From: John Stone (johns_at_ks.uiuc.edu)
Date: Tue Nov 09 2010 - 11:16:54 CST

Hi,
  I'm behind on emails because I have a bunch of paper reviews to get done
today, but here's a quick comment:

SLI should help VMD, but in order to benefit, you may have to enable
display list caching of the scene geometry. This is done by selecting
Display->Cachemode->On in the main VMD window, or by running the command
(note capitalization):
  display cachemode On

This will download _all_ of the VMD scene geometry onto the GPU and keep
it there, which can be a huge performance boost when rendering mostly-static
molecular structures over a remote network connection. It is also beneficial
for the graphics driver since the driver can pre-optimize the displayed geometry
and reuse it over and over every time the scene is redrawn.

The main drawback of enabling display list caching in VMD is that if you are
animating a trajectory, you'll take a 2x performance hit in some cases since
the graphics driver is usually doing extra "optimization" work that is not
amortized over multiple redraws. So, while it's great for large static
structures, it's probably not what you want when animating a trajectory.

Future versions of VMD will benefit much more from SLI, but probably not
until the last bits of immediate-mode OpenGL are gone from the rendering
pipeline in VMD...

Cheers,
  John

On Tue, Nov 09, 2010 at 11:51:32AM -0500, Axel Kohlmeyer wrote:
> On Tue, 2010-11-09 at 17:43 +0100, Prskalo, Alen-Pilip wrote:
> > ***************************************
> > > important, graphics card. As stated, I'll take SLI, so if one GPU is to
> > > little, I'll take the advandage of havind the other one also. I still
> >
> > not sure, if there actually is a benefit from SLI on linux /w opengl.
> > a colleague tried it and didn't see anything.
> >
> > ***************************************
> >
> > Does anyone have an opinion onto if two graphics cards connected by SLI would improve the VMD representation, has anyone tried it out?
> > As already said, I am buying some workstations now, so if would like to know should I buy one or two Nvidias and I don't want to throw money out of the window if it doesn't help ;-)
>
> google is your friend:
>
> as far as i understand you would need specific support
> in the application to benefit from sli, but it looks even
> for full screen linux apps that do, it is not worth it.
>
> http://www.nvnews.net/vbulletin/showthread.php?t=153283
>
> axel.
>
> > I agree that a representation of 1M atoms in CPK doenst make sense, I simply wanted to know what I can expect.
> >
> > Thanks in advance
> >
> > Alen Prskalo
> >
> > -----Original Message-----
> > From: owner-vmd-l_at_ks.uiuc.edu [mailto:owner-vmd-l_at_ks.uiuc.edu] On Behalf Of Axel Kohlmeyer
> > Sent: Montag, 8. November 2010 19:39
> > To: Prskalo, Alen-Pilip
> > Cc: vmd-l_at_ks.uiuc.edu
> > Subject: Re: vmd-l: Graphics card for VMD, number of atoms, file format
> >
> > On Mon, Nov 8, 2010 at 12:38 PM, Prskalo, Alen-Pilip
> > <Alen-Pilip.Prskalo_at_mpa.uni-stuttgart.de> wrote:
> > > Hi all,
> > >
> > >
> > >
> > > i have allready posted a few times here, I got help quite fast, so I'll try
> > > it again.
> > >
> > >
> > >
> > > Firstly, i indent to by a Nvidia workstation, among other software i
> > > especially indend to run VMD on it. Components should be as following: 2
> > > Intel 6-Cores, around 32 GB RAM, Nvidia mainboard with SLI, and most
> >
> > > important, graphics card. As stated, I'll take SLI, so if one GPU is to
> > > little, I'll take the advandage of havind the other one also. I still
> >
> > not sure, if there actually is a benefit from SLI on linux /w opengl.
> > a colleague tried it and didn't see anything.
> >
> > > have some doubts, either to got with Nvidia Quadro FX 5800 4 GB or Nvidia
> > > GeForce GTX 480, I must addmit i dont know the difference between the Quadro
> > > and GeForce card so I'll take the best one of both models. Quadro is
> > > around 2500 Euro and GeForce arround 350 Euro, so the equation seems to be
> > > very simple. The reason why i doubt the Quadro card (besides the price ;-)
> > > is the fact that a colleaque of mine bought one workstation with a Quadro FX
> > > 5800 in it and was dissapointed, so was I, i tried to rotate a a system
> > > sonsisting out of 1 000 000 atoms in PDB file format. As long as i stayed in
> > > line representation, everything worked fine, but as soon as i used CPK, the
> > > PC hanged, i could not move the system as i wished, i had to wait for
> >
> > this will happen with _any_ card. the Quadro FX5800 is pretty powerful.
> > its graphics performance is roughly equivalent to a GTX 280. so the
> > GTX 480 should be faster, but also _much_ hotter.
> >
> > to get better performance on high end graphics cards you should
> > enable GLSL. this way, your CPU only needs to compute the centers
> > and radii of spheres and GPU does the rest. that being said. for a 1M
> > atom system, you will alway have the lines or points rep be much faster.
> >
> > i regularly visualize systems that are up to 10x larger and always use
> > points or lines at point size 5 or larger together with depth cueing rather
> > than GLSL as that is the only way to operate smooth enough. rendering
> > those systems on the screen with a more detailed rep is more distracting
> > than helpful, IMO. that being said, if i look at subsets of the system, i
> > certainly use the higher quality visualization reps.
> >
> > > minutes until a rotation was performed. Or did i expect to much, could
> > > someone tryed to to this on their own PC with a good graphics card and tell
> > > me if i works.
> >
> > you are expecting far too much. you should also consider that with
> > 1M atoms, you don't have much room for each atom, so what is the
> > point to spend a lot of computing effort on a sphere or cylinder if
> > that results in only a few pixels.
> >
> >
> > > Secondly: would a a Tesla card accelerate VMD graphics if I used it for
> > > visualisation only and dont perform any acctual GPU calculations? So, to
> > > have a Tesla and a Quadro for instance.
> >
> > nope.
> >
> >
> > > Third question: I perform MD calculations on a HPC cluster, the output i get
> > > are checkpoint files, basically snapshots of the system for specific time
> > > step. The file format is of following: 10 columns (atom number, atom type,
> > > mass, x, y, z, v_x, v_y, v_z, E_pot). For a a smooth visualisation with VMD
> > > i transform each file into a PDB file format and then glue them together
> > > with an "END" inbetween each frame. This is recognized by VMD as a time
> > > scale so i can run it as a movie. The problem is offcourse that i have a
> > > huge single file, depending on the number of atoms and number of individual
> > > frames, so VMD gets very slow. As i was told by Axel Kohlmeyer as a respond
> > > on to my previous posts, it is due to the fact that PDB is first of all a
> > > text format, having the data into binary format would improve the
> > > performance. VMD uses some kind of a start configuration and a trajectory
> > > file, both in binary format. Could someone explain me how i get my bunch of
> > > files into this binary format(s).
> >
> > write a converter, or have your MD code output .dcd or .xtc directly,
> > or write an interface to the VMD molfile plugin library.
> >
> > axel.
> >
> > >
> > >
> > > I thank you all in advance for any help i get.
> > >
> > >
> > >
> > > Regards
> > >
> > >
> > >
> > > Alen Prskalo
> > >
> > >
> > >
> > >
> > >
> > > Dipl.-Phys. Alen-Pilip Prskalo
> > >
> > > Institut für Materialprüfung, Werkstoffkunde und Festigkeitslehre IMWF
> > >
> > > Universität Stuttgart
> > >
> > > Pfaffenwaldring 32
> > >
> > > 70569 Stuttgart
> > >
> > > Tel: +49 711 685 62579
> > >
> > > Fax: +49 711 685 62635
> > >
> > > Email: alen-pilip.prskalo_at_mpa.uni-stuttgart.de
> > >
> > > Email: alen-pilip.prskalo_at_imwf.uni-stuttgart.de
> > >
> > >
> > >
> > >
> >
> >
> >
>
> --
> Dr. Axel Kohlmeyer
> akohlmey_at_gmail.com http://goo.gl/1wk0
>
> Institute for Computational Molecular Science
> Temple University, Philadelphia PA, USA.

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078