K atoms). Being in a
> hurry, for publication reasons, I was planning to make use of CUDA GPU.
>
> Did any of you tried to run MD with a similar set-up? Any advice on
> which EC2 instance would be best in terms of ROI/computational
> time/storage space? Do you know where I could find any benchmark for
> AMBER runs on AWS instances?
>
> Kind regards and thank you for your help
>
> Luca
>
--
Dr Luca Codutti
Carlomagno Group
Universität hannover - BMWZ
Schneiderberg 38, PLZ30167
Tel. +49 511 762 16353
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 22 2018 - 04:00:02 PDT
