[Chimera-users] odd chimera benchmark performance

[Tom Goddard]

Hi Dan,

  Silhouettes can kill performance, sometimes 100x slower, but more often just 3x or 5x slower.  It uses an obscure opengl feature (glPolygonMode) that is probably preventing hardware acceleration with your driver.  I would expect depth cue to have no effect.  Factory Chimera settings have silhouettes off, depth cue on.  You must have saved the setting for depth cue on (it ends up in your Chimera preferences file).  One other setting to avoid is multisampling (Chimera menu Tools / Viewing Controls / Effects) — this can be a killer if your driver does not support it well.

	Tom


> On Aug 13, 2015, at 5:13 PM, Dan Lin  wrote:
> 
> Hi Tom, Greg:
> 
> Thanks for replying.
> 
> I am pretty sure I do have proprietary driver installed.
> 
> I checked in the report a bug dialog and got:
> 
> OpenGL Vendor: NVIDIA Corporation
> OpenGL Renderer: GeForce GTX 560 Ti/PCIe/SSE2
> OpenGL Version: 4.2.0 NVIDIA 304.116
> 
> However, I was checking if there was anything else that would affect performance, and it looks like my version of chimera has silhouettes and depth cueing on by default. When I turned both those effects off I saw some pretty big improvements in benchmark and real-life performance:
> 
> Old molecule benchmarks:
> 6249.5	158.7	1.5	0.9	27.4	0.4	89.3
> 
> New molecule benchmarks:
> 369581.3	164.7	201.1	130.3	2145.8	51.3	103.3
> 
> Which I guess explains the performance differences, unless the benchmarks on the website were with both of those effects on as well. Are there perhaps any other effects I could be on the lookout for that would also drain performance when dealing with millions of atoms?
> 
> Cheers,
> Dan
> 
> On Thu, Aug 13, 2015 at 4:59 PM Tom Goddard wrote:
> You probably don’t have the Nvidia graphics driver installed, so you are getting slow Mesa rendering.  To check this use Chimera menu Help / Report a Bug… and look in the “Gathered Information” box where it will identify the OpenGL renderer.  For instance on my Mac it says:
> 
> OpenGL Vendor: NVIDIA Corporation
> OpenGL Renderer: NVIDIA GeForce GTX 680MX OpenGL Engine
> OpenGL Version: 2.1 NVIDIA-10.4.2 310.41.35f01
> 
>   Tom
> 
> 
> > On Aug 12, 2015, at 11:53 PM, Dan Lin wrote:
> >
> > Hello:
> >
> > I was wondering if anyone had any experience tracking down performance problems in chimera.
> >
> > I've noticed that I am getting really sluggish performance when managing a lot of atoms, even when everything is shown as backbone traces only.
> >
> > To take a look at the performance, I ran the benchmark and got the following numbers:
> >
> > Volume benchmark scores
> >  surface    mesh  contour  solid  recolor
> >      981    1291     280     589     393
> >
> > Null  Wire    Stick   BStick  Ribbon  Sphere  Ops
> > 6249.5        158.7   1.5     0.9     27.4    0.4     89.3
> >
> >
> > compared to a benchmark posted on the website of the same graphics card:
> >
> > 2125 2026 302 512 491
> >
> > 213 492 452 3259 169 128
> >
> > Apparently I'm getting awful benchmarks for most of the molecule benchmarks, while the volume performance is pretty comparable. I re-ran the molecule benchmarks to make sure they were reproducible and got pretty much the same numbers.
> >
> > Would anybody have any suggestions for figuring out what's going on?
> >
> > For reference, I'm running ubuntu 12.04 with up to date nvidia drivers for a GTX 560 Ti.
> >
> > Thanks,
> > Dan
> >
> >
> > _______________________________________________
> > Chimera-users mailing list
> > Chimera-users at cgl.ucsf.edu <mailto:Chimera-users at cgl.ucsf.edu>
> > http://plato.cgl.ucsf.edu/mailman/listinfo/chimera-users <http://plato.cgl.ucsf.edu/mailman/listinfo/chimera-users>
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://plato.cgl.ucsf.edu/pipermail/chimera-users/attachments/20150813/2e9dea9f/attachment.html>