[Chimera-users] Maximizing performance with large volume data maps

[Thomas Goddard]

Hi Ryan,

   Take a look at the Chimera graphics benchmarks page to get an idea 
how various graphics cards perform.

	http://plato.cgl.ucsf.edu/trac/chimera/wiki/benchmarks

If you are using surface rendering in volume viewer then the relevant 
number in the benchmarks is the surface score.  It indicates the size N 
of a cube that can be rendered at 10 frames per second where each face 
of the cube is rendered as 2*N*N triangles.

   For your map (from the size I guess a virus) you are probably getting 
slower frame rates than for the benchmark cube at 550^3 size because the 
virus contour surface has onion-like internal layers.  Even though those 
layers aren't visible they take lots of time in the rendering.  You 
could clip the map (use Favorites / Side View and move left yellow 
vertical line to the right) to see the internal structure.  If there are 
many layers at the contour level you like and your primary interest is 
the protein layer of the capsid you might consider making a new map 
where you zero the interior (genome part) of the virus.  You can perhaps 
use Chimera volume eraser or the "shape sphere" command and "mask" 
command.  This will make the surface have many fewer triangles giving 
faster rendering.

   I think your graphics rendering rate is not being limited by graphics 
memory, but rather by GPU speed.  A look at the benchmark web page at 
cards with different amounts of memory will give you an idea of what you 
can expect with more memory.

   The benchmark test cube has 12*N*N triangles so a benchmark surface 
score of 1000 means 12 million triangles are rendered 10 times per 
second -- ie. 120 million triangles per second.  You can see how many 
triangles per second you are getting with your map.  Select the map 
surfaces (ctrl-click) and use Actions / Inspect.  The number of 
triangles will be listed at the bottom.  Then Tools / Utilities / 
Benchmark will allow you to measure the frame rate.  Clear the selection 
before measuring the frame rate because the selection outline is 
rendered with a 5-pass drawing algorithm that greatly slows the 
rendering.  Use the "Measure frame rate continuously" checkbutton.

	Tom


-------- Original Message --------
Subject: [Chimera-users] Maximizing performance with large volume data maps
From: Ryan Rochat
To: chimera-users
Date: 12/4/09 7:15 AM

> Hello, I was wondering how much video memory Chimera is capable of 
> accessing.  I currently have a 898Mb videocard that begins to slow down 
> with a 550x550x550 map (at step size 1).  This map is actually a 
> downsample of the true 1100x1100x1100 map coming in at 5GB.  I was 
> contemplating going for a much larger video card at 4GB, however I was 
> curious to what extent I could expect to see acceleration using Chimera 
> with such a massive dataset (600+MB) and video card.
> 
> Thank you,
> Ryan H. Rochat MS
> MD/PhD Candidate (Baylor College of Medicine)
> Graduate Student SCBMB (Baylor College of Medicine)
> Chiu Lab
> One Baylor Plaza N420 Alkek
> Houston Texas 77030
>