Michael T. Jones (mtj++at++babar.asd.sgi.com)
Mon, 29 Apr 1996 15:31:16 -0700
Hi John!
> Thanks for your thoughts and comments. I felt it best to not go on-line
>with a response to your Performer newsgroup as this discussion is not a
>Performer related issue, per se and I don't like to violate the spirit of that
>forum (e.g., Performer focus). As such, I'd assume SGI would prefer to keep
>this off-line from info-performer; if you don't care, you're certainly free to
>forward it there if you want!
Well, considering how little your long OpenGVS advertisement had to do with
Performer, that we have nothing to hide about our performance, and that the
Performer users are probably curious to learn of the relative performances
between various IG's and software environments, I'll go ahead and forward it
since you say it's OK. Perhaps there are non-Performer users on the mailing
list who will speak-up and offer to implement your test databases on their
machines/systems and an interesting dialogue will result.
> Anyway, some clarifications for you regarding your recent email. You wrote:
> > ...The Gemini IG test suite will serve as an interesting benchmark since
> > it's open, in the following ways:
> :
> : 1. It can be run on multiple machines using OpenGVS. It would be a
> : way to verify and judge implementation and hardware. This is the
> : test that a GVS user would want to make, since it is a measure of
> : virtue for the range of available options.
>
> Comment: the benchmarks are expected to be of interest to both GVS users and
>basically *any* other interested party (user, hardware manufacturer, whatever)
>since these benchmarks (as you pointed out) will help provide another useful
>"data point" as to the potential suitability of the underlying graphics
>hardware for computer image generation use.
We see this point very differently. The GVS versions only speak to GVS speeds
on the machine. They do not measure the machine. That would be like a ISV PHIGS
company ranking workstations on drawing a car model based on their PHIGS
implementation. The fastest one would be important to that ISV's PHIGS users,
but no one else (not even other PHIGS users), unless...
>
> : 2. It can be run using multiple software libraries (OpenGVS, dVS,
> : VistaWorks, SENSE8, Performer, etc.) on the same machine and
> : database. This is what an SGI user would want to test since it
> : is a measure of virtue for the range of available options.
>
> Comment: we only run on top of OpenGL, Direct3D, or the manufacturer's native
>rendering API for those graphics implementations that support neither natively
>(e.g., Real3D, 3DFX). Thus, to clarify, we do *not* run on top of Sense8, dVS,
>Performer, or any other higher level APIs as I believe you suggested. Any
>hardware manufacturer that uses a DLL under Windows NT or Windows 95 will be
>able to run the binaries as long as the DLLs have been installed on their PC
>(DLLs are much like SGI's shareable libraries).
...the test environment is as I thought you said before, an open one. This
means that there is some database, and some task, where there as a database in
OpenFlight, DWB, Wavefront, IV, or some other public format and some set of
moving models with a log file of positions for each with time stamps
(eyepoints too). There is no need for GVS code, for binaries, etc. Just the
databases and example information so that people porting the tests to other
environments can make the test measurements fairly.
Gemini would implement these tests in OpenGVS (this is test #1) but I would do
it in Performer. Companies like SENSE8, Loral, etc. would of course do it in
the software environment they produce. This is test #2. Without test #2, your
"universal testbed" is merely an OpenGVS test suite and is of no interest to me
or Silicon Graphics. With test #2, your work becomes valuable in that the
results from test #1 are validated -- a user knows much weight to give to a
GVS-mark based on how well GVS does on that machine compared to other toolkits.
Based on the GVS-mark, users can extrapolate performance of machines or
alternate toolkits by multiplying by the speed of the faster or chosen toolkit
by the GVS one on the same machine/database benchmark pair. (The factor would
be 1 of course everywhere that OpenGVS is the fastest, so the results will be
good for marketing for at least one of the companies participating in the
benchmarks.)
I have accepted the head-to-head test that you proposed. Are you backing out of
it now? I don't see why you would do that after making such a big splash in
the Performer mailing list with your original announcement.
> You wrote:
> : ...I'm eager to get the Gemini Technology test suite and
> : have the Performer team get it up and running. If we can get it here
> : at SGI in time, we can have numbers for test #2 in time for John
> : Archdeacon's presentation at the Image Society meeting in June.
> : When can we get the software and databases, what are the methods by
> : which timings should be taken (resolution, #samples, etc.) and what else
> : do we need to help you compile this data?
>
> Comment: I plan to run the tests under IRIX 6.2 on a 250 MHz Onyx RE2 with
>two RM4 boards (the baseline re2stone as you may recall). At last check, I
>don't believe the LA, Irvine, or San Diego offices of SGI have such a system
-- >can you arrange to loan us one through Image? If not, I plan to use one of my >customer's systems (which does have the baseline configuration) but your >support in this area would help alot if possible. Second, I do plan to publish >the binaries on-line at our home page in a compressed TAR file format for IRIX >users as well as appropriate versions for Windows NT and '95 users as well. >I'll keep you posted as to when the IRIX version is available (sooner if you >can help us get the right hardware here... if not, a trip to my customer's site >or to Mountain View!)Are we talking about the full, open IG test approach (tests #1 and #2) or the OpenGVS-only approach that you seem to retreating to? If it's the real thing then I think I need to find a way to help you since it could clear up so many misconceptions.
> Thanks - John > > >-- End of excerpt from John Archdeacon
I hope that I misunderstand your new position. I urge you to follow through with the plan to make the databases and eyepoint/vehicle information avaliable as your test suite. Just printing a list of OpenGVS throughputs and claiming that you've measured the soul of InfiniteReality, CompuScene-VI, ESIG-4000, and so on is not much to base in IMAGE presentation on, nor is it something that SGI, Lockheed-Martin, or E&S would seem likely to embrace (speaking for myself, not for other image generator vendors).
Michael Jones
--Be seeing you, Phone:415.933.1455 Fax:415.965.2658 M/S:8U-590 Michael T. Jones Silicon Graphics, Advanced Systems Division mtj++at++sgi.com 2011 N. Shoreline Blvd., Mtn. View, CA 94039-7311 "Du musst Amboss oder Hammer sein" -- Goethe
This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:52:49 PDT