[vtkusers] debug volume rendering performance with volumepro1000

Lisa Avila lisa.avila at kitware.com
Fri Jan 16 13:09:35 EST 2004


Hi Daniel,

It has been a while since we have done performance tests but your numbers 
seem like what I would expect. We were experiencing about 3 to 6 frames per 
second with a 512x512x200 data set that we used for testing with the 
VolumePro 1000 board. Keep in mind that the frame rate will be impacted by 
the display size (the more pixels to be covered, the more work the board 
must do to generate the image). There is also the issue that VTK is 
sometimes increasing your image size due to a board limit that adjacent 
rays cast can not intersect the volume "too far away" from each other. This 
is an ugly calculation that sets the minimum image size that can be 
requested to avoid this error from VLI.

We found these frame rates in VTK matched those obtained with the Revli 
application. VTK does not do a lot of work per render - just take the 
returned data and map it on a polygon for display. I suspect that TeraRecon 
may have done more work on the VLI interface in the past year or two for 
performance reasons - it is possible that VTK is not taking advantage of 
some "recent" change that could improve performance.

You best method of improving performance is to use a vtkLODProp3D with two 
resolutions of your data and two mappers. This will render the low res data 
during interaction and fill in with the high res data when you stop 
interacting. This will allow for smaller image sizes and faster performance.

Lisa





At 07:18 AM 1/16/2004, Dr. Daniel James White PhD wrote:
>Hi All,
>
>We have recently acquired a volumepro1000 volume graphics board to speed 
>up the volume rendering of our large data sets in MayaVi,
>which is a VTK application for data visualisation with a user interface 
>written in python.
>
>Using vtkVolumeProMapper in place of vtkRayCastMapper or 
>vtkTextureMapper2D we do indeed see an increase in rendering performance.
>The raycastmapper is much slower than the volmepromapper, but the 
>TextureMapper2D is only 2 or three times slower.
>We get frame rates between 10 and 1 frame per second using the volmepromapper.
>This board should give 30 fps!
>
>Our data sets are smaller in all 3 dimensions than the  512x512x512 limit 
>that I believe the volumepro1000 board can handle with its on board 
>memory,  so the data should not be "paged" by the board.
>
>The board is in a short/33MHz PCI slot, not the recommended long/66MHz 
>slot recommended.
>Would this cause the slow performance? I think I saw that it might only 
>give a 10% performance it, but I could be wrong.
>
>Otherwise, what is the best way of going about finding bottlenecks in the 
>VTK and hardware rendering pipeline?
>How do I actually test the speed of each part of the pipeline to find the 
>bottleneck?
>
>
>Possible places might be
>1 PCI bus between the processor and the volumepro1000 card,
>2 python,
>3 VTK c++,
>4 ATIRadeon8500 graphics card,
>
>One extra bit of information I have is due to the volume pro board having 
>a small defect which causes it to squeak every time the display is 
>refreshed/redrawn. So I hear a noise for every frame update.
>It kind of sounds like the volumepro1000 board is only rendering as many 
>frames per second as the graphics card is displaying (or at least doing 
>something every time the display updates..... the noise comes from a power 
>converter apparently).
>Is this because that is the fastest frame rate the volumepro1000 can 
>handle under these conditions (it should be faster according the the 
>documentation), or is it that VTK is not sending the rendering 
>instructions to the volumepro1000 board fast enough, or something else?
>
>My system is redhat linux 8, VTK and MayaVi from very recent CVS, athalon 
>1.5GHz processor, ATIRadeon8500, volumepro1000 512MB in 33MHz PCI slot 
>with linux vli3 library.
>
>Any debugging suggestions very welcome!
>
>cheers
>
>Dan
>
>Dr. Daniel James White BSc. (Hons.) PhD
>Cell Biology
>Department of biological and environmental science
>PO Box  35
>University of Jyväskylä
>Jyväskylä FIN 40014
>Finland
>+358 (0)14 260 4183 (work)
>+358 (0)414740463 (mob)
>
>http://www.chalkie.org.uk
>dan at chalkie.org.uk
>white at cc.jyu.fi
>_______________________________________________
>This is the private VTK discussion list. Please keep messages on-topic. 
>Check the FAQ at: <http://public.kitware.com/cgi-bin/vtkfaq>
>Follow this link to subscribe/unsubscribe:
>http://www.vtk.org/mailman/listinfo/vtkusers





More information about the vtkusers mailing list