Michael Brown (mbrown@stranj)
Mon, 23 Nov 1998 12:54:13 -0800
A couple of comments.
Chikai's suggestion is a very good one, and while it will not give perfect
results when you integrate geometry, the geometry will be rendered correctly
and with tri-linear interpolation (O.K., bi-linear with a single image), it
should still look pretty good.
The other thing that I have seen done is to use Dynamic Video Resizing to
stretch a window to fit an entire display. This has the same effect, but
does not require a readback from the Framebuffer to the memory so it may
improve your frame rate slightly. Again, you get very good interpolation,
and the displayed results are quite acceptable. Of course, with stereo
imagry you want the edges of the geometry to be as crisp as possible to
improve depth perception, and any type of re-sizing will introduce a very
limited amount of fuzz.
Finally, your configuration is a little light for maximum performance stereo
volume rendering. Unless you are using a very complex volume decomposition,
Onyx2 performance is 100% driven by texture fill rate. (If you find that you
are limited by the calculation of the geometry, you could use the same geometry
for the painting plans for both the right and left eye - probably defined by
the nose. This gives you a correct image for each eye, and is only very
slightly less perfect than recalculating the geometry for each eye. However, I
doubt
that this is an issue with your data set.
Your system has a single RM7-64 with which you are driving two seperate high
resolution displays in order to get right and left eye viewpoints. If you
zoom to full screen for each eye, you are drawing 2.5M pixels per rendering
plane, and if you are drawing 100 planes, then you would have 250M pixels for
each time step. This is a lot to ask of a single RM, and I would expect less
than 0.5 frame per second in this mode versus 8 frames per second when
rendering a stereo 320x256 window and expanding it or 4 frames per second
when rendering a 640x512 window and expanding it.
To maximize performance on rendering of volumes of <=64MB in size, then you
should have as many Raster Managers as possible since they will give you a
linear speedup in performance.
At SC98, Ian Foster from Argonne was using Volumizer in a Pyramid Systems
CAVE to look at volume data in stereo while driving a 1280x1024 stereo
display. That system had 4RM's per pipe, and he was getting much more
than the 2 frames per second he would have achieved without one of the two
resizing approaches.
I hope this helps.
Regards,
Michael
On Nov 23, 13:16, Robert van Liere wrote:
> Subject: rendering speeds
>
> For what its worth: we've got Volumizer running in a VR
> setting (i.e. head tracked, stereo, spatial input, large
> back projected display). We are looking at cell data
> comming from confocal microscopes.
>
> To be a succesfull VR environment, want to run at least
> 10 frames / sec minimum (in stereo). Frame rates are quite acceptable
> for sets with res 128x128x32. Frame rates are to low for sets of
> resolution of 256x256x64.
>
> Question: whenever we zoom into the volume, the frame rate slows
> down substantially. Why is this? (Perhaps because the triangles get
> very large (i.e. rasterizition bottlenecks) ?) Can we do something to
> improve the rendering speeds for these cases ?
>
> Any info will help,
> -- Robert van Liere
>
> ; uname -a
> IRIX64 perseus 6.5 05190003 IP27
> ; /usr/gfx/gfxinfo
> Graphics board 0 is "KONAL" graphics.
> Managed (":0.0") 2560x1024
> Display has 2 channels
> 4 GEs (of 4), occmask = 0x0f
> 4MB external BEF ram, 32bit path
> 1 RM7 board (of 1) 1/0/0/0
> Texture Memory: 64MB/-/-/-
> Small pixel depth
> 32K cmap
> Channel 0:
> Origin = (0,0)
> Video Output: 1280 pixels, 1024 lines, 72.00Hz (1280x1024_72.vfo)
> Channel 1:
> Origin = (1280,0)
> Video Output: 1280 pixels, 1024 lines, 72.00Hz (1280x1024_72.vfo)
>-- End of excerpt from Robert van Liere
-- Michael Brown, Business Development Manager Advanced Graphics Marketing, Silicon Graphics Computer Systemsemail: mbrown@engr.sgi.com phone: +1(650)933.35.48 fax: +1(650)964.86.71
This archive was generated by hypermail 2.0b2 on Mon Nov 01 1999 - 10:49:29 PST