OpenGL performance problems on DEC boards

OpenGL performance problems on DEC boards

Post by Anders Hjel » Wed, 08 Jan 1997 04:00:00



Posting this question here is a long shot, but if I am lucky someone can
help, by giving new ideas, or maybe by having solved a similar problem
before.

I am developing 3D graphics for a medical application, where grey scale
images from computer X-ray tomography or magnetic resonance are the basic
objects. From these images various volumes like organs and bony structures
are defined. In the 3D display we show these defined volumes, and
additionally one or more grey scale images.

Typically the objects are rendered as triangulated surfaces with a few
thousand triangles, lit and smoothly shaded. No alpha blending is used. The
grey scale images are displayed as texture decals pasted on squares, no
lighting etc.

Occasionally, the user wants to rotate the view, and in order to speed up
the display we show everything as non-lit wireframes during the rotation.

Our application runs on DEC AlphaStations. For the graphics, at a
reasonable price/performance level, we use the DEC ZLXp-L1 board (Hardware
acceleration of OpenGL, but not of textures, quite old board), but for new
installations DEC wants us to use the PowerStorm 4D20 (No hardware
acceleration at all), since on fast machines (like the AlphaStation 500)
the 4D20 board performs as good as or better than the ZLXp-L1, and 4D20 is
much cheaper.

Our problem is that we do not get the performance we should from the 4D20.
If we use a small test program just doing the display described above, the
two boards are roughly equal in performance (as a matter of fact the 4D20
is much faster in the wireframe mode; when the texture image is included
the ZLXp-L1 is faster since the CPU does the texture at the same time as
the board does the rest). But within our large medical application the 4D20
performance is much, much slower than the ZLXp-L1 and much slower than for
the stand alone test program, eventhough everything else in the
configuration is the same!

We cannot find the reason for this performance difference: In a small
program the 4D20 performance is good , but in a large application the speed
is lost. The large application has allocated 5 - 10 widgets, and several X
pixmaps. However, we cannot detect any swapping between the board and the
main RAM or something similar. And it seems like it is the triangulated
volumes that cause the trouble, the texture image takes roughly the same
time for all cases.

Furthermore, we are not able not obtain a double buffered 24-bit visual on
the 4D20, only a 12-bit double buffered or a 24 bit single buffered one.

We are running under OpenVMS 6.1, DEC Open3D 3.3(?). (OpenVMS is the reason
for why we cannot use PowerStorm 4D40T,50T,60T).

A hint on why our large application slows down the PowerStorm 4D20, anyone?

Any help is appreciated, thanks.
Anders Hjelm


 
 
 

OpenGL performance problems on DEC boards

Post by Ben Cheatha » Wed, 08 Jan 1997 04:00:00



<snip>

> A hint on why our large application slows down the PowerStorm 4D20, anyone?

Could you simply be PCI limited?  In otherwords, once you hit the PCI
limit (because you are drawing lots more triangles) both boards start
to perform equivalently.

--Ben

 
 
 

OpenGL performance problems on DEC boards

Post by Andy Vespe » Fri, 10 Jan 1997 04:00:00



> Our application runs on DEC AlphaStations. For the graphics, at a
> reasonable price/performance level, we use the DEC ZLXp-L1 board (Hardware
> acceleration of OpenGL, but not of textures, quite old board), but for new
> installations DEC wants us to use the PowerStorm 4D20 (No hardware
> acceleration at all), since on fast machines (like the AlphaStation 500)
> the 4D20 board performs as good as or better than the ZLXp-L1, and 4D20 is
> much cheaper.

The 4D20 board (some rendering acceleration) performs as well as the
ZLXp-L1
(more acceleration) because the power of the main CPU is applied to the
graphics processing.

Quote:> If we use a small test program just doing the display described above, the
> two boards are roughly equal in performance (as a matter of fact the 4D20
> is much faster in the wireframe mode; when the texture image is included
> the ZLXp-L1 is faster since the CPU does the texture at the same time as
> the board does the rest).

I'd guess that this test program uses very little CPU power for itself.

Quote:> But within our large medical application the 4D20
> performance is much, much slower than the ZLXp-L1 and much slower than for
> the stand alone test program, eventhough everything else in the
> configuration is the same!

Does your application use up a fair amount of CPU?
MONITOR/PROCESS/TOP_CPU
should show what processes are using the CPU.

Quote:> Furthermore, we are not able not obtain a double buffered 24-bit visual on
> the 4D20, only a 12-bit double buffered or a 24 bit single buffered one.

The Open3d version 3.3 release notes mentions a flag named
Export24BitDoubleBufferVisualOnZLXE that causes the server to export
a 24-bit double buffered visual. On VMS this gets put in the file
SYS$MANAGER:DECW$XSERVER_O3D.CONF. This should work for the 4D20 but I
don't believe it is supported or documented for that device. It makes
buffer swapping more costly as it is now a copy rather than changing
which buffer is displayed.

Quote:> We are running under OpenVMS 6.1, DEC Open3D 3.3(?). (OpenVMS is the reason
> for why we cannot use PowerStorm 4D40T,50T,60T).

This will, unfortunately, limit your options in the future as well as
now.

--
Andy V (A Digital spoke, not a Digital Spokesperson)

OpenGL Alpha Geek

 
 
 

1. DEC->SGI OpenGL display problem

  Hi,
I'm trying to display the OpenGL output of our application from DEC Alpha
machine to SGI. It works just fine locally on both platforms as well as from SGI
to DEC ( display on DEC, process on SGI ). But displaying on SGI from DEC causes
GLXBadContextTag error and nothing ever shows up in the window. Sometimes it
would even cause the Xsgi server to crash. I traced it down to glXUseXFont call:
without this call everything else works. I also found that the example program
xfont.c which comes with the OpenGL ( Open3D ) and contains that call also has
the same problem ( so it's not just me.. ). I hoped that the upgrade from Open3D
2.5 to 2.6 would fix this, but it didn't help. I'm not sure whom should i
contact, so i would appreciate if someone from DEC or SGI comments on this..
 Thanks in advance.


2. quality graduations

3. problem with DEC Alpha, OpenGL, and threads

4. Collision Detection

5. Performance Problems with OpenGL on WinNT

6. FREE Filters available

7. OpenGL line vs. quad performance problem

8. Audio and Ulead DVD Workshop

9. Performance of Mesa on dec-alpha

10. FS: DEC Powerstorm 4D30T /31MB 3Dvideo board

11. WTB : OLD DEC BOARDS

12. STB Lightspeed 128 video board..........Hows the Performance?????

13. mother board performance question(Dual vs. Quad)