heavy atom plotting

Hi All,

Right now i’m running MD simulations of up 50 millions atoms , and probably I will go up to > 500 millions atoms.
I use AtomEye , but I notice it becomes slow (it crash) . There is also a BETA Parallel version, but I think wont work. The problem
is that now I Need a Heavy user Atoms plotter ! . Can someone please tell me what are the options available ?
Also what hardware may be required i.e Huge GPU Cards and that weird stuff , luckily I can buy one of those ! .
Also I may have an account at TACC (Texas Advance computer center) , does someone know if they give support for heavy
graphics needs ?

Thanks alot
Oscar G.

Hi All,

Right now i'm running MD simulations of up 50 millions atoms , and probably
I will go up to > 500 millions atoms.
I use AtomEye , but I notice it becomes slow (it crash) . There is also a
BETA Parallel version, but I think wont work. The problem
is that now I Need a Heavy user Atoms plotter ! . Can someone please tell
me what are the options available ?

not many.

Also what hardware may be required i.e Huge GPU Cards and that weird stuff ,
luckily I can buy one of those ! .

your problem is not due to the graphics card, but due to main memory.
i know that VMD for example has been streamlined to handle large
quantities of atoms, but for 500 million atoms, you will need a _huge_
amount of RAM to simply keep one single frame in memory.
you can check this with some simple back of the envelope math:

to simply store the coordinates for 500 million atoms in single
precision floating point format, you need almost 6GB of RAM.
double for double precision. add to that information about other
per atom properties labels, color, radius, element, mass and
whatever is needed to make graphics representations of it,
and you'll find yourself in dire need of as much RAM as you
can stuff into a single machine, or a very smart, distributed
data parallel visualization software.

have you checked out the new "dump image" option in lammps?

Also I may have an account at TACC (Texas Advance computer center) , does
someone know if they give support for heavy
graphics needs ?

define "support"! many centers will be happy to have high-end
users, but most of the "support" staff has about as much
experience with juggling such large quantities of data as you
have. few people have looked at this large atomic data sets.

cheers,
    axel.

Wait … My Math Skills are Weak … Let me see if i got it :
if i have 8 millions atoms, and each one its represented by a coordinate (x,y,z) and further assumption of using a single precision floating point number ( 8 byte in memory) then
I have (50010^6)38 = 12000000000 bytes , now If 1 byte = 9.3110^-10 GB , Then (50010^6)(9.31*10^-10)83 that equals 11.172 GB of RAM memory are needed mmmmm I see

Wait ... My Math Skills are Weak ...... Let me see if i got it :
if i have 8 millions atoms, and each one its represented by a
coordinate (x,y,z) and further assumption of using a single precision
floating point number ( 8 byte in memory) then
I have (500*10^6)*3*8 = 12000000000 bytes , now If 1 byte =
9.31*10^-10 GB , Then (500*10^6)*(9.31*10^-10)*8*3 that equals 11.172
GB of RAM memory are needed mmmmm I see

actually, single precision floats are 4 byte,
but that doesn't change the general message.

axel.

How about using VisIt – see here (https://wci.llnl.gov/codes/visit/ )
This is a parallel visualisation program – I have used it for up to 10^10 atoms

Nigel

I briefly read the manual of VisIt and also found another option call ParaView . Unfourtounately the visualisation of dump files generated from LAMMPS its not directly compatible i.e template Format not supported. However i am sure its feasible to create an script to convert the dump files . May I Ask if someone may be willing to share its Home-Made script (Please) ? . Else I will be in the need of creating my own script which will be time consuming sighs … (as usual)

Ps : Hi Nigel , and thanks for the INFO !

I briefly read the manual of VisIt and also found another option call
ParaView . Unfourtounately the visualisation of dump files generated from
LAMMPS its not directly compatible i.e template Format not supported.
However i am sure its feasible to create an script to convert the dump files
. May I Ask if someone may be willing to share its Home-Made script (Please)
? . Else I will be in the need of creating my own script which will be time
consuming sighs ... (as usual)

the first place to look is the pizza.py toolkit.
http://www.cs.sandia.gov/~sjplimp/pizza.html

cheers,
    axel

Oscar,

I am not familiar with ParaView.

However, VisIt will read a LAMMPS dump (of atom style atomic) directly; no conversion needed. Your only issue is that you will be producing ascii files, which can be quite big for 500M atoms; you may want to consider carefully which variables to dump. Also if you want to see atoms as spheres, rendering will take some considerable time. At this number, however, you can visualise each atom as a point just so you can work out what you are looking at. Further, VisIt can be driven by a script (it has a python interface) so you can set the work up as a batch job to be run on your cluster.

Nigel

Re: the size of file issue, you can also ask LAMMPS to
write out a gzipped dump file by using a file name
like mydump.gz

Steve