VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Fri Apr 14 2017 - 05:03:15 CDT

Got to my next airport, so I'll add a bit more commentary:
  Beyond seeing what the particular visual artifact is that is described
in your prior note (which didn't seem to make it to VMD-L?), I would warn
that you have to be careful with regard to which version of CUDA you
use when pairing that with OptiX. Certain OptiX versions only work
correctly with certain versions of CUDA, and the same goes for
particular GPU hardware. The symptoms of straying from the supported
version/hardware configurations can range from things not working to
various bizarre rendering issues. Also be careful that you don't accidentally
compile with one CUDA version but then have a different version earlier
in your LD_LIBRARY_PATH, or something similar. We've been doing lots
of movie rendering using the OptiX batch renderer in VMD 1.9.3 and
and I haven't seen any shadow rendering issues such as the one you
alluded to you the note you included below, so I would begin by
carefully checking your software versions first.

Cheers,
  John

On Fri, Apr 14, 2017 at 02:29:36AM -0500, John Stone wrote:
>
> Josh,
> I have no idea what issue you're describing since I haven't seen it myself
> It would be useful to get example images that display the problem so
> I can have a look, but I wouldn't expect the old code to change much if
> anything.
>
> Cheers,
> John
>
> On Fri, Apr 14, 2017 at 01:53:52AM +0000, Vermaas, Joshua wrote:
> > Alright, I got fed up enough to try and fix this behavior. I failed
> > miserably at correcting it myself, so instead I have a quickish hack for
> > others that experience this: copy over OptiX* files from the src
> > directory of a VMD 1.9.2 install to replace the same files in the src
> > directory of VMD 1.9.3. This means you'll loose any of the VR-specific
> > changes made for 1.9.3, since you are effectively going back to the
> > 1.9.2 rendering path. You'll need to patch FileRenderList.C to
> > accommodate this too.
> >
> > Instead of:
> > if (optixdevcount > 0) {
> > // Emit a console message during OptiX renderer instantiation
> > // since the JIT compilation and linkage of the 256+ shaders may
> > // take several seconds on machines with several GPUs...
> > if (vmdapp->nodecount == 1) {
> > msgInfo << " Compiling "
> > << OptiXRenderer::material_shader_table_size()
> > << " OptiX shaders on " << optixdevcount << " target GPU"
> > << ((optixdevcount > 1) ? "s" : "") << "..." << sendmsg;
> > }
> >
> > add(new OptiXDisplayDevice(vmdapp, 0));
> > #if defined(VMDOPENGL) && defined(VMDOPTIX_INTERACTIVE_OPENGL)
> > add(new OptiXDisplayDevice(vmdapp, 1, cluster_dev));
> > #endif
> >
> > You go back to the simpler code from 1.9.2:
> >
> > if (optixdevcount > 0) {
> > add(new OptiXDisplayDevice(vmdapp, 0));
> > #if defined(VMDOPENGL) && defined(VMDOPTIX_INTERACTIVE_OPENGL)
> > add(new OptiXDisplayDevice(vmdapp, 1));
> > #endif
> >
> > After those changes are made, recompile. Rendering shadows then behaves
> > correctly when calling the non-interactive renderer. As a happy
> > accident, VMD startup times are reduced, since there isn't any shader
> > compilation at startup.
> >
> > -Josh
> >
> > On 03/31/2017 01:25 PM, Vermaas, Joshua wrote:
> > > Heya folks,
> > >
> > > Has anyone else noticed strange behavior when using ambient occlusion &
> > > shadows coupled with the TachyonLOptiXInternal renderer in VMD 1.9.3?
> > > Relative to the Interactive renderer (see screenshots attached), the
> > > shadow edges look too straight, and are just wrong sometimes (compare
> > > the edges on the bottom of the rightmost aromatic ring, for example).
> > > The differences are huge compared to what they used to be in VMD 1.9.2.
> > > Interestingly the render times have gone down in VMD 1.9.3 as well, so I
> > > suspect internally its just a matter of not enough rays being cast in
> > > the current version, but I haven't found the knob to turn to get the
> > > look I want back. If I'm the only one experiencing this behavior, I can
> > > just go back to VMD 1.9.2 when I need to render animations, but if I'm
> > > not, I'd like to see what workarounds other folks have discovered.
> > >
> > > Thanks!
> > >
> > > -Josh Vermaas
> >
>
> --
> 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/ 

-- 
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/