VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Wed Apr 17 2019 - 08:41:53 CDT

Bart,
  I'm traveling so I only have a minute to answer, but I'll try
and give you a brief answer and add more details later if you're curious.
The main reason the ray tracing engines run faster than OpenGL is that
they benefit from spatial acceleration structures (bounding volume
heirarchies) that accelerate the core ray-object intersection search
by many orders of magnitude. Effectively, the acceleration structures
convert ray tracing from a linear time algorith with respect to the
number of objects in the scene, into a log(N) algorith with respect
to the number of objects in the scene. Rendering time is still roughly
linear with the number of pixels however, so you end up with someting
that's roughly Npixels Log(Nobjects) or so.

It is my goal to (optionally) replace rasterization with ray tracing in
the default VMD display, but to achieve this I still need to overcome
some performance hurdles associated with the construction and updates
to the ray tracing acceleration structures. Some very significant
ray tracing software API technology evolution is currently underway, and
I expect that I should be able to achieve this in the coming year if
things go as I expect.

I'm also expecting to speed up the rasterization display in VMD tremendously
in the near future by virtue of a native Vulkan renderer that won't
be hindered by unfortunate workarounds for buggy drivers, particularly
on platforms like MacOS X that barely have any OpenGL support anymore.

Best,
  John Stone
  vmd_at_ks.uiuc.edu

On Wed, Apr 17, 2019 at 01:02:46PM +0200, Bart Bruininks wrote:
> Hello dear VMDers,
> I am currently working at the Molecular Dynamics lab of Groningen (the
> Martini lab) and we are testing the new visualization node with VMD
> 1.9.4.x. Most things are working just fine and we are exploring the
> possibility of the P6000RTX cards. However, we already have very
> reasonable real time interactive ray-tracing on the V100 cards using the
> AVX acceleration (15 fps for 10 million atoms with AO and shadows and
> quicksurf all). During the testing of the node I ran into an unusual
> problem. The rasterized 'quick' rendering which VMD uses when you want
> to quickly look at your system and the default display, runs at a much,
> much much lower fps (0.3) than the interactive ray-tracing (15).
> Actually, if I could also change the frame and the lighting positions, I
> see no use in not having the option to make the interactive ray-tracing
> my default display. I was wondering if you guys thought about this and
> if there is an explanation as to why the ray-tracing rendering is faster
> than the rasterization, and if so why are we even still using the
> rasterization?
> Cheers,
> Bart Bruininks 

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