home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!saimiri.primate.wisc.edu!ames!sgi!fido!tuolumne!akin
- From: akin@tuolumne.asd.sgi.com (Allen Akin)
- Newsgroups: comp.graphics.opengl
- Subject: Re: Software speed?
- Date: 11 Jan 1993 19:52:07 GMT
- Organization: Silicon Graphics, Inc. Mountain View, CA
- Lines: 51
- Message-ID: <1isj57INNp2d@fido.asd.sgi.com>
- References: <1993Jan8.174025.21761@eye.com>
- NNTP-Posting-Host: tuolumne.asd.sgi.com
-
-
- In article <1993Jan8.174025.21761@eye.com>, erich@eye.com writes:
- > Lines: 13
- >
- > I would like to know how fast the provided code for the OpenGL API performs.
- > That is, if you get a level 2 commercial license, how fast is the software
- > only version of the X11 sample implementation for shaded polygons? I'm
- > interested in figures on any machine - GPC type figures are fine, or any other
- > 3D illuminated polygon benchmark you guys at SGI might have (Bill Glazier, are
- > you still out there? You seem to be in the know).
-
- As Kipp said... The sample OpenGL implementation is intended to be
- functionally complete and readily customizable, but it isn't optimized
- to the extent that you could ship it as a product off-the-shelf.
- Therefore measuring its performance isn't very meaningful -- you'd be
- getting some measurement of the schedule pressures faced by its
- implementors :-), but not any measurement of the intrisic capabilities
- of the API or the software or the hardware on which it's run.
-
- FYI, the sample OpenGL implementation is being used as the basis for
- SGI's product implementations, so we're pretty confident that it
- doesn't present any unreasonable limits to optimization. Some of the
- changes we make for our products will be fed back into the sample
- implementation, so it will continue to improve. Perhaps the other
- licensees will contribute optimizations, too.
-
- Sorry we have to be this cautious. People do use performance numbers
- out of context, so it pays to avoid misunderstandings. Could you give
- us a clue as to what questions you're trying to answer? Perhaps that
- would help us provide more useful information.
-
-
- > I'm also interested in how this compares to SGI hardware speeds, i.e. how
- > much does various dedicated graphics hardware buy you vs. just using the raw
- > compute power of the machine and a software only render? 100x? 10x? 1.5x?
- > Does anyone else out there have any feel for this?
-
- Could you give me a little more context?
-
- As I understand your question, the answer depends pretty drastically on
- which model of graphics hardware and which graphics primitive (among
- other things) you'd like to use for comparison. I've seen cases in the
- past where adding graphics hardware yielded performance improvements of
- greater than 1000X (textured polygons) and performance degradation of
- 0.5X (image transfer).
-
- The general question of using dedicated hardware vs. general-purpose
- hardware, and the resulting cost tradeoffs for particular markets, is a
- very interesting one.
-
- Allen
-