💾 Archived View for gemini.spam.works › mirrors › textfiles › internet › FAQ › 3dcfaq1.txt captured on 2022-03-01 at 15:43:10.
-=-=-=-=-=-=-
========================================================================
PC 3D Graphics Accelerators FAQ
v 0.6
9 August 1995
========================================================================
<http://www.cs.columbia.edu/~bm/pc-3dcards.txt>
<ftp://ftp.cs.columbia.edu/pub/bm/pc-3dcards.txt>
Changes and additions to:
Blair MacIntyre
bm@cs.columbia.edu
Work:
+1 212 939 7087
Columbia University, Dept. of Computer Science
500 W 120th St., Room 450
New York, NY 10027
USA
This document provides a partial -- and not comprehensive -- list of
3D graphics accelerators for PCs. In particular, I am interested in
PCI-based accelerators. You can help make it more comprehensive by
sending me additional information and/or updates. This information is
abstracted and condensed from the posts of many different
contributors. No guarantees are made regarding its accuracy.
This document is available via WWW. See above URL.
The original motivation for this document came from a list of
accelerators compiled by Chris Hinch, Management Information Systems,
Dunedin City Council, PO Box 5045, Dunedin, New Zealand (chris@dcc.govt.nz)
----------------
Changes since version 0.5
Q02) added pointers to the OpenGL home page and the Viewperf ftp directory
Changes since version 0.4
Changed Entries:
GLINT04) Fixed area code in telephone number
Changes since version 0.3
New entries:
B07) Division's VPX Image Generator and ProVision Systems
B08) Reality Simulation Systems Little Squirt
Changes since version 0.1
New entries:
Q02) How does the performance of these cards compare to graphics
workstations such as those produced by SGI, Sun, DEC, etc?
CS03) S-MOS SPC1500 Geometry Processor
B05) Oki TrianGL
B06) Intergraph GLZ Series
Changed entries:
CS01) Changed address of www site and added email contact.
GLINT01) added pricing information
B01) Historical correction.
B04) Small change (VLbus parts may not exist)
---------------
TODO:
- organize the information a bit better, remove some extraneous or
duplicate info.
- answer some frequently asked questions
- get more hard data comparing the boards, such as the output of the OpenGL
viewperf program. If any manufacturers want to send me a demo
unit, or donate one to our group, my address is above.
- add more info, as people request and provide it!
==================================================================
Contents of Part 1
Introduction
Disclaimer
Section I: Frequently Asked Questions
Q01) Are there any 3D graphics accelerators for PCs?
Q02) How does the performance of these cards compare to graphics
workstations such as those produced by SGI, Sun, DEC, etc?
Section II: 3D Chipsets Availability
CS01) 3DLabs GLINT 300SX
CS02) Lockheed Martin Real3D (not yet available)
CS03) S-MOS SPC1500 Geometry Processor
Section III: GLINT 300SX-based Boards
GLINT01) Omnicomp 3DEMON
GLINT02) Fujitsu Sapphire 2SX (?)
GLINT03) ELSA GLoria (not yet available)
GLINT04) Media Labs Inc. 3DMedia
GLINT05) SPEA FireGL (not yet available)
GLINT06) Force Inc. 3DE-300SX (?)
GLINT07) Densan PCI-300SX (?)
Section IV: Real3D-based Boards
R3D01) Lockheed Martin R3D/100 (not yet available)
Contents of Part 2
Section V: Other boards
B01) AccelGraphics AG300
B02) ISC PowerGL
B03) Evans & Sutherland Freedom (not yet available)
B04) Matrox Impression Plus
B05) Oki TrianGL
B06) Intergraph GLZ Series
B07) Division's VPX Image Generator and ProVision Systems
B08) Reality Simulation Systems Little Squirt
==================================================================
Introduction
The original motivation for compiling this list was to compare
the features of various 3D graphics accelerators becoming available
for microcomputers. In particular, I am interested in PCI-based
cards which support OpenGL under Windows NT, Linux, Solaris x86
(none yet!), etc. and support stereo head mounted displays.
(In fact, of the currently available boards, only one qualifies.
See if you can figure out which one.)
Information which falls in that category will obviously be more
complete. Other information will be included if someone sends it
to me, but I won't be actively researching it myself. As such,
there is little or no information right now about 3D accelerators for
high-end machines such as those produced by SGI, Sun, HP, DEC,
etc, except as the ones listed may also work in those machines.
==================================================================
Disclaimer
I haven't seen or used many of these cards, and all figures are the
manufacturers own, as near as I can verify. I am not promoting
any one card or company over any other. Any anecdotes, speed
comparisons, etc., that I collect will be included on a purely
informational level. USE AT YOUR OWN RISK. Of course,
I will not include anything I do not believe or know to be false.
No representation is made as to the accuracy of these
figures/technical specs. If you are going to buy/bet the
farm/otherwise invest in one of these products, CONTACT THE
MANUFACTURER first and make sure that what you want is what you'll
get!!!
CAVEAT EMPTOR!!!
All trademarks etc are copyright of their respective owners etc. etc.
Furthermore, many of these boards are not yet available, so
you should find out delivery dates before getting your heart set on
any given one.
====================================================================
Section I: Frequently Asked Questions
====================================================================
Q01) Are there any 3D graphics accelerators for PCs?
I'm glad you asked. There are quite a few, with more coming out all
the time. See the rest of this document for more information.
(Ok, it's a stupid question, but I needed _something_ in the FAQ
section)
Q02) How does the performance of these cards compare to graphics
workstations such as those produced by SGI, Sun, DEC, etc?
This question is difficult to answer because there are so many factors
affecting the overall 3D performance on a PC. Most importantly, while
many of the boards listed below have impressive numbers in terms of
maximum number of polygons/second, most only accelerate rendering.
What this means is that you are still bounded by the speed at which
your processor can transform a 3D data stream to a 2D perspective with
lighting and material effects and then ship these numbers over the bus
to the graphics card. So, the processor, processor speed, bus type
and speed, software design, etc., all effect the final throughput.
Thus, a board may perform differently on two different machines that
have the same "specs" (ie. 90Mhz Pentium, PCI bus) if the bus was
designed differently, or if there are other cards on the bus taking up
bandwidth.
What is really needed is a set of comprehensive benchmarks that give
some more realistic numbers about the various combinations of
hardware. The NCGA (National Computer Graphics Association) GPC
(Graphics Performance Committee) (comprised of the major vendors along
with UNC and SDSC) is an organization primarily concerned with
developing benchmark standards and performance characterizations for
graphics hardware.
Check out:
http://sunsite.unc.edu/gpc/gpc.html
for information about the GPC activities.
Unfortunately, the GPC Quarterly is about $200/year, which is beyond
my budget. Thus, to get their benchmark results, you'll have to talk
to them. However, the benchmarking tool they use to measure OpenGL
performance, called Viewperf, is available for free. From the README
that comes with Viewperf:
What is this Thing Called 'Viewperf'
-----------------------------------
Rob Putney, OPC Vice Chairman
4/21/95
Viewperf is a portable OpenGL performance benchmark program
written in C. It was developed by IBM. Viewperf provides a vast amount of
flexibility in benchmarking OpenGL performance. At the current time,
the program runs on most implementations of UNIX, Windows NT, and OS/2.
The OpenGL Performance Characterization (OPC) Committee has endorsed
Viewperf as its first OpenGL benchmark and results
were first published using it in 4Q94 version of The GPC Quarterly.
OPC member companies have ported the Viewperf code to their operating
system and window environments. The OPC Committee maintains a single
source code version of the Viewperf code that is available to the
public.
In future versions of this FAQ, performance measurements will be added
in another section at the end. Stay tuned. If you have any
measurements using Viewperf, please send them to me (bm@cs.columbia.edu)
Viewperf can currently be obtained from
ftp://net1.uspro.fairfax.va.us/pub/gpc/opc/viewperf/
If that directory does not work, try looking on the OpenGL home page at
http://www.sgi.com/Technology/openGL/opengl.html
which should contain a pointer to ViewPerf.
====================================================================
Section II: 3D Chipsets available now or in the near future.
There are numerous 3D chipsets being worked on, but these 2 happen to
be the ones I have detailed information both on the chips and on
boards built using them. If I get more information on other chips,
I will include them here.
====================================================================
CS01) 3DLabs GLINT 300SX
First, a comment about the GLINT 300SX chip, which I will make because
it was not clear to me when I first read their specs. They mention
texture mapping acceleration, but on the 300SX texture maps are held
in host memory and per pixel information is sent to the chip which
performs the necessary interpolations and applications. For the
upcoming 300TX, the texture maps are held in local memory on the
graphics card (the "localbuffer") and the host just sends per vertex
information to the chip which performs all the texture mapping
calculations.
Information on the 300SX can be obtained from the 3DLabs home page at
http://www.3Dlabs.com/3Dlabs/
or by sending email to info@3Dlabs.com.
The technical specs below were copied directly from the page
http://www.3Dlabs.com/3Dlabs/300SXinf.html
GLINT 300SX Overview
The GLINT 300SX high performance graphics processor combines workstation
class 3D graphics acceleration and state-of-the-art 2D performance
in a single chip.
GLINT is capable of processing 300,000 shaded, depth buffered
and anti-aliased polygons/second. The chip provides complete 32-bit
color, 2D and 3D acceleration, an on-chip PCI-compliant local
bus interface and integrated LUT-DAC control, making a complete
graphics subsystem possible with minimal chip count.
All the rendering operations of OpenGL are accelerated by the
GLINT 300SX, including Gouraud shading, depth buffering, anti-aliasing,
alpha blending, and texture mapping.
3D ACCELERATION
o 100% OpenGL compliant rendering operations in hardware
o Point, line, rectangle and polygon primitives
o Gouraud shading, depth buffering, anti-aliasing, dithering,
depth cueing, texture map filtering and alpha blending
o 2.5 Giga operations per second
o 300K Gouraud shaded, depth buffered triangles/sec
2D AND GUI ACCELERATION
o True color acceleration of window systems such as Win32 and
X11
o Accelerated bitBLT, line drawing, fills, text and window clipping
SPECIAL FEATURES
o 64-bit hyper-pipelined architecture
o Vertex level interface
o 112-bit pixel datapath to memory
o High quality 16 and 8-bpp dithering of 24-bit graphics images
o Fast frame and depth buffer clears
o Shared framebuffer interface for easy multimedia integration
o VRAM block fill and bit masking
o Anti-aliasing for high quality images at 4x4 and 8x8 sub-pixel
resolution
DISPLAY MODES
o Resolutions up to 2560x2048
o 8, 16 or 32-bits per pixel RGBA and 4 or 8-bit color indexed
o Packed 8, 16 and 32-bit pixels
o Supports advanced modes: double buffering, stereo and overlays
GLINT 300SX Graphics Architecture
The GLINT 300SX delivers 3D and GUI rendering functionality to
improve the performance and quality of graphics applications and
APIs.
o Gouraud shading
o Depth buffering
o Anti-aliasing
o Dithering
o Depth cueing and fog
o Texture mapping
o Scissor and stipple masking
o Alpha blending
o Logical Operations (e.g. XOR)
The unique hyper-pipelined architecture and 112-bit memory datapath
ensure that multiple rendering operations can be combined with
no performance penalty - allowing a Gouraud shaded polygon to
be processed just as fast when depth buffering, dithering and
anti-aliasing are enabled.
FRAMEBUFFER CONTROLLER
The 64/32-bit framebuffer controller directly interfaces to VRAM
or DRAM framebuffers and supports a wide range of pixel depths,
resolutions and memory sizes - enabling systems to be tuned to
meet specific price/performance targets.
O 1-32 MBytes VRAM or DRAM
O 8,16 or 32-bit RGBA color
O 4 or 8-bit color indexed
O 2, 4, 8 pixels per 64-bit word
O 640x480 to 2560x2048
O VRAM arbitrary shape block fill
O VRAM bit masking
O Framebuffer bypass mode
LOCALBUFFER CONTROLLER
The localbuffer holds optional off-screen information for
each pixel including; depth (Z) values, stencil data, arbitrary
window clip planes, plus a control field for fast window clearing.
The format for each field is user defined, up to a maximum of
48 bits.
O 0-48 MBytes DRAM
O 0-48-bits wide
Depth: 0, 16, 24, 32-bit
Stencil: 0, 4, 8-bit
Window clipping: 0, 4-bit
Fast depth clear: 4, 8-bit
O Localbuffer bypass mode for direct access by external devices
INTEGRATED VIDEO TIMING
An internal timing generator and VRAM transfer controller
are incorporated into the GLINT 300SX processor. Advanced systems
requiring overlays, stereo, interlace or very large framebuffers
can be supported using an external timing generator.
SHARED FRAMEBUFFER
The GLINT 300SX's on-chip shared framebuffer interface is
a fully integrated implementation of S3's interface, allowing
GLINT to use video and multimedia co-processors, or for the GLINT
to be used as a 3D coprocessor to an S3 device.
UPGRADE PATH
The GLINT 300SX is the first of a family of GLINT graphics
processors that are designed to meet the demands of the graphics
designer. The GLINT300TX, is a second generation pin compatible
processor that enhances the SX's texture capabilities - supporting
local texture storage and pixel address generation for all OpenGL
texturing modes.
NOTICE: 3Dlabs reserves the right to change the above specifications
without notice.
From looking at the various boards, it appears that typical
performance quotes are in the following range:
3D Gouraud shaded, depth buffered, window clipped, stippled,
dithered, and alpha tested rendering performance with
24-bit RGB color depth is the following:
300K 25 pixel triangles/sec
500K 10 pixel vectors/sec
8M independent points/second
500K 9x13 pixels characters/second
2D Graphics performance
1M 10 pixel vectors/sec
500K 10x10 pixel, flat shaded rectangles/sec
500K 9x13 pixels characters/second
Scroll Rate 48M pixels/second - 8 bits/pixel
====================================================================
CS02) Lockheed Martin Real3D
Information on the Lockheed Martin Read3D chipset was obtained from
their WWW server at the following URL:
http://www.mmc.com/real3d/
Check there for more information, or call 1-800-393-7730, or send
email to real3d@mmc.com
The following is taken from a press release found at that site:
The REAL3D(tm) chipset is based on real-time computer image generation
technology that gives a combination of dynamic response and realism
previously available only on dedicated graphics workstations and high
end custom image generators.
The first product will be the R3D/100 graphics accelerator. Key
performance attributes of the new chip set include an embedded 100
MFLOPS geometry processor, pixel write rates of up to 33 million
pixels per second, up to 750,000 polygons per second, line processing
up to 1.5 million per second, and provides up to 192 color texture
maps (128 x 128 mipmapped) in real-time.
The new product is an outgrowth of Lockheed Martin's proprietary
computer graphics technology previously used in high performance
military simulation, engineering research and training applications
first developed for astronaut training and military flight
simulators.
The chip set provides faster processing through its patented hardware
design which incorporates geometry processing, rasterization and
texture mapping. The R3D/100 embedded floating point geometry
processor removes significant processing burden from the host CPU. The
patented texture processor applies color mipmapped texture to polygons
in true 3D corrected perspective.
Designed as a true polygon processor with texture processing and
scaleable texture memory from the outset, the R3D/100 chip set
includes dedicated hardware acceleration of mipmapped texturing that
provides continuous high fidelity image quality. This chip set
simulates spotlights, fog and realistic curved surfaces. Additionally,
improved image quality is provided with multi-pass anti-aliasing.
The R3D/100 chip set directly interfaces with Microsoft 3D/DDI and
supports all 3D/DDI-compliant APIs, such as OpenGL(tm) and comes with
device driver software and a device driver kit.
====================================================================
CS03) S-MOS SPC1500 3D Graphics Geometry Processor
S-MOS Systems is offering the industry's first commercially
available single chip 3D geometry accelerator for the PC marketplace.
The SPC1500 will work together with any 3D rendering processors to
accelerate 3D graphics up to 300K+ lighted triangles per second.
Why is geometry processing acceleration important?
Rendering processors alone, even on Pentium class machines, can not
deliver true workstation class performance. This is because even the
fastest CPUs today are quickly overcome by the amount of algorithmic
processing necessary to transform a 3D data stream to a 2D perspective
with lighting and material effects.
The SPC1500 integrates a 150Mflop dual floating point engine, an
on-chip DMA processor, on-chip ROM to store the algorithm instruction
set, and an on-chip RAM cache to store customer instructions in a
single 223 pin PGA package.
Easy integration
S-MOS Systems provides a complete design package for the system
integrator and board vendor.....
PCI Reference Designs
S-MOS Systems provides reference designs and evaluation kits showing
the capabilities of the SPC1500 running OpenGL on the WindowsNT
platform. Full technical manuals, electronic gerber files, and
applications notes are available.
Full Software Support
WindowsNT and OS/2 Support
S-MOS Systems provides OpenGL compliant software and system utilities
for the WindowsNT platform. This software supports any rendering board
(2D or 3D) that supports 3D DDI. It will also supports direct data
transfer to the leading 3D rendering board vendors.
MacOS Support
S-MOS Systems will will support software drivers and system utilities
for Quickdraw3D.
An Acrobat version of the SPC1500 Datasheet is available via their page at
http://www.smos.com/standard/spc1500.htm
For more information contact
Giri Venkat, Product Marketing Engineer, S-MOS Systems
(E-Mail: venkat@smos.com).
====================================================================
Section III: GLINT 300SX-based Boards
The descriptions of the boards in this section will attempt to give
information not common to all GLINT 300SX-based boards.
====================================================================
GLINT01) Omnicomp 3DEMON
FROM THE SPEC SHEET:
3DEMON 3-D Graphics Accelerator for PCI Local Bus Technical Brief
The 3DEMON provides workstation class 3-D graphics and high
performance 2-D acceleration using the GLINT (tm) 300SX 3-D processor
from 3Dlabs.
Frame buffer: