R: Re: Re: Re: The best equipment for MD simulations

LAMMPS LAMMPS Mailing List Mirror
1

I'm really sorry to receive similar responses! Only recently my university is
working on molecular dynamics and the plans to build a suitable computer
center was born recently and will develop slowly and with appropriate caution!
At this very preliminary stage, being inexperienced in computer science, I
thought I'd write here, with extreme humility, hoping to get some general
information from people who are more experienced than me, who use specific
equipment for MD simulations. My goal was to have a simple benchmark. I'm sorry
I was misunderstood!

----Messaggio originale----
Da: [email protected]
Data: 3-ott-2013 19.19
A: "[email protected]..."<[email protected]...>
Cc: "LAMMPS Users Mailing List"<[email protected]>
Ogg: Re: Re: Re: [lammps-users] The best equipment for MD simulations

I try to rephrase the question. Can you tell me the characteristics of the

most

powerful machines used today to simulate larger systems (especially metals

and

polymers) as done in this page

....and i repeat. none of that matters unless you don't mind wasting
lots of money. vendors *will* screw you big time, if you don't know
better and force them to build the machine you need. if you just clone
somebody else's machine, you will likely not get the machine that is
good for you, and particularly good for really large simulations.

on top of that, you have to deal with all the nasty little details of
managing a procurement process on a larger scale. i have by now
overseen three very large ones (one of which landed a machine in the
top500) and can tell you from personal experience that *every* detail
matters.

i can describe an attitude like yours only as irresponsible.

axel.

LAMMPS Molecular Dynamics Simulator ?

I wrote here hoping that someone could give me the characteristics of the
machines that he personally used for simulations of larger systems.

----Messaggio originale----
Da: [email protected]
Data: 3-ott-2013 18.34
A: "[email protected]..."<[email protected]...>
Cc: "LAMMPS Users Mailing List"<[email protected]>
Ogg: Re: Re: [lammps-users] The best equipment for MD simulations

or to express this in different terms. what you are asking me is the
same as asking me for recommending you the exact brand, model, make,
color, number of seats, engine, transmission, interior design for a
car without telling me what kind of a driver you are, what kind of
roads you have, what kind of service stations are in your neighbor
hood, what size of garage you have, what the typical weather is. all
you tell me is that you want to drive fast, want a car for racing and
have a "lot" of money.

axel.

Thanks for your precious advice Axel! As I said, I'm no expert on

computers and

I'd like to know, indicatively, the main features of a great

supercomputing

center about MD, before talking with likely suppliers. Until today I

used

only

my PC (a common dual core) for MD simulations of little systems and I've

no

idea of characteristics of a cluster for example. My University's budget

is

very high. Can you give me any information about main features (number

of

processors, CPU, GPU, frequency...) of a machine able to simulate

systems

of

great dimensions?

you are not making sense. i just explained why it is impossible to
answer at this level of detail without spending a significant effort
on researching the boundary conditions, testing and benchmarking
representative input decks and talking to the people that are supposed
to operate the machine.

anything else would be grossly unethical, particularly with a "high"
budget (whatever that means).

axel.

----Messaggio originale----
Da: [email protected]
Data: 3-ott-2013 17.13
A: "[email protected]..."<[email protected]...>
Cc: "LAMMPS Users Mailing List"<[email protected]>
Ogg: Re: [lammps-users] The best equipment for MD simulations

Hi,
I'm Caterina, a researcher of Mechanical Engineering Departement of
University of Calabria, Italy. My Departement has to buy a complete

equipment

(machine, post-processors...) in
order to run MD simulations with LAMMPS. Can you give me any

suggestion?

What

is the most powerful machine on the market today? What kind of machine

can

simulate systems of very high dimensions? What features should it

have?

My

University would like to buy the best equipment on the market.
I'm no expert on computers and I would be happy to receive your

advice!

2

I'm really sorry to receive similar responses! Only recently my university is

that is because you are asking the same question and there is only one
correct answer to it.

working on molecular dynamics and the plans to build a suitable computer
center was born recently and will develop slowly and with appropriate caution!
At this very preliminary stage, being inexperienced in computer science, I

computer science has nothing to do with this. we're talking here about
common sense and specific experience.

thought I'd write here, with extreme humility, hoping to get some general
information from people who are more experienced than me, who use specific
equipment for MD simulations. My goal was to have a simple benchmark. I'm sorry
I was misunderstood!

there *is* no simple answer outside of: "it depends"

you can run LAMMPS on a very large variety of machines and on most of
them quite reasonably.
you asked for very specific details about hardware, but if you'd do a
survey of what people are using LAMMPS on, you would get a very large
range. you could run LAMMPS on a raspberry pi, or a cell phone on or
on a gigantic supercomputer. clusters are usually purchased with the
intent of servicing a variety of users. with as flexible an
application as LAMMPS the demands in terms of CPU power vs.
communication, total memory, CPU cache, number of cores per node, with
or without GPU, even if you restrict yourself to just one field
"polymer".

now if you ask: "what is the most commonly used hardware?" that is a
cat of a different color. or what kind of hardware are people the most
happy with. or which machines are the most usable ones.

you saw the story from andrew. while he was running on a machine that
would be very capable for his needs, the configuration and usage
policies were not compatible with his simulation approach. it would be
perfectly possible to rig the very same machine to run very well for
him.

the only way to get an impression of what is good and what not is to
get an account on different machines and try it out for yourself.
without practical hands-on experience you are flying blind and it will
be a problem.

i am dealing with running parallel applications on linux clusters for
over 15 years now and have seen a lot of people in the very same
position that you are in and asking the very same questions that you
are asking.
please understand, that i am using drastic language, because i know
from experience that you are asking the wrong questions and while it
is very frustrating for you now, i am trying to help you avoiding a
much greater frustration and disappointment down the road.

axel.