The 3D Chipset Wars: A Chronicle of the Past, Present, and Future
-- Nicholas Paufler
It’s been a long time since a piece of hardware, other than CPU’s, RAM, and
hard drives, has developed as rapidly as modern video cards. What’s fueling this
rapid advance in technology? Games, pure and simple. Not since the day the original Wing
Commander was released has there been such a rush to adopt new hardware. In The early
90’s, the advent of true sound spawned dozens of competing standards. There
was no marketing hype, no benchmarks, just the promise that games could really only be
experienced in glorious 8 bit sound. Roland, AdLib, and yes, even the SoundBlaster
redefined how we entertained ourselves on our PC’s. Today, however, a new battle for
supremacy is being fought, but it is not aural stimulation (although that battle is
starting to heat up again, as well…), but rather … THE 3D CHIPSET WARS.
I can remember my father telling me about the time when he was working in a hospital,
and the classic game Zork became available for PC. Suddenly, everyone decided they
needed a PC in their office to get their work done. This was almost 20 years ago, and the
masses were enthralled by the white house, the mailbox, and the infamous grue, all
rendered in glorious … text. The CGA (Color Graphics Adapter) standard first brought
color to home PC’s. With only 4 colors to work with, true greatness was not realized
until EGA (Enhanced Graphics Adapter), and with it, a huge palette of 16 colors.
King’s Quest, Space Quest, Quest for Glory, and their plethora of sequels were all
initially developed for this standard. Shortly after, VGA (Video Graphics Array) brought
about a revolution. A palette of 256 colors (8 bit) meant video was finally possible, and
graphics were better than ever. Finally SVGA, or Super VGA brought color depths of 16bit
(65,535 colors), and 24 bit (16.7 million color). Full motion video with TV quality and
photorealistic images were only the beginning. And then, for the most part, it all
stopped.
16.7 million colors are far more than the human eye can differentiate between, so it is
unlikely that any higher color depths will be adopted. Video card manufacturers developed
faster RAMDAC’s (Random Access Memory Digital-to-Analog Converter), which were able
to display higher resolution images, as well as integrated additional RAM, but little more
could be done. In all likelihood, the world would have continued on its merry way, with
nary a second thought, save for a little game called Quake. When iD Software
released this instant classic in 1996, frag fests erupted all over the world. It would be
inaccurate to say that it was Quake alone that spurred the development of 3D accelerators,
but it did play an integral role. All of this real time 3D geometry had engineers
pondering how they could allow for greater speeds. Professional 3D modelers and CAD
designers already had cards that offloaded rendering tasks from the CPU, but these were
primarily rendering still images and were prohibitively expensive.
What the world needed was a card that could both accelerate rendering tasks and be
cheap enough to be accessible to the average consumer. Just like any other new
development, the road to victory was littered with failed attempts. Can anyone remember
the Cirrus Logic Laguna 3D? No? Didn’t think so. These first generation 3D
accelerators barely made a noticeable difference, as they were most often handling all of
the rendering tasks via software drivers, not exactly an efficient solution. There were a
few moderate success stories along the way, but only one truly stands out, as its name is
legendary. The 3Dfx Voodoo. Unlike many of their competitors, 3Dfx only
manufactured the chipset, and produced a set of reference designs and drivers, which they
supplied to other manufacturers, such as Diamond and Canopus. These other companies
released Voodoo based cards such as the Monster 3D and the Pure 3D. Here were cards anyone
could throw in their computer and have arcade quality graphics for a nominal price. It was
an instant hit, and the Voodoo chipset became the de facto standard for 3D gaming. While
the Voodoo was an attractive card to consumers, it was developer support that solidified
its dominance. Microsoft’s Direct3D API was still in its infancy, and the OpenGL API
had not yet been adapted for gaming. 3Dfx’s masterstroke was their own proprietary
API, Glide. Programmers loved it, as it was brilliantly written and allowed for rapid
development, yet was both robust and fast.
For all its success, the Voodoo chipset was only designed as an add-in card, meaning it
required an existing 2D video card to function. In an attempt to integrate their Voodoo
chipset with a 2D core and develop a 2D/3D combination card, 3Dfx released the Voodoo
Rush. In the first of many mistakes to come, it was declared to be a failure by gamers and
developers alike. While it was similar to the Voodoo, there were serious differences. The
Voodoo Rush was noticeably slower, and games that fully supported the Glide API often
required patches to make them function with a Rush. All the while, other hardware
manufacturers were entering the fray. Canada's own ATI released many video cards, such as
the Rage, Rage II, and Rage Turbo, but all failed to catch on. While ATI bragged about
synthetic benchmarks, the question was raised as to the validity of present day 3D
benchmarking suites. While tests often showed ATI’s offerings to be faster on paper,
in real world applications (read: games), Rage based cards were soundly trounced by the
Voodoo. The dominant force in the business and graphics development field was also a
Canadian based company, Matrox. Their Millennium and Millennium II cards were expensive,
but well suited to applications such as Photoshop, where their high refresh rates and
resolutions gave graphics designers greater flexibility. They were, however, useless for
games. There first attempt was the Matrox Mystique, which was not nearly as fast or as
feature rich as the Voodoo, yet still experienced a little success. The little known
Matrox m3D followed, but was a huge flop. The Matrox advertising campaign preached that
framerate (the number of times per second the screen is updated) was everything, and that
it did not matter if image quality suffered, or certain features were not present.
Unfortunately, framerate is not everything, and lacking key features that its older
competitor offered, it never caught on.
Arguably, it can be said that the true second generation of accelerators came about
around the release of 3Dfx’s Voodoo 2 chipset. Offering higher resolutions,
significantly greater speed, and a new feature called SLI (ScanLine Interleave), 3Dfx had
another hit on their hands. What truly set the Voodoo 2 apart from its competition was its
SLI mode, which allowed a user to put two Voodoo 2 based cards in their computer, connect
them with a cable, and nearly double the speed at which they could play. SLI works by
having each card render alternate scanlines, so each card only needs to render half of the
screen. While it was prohibitively expensive at first, costing upwards of $600 for a pair
of 12 meg Voodoo 2’s, power gamers snapped them up. Around this time, the second
generation of competitors finally began challenging 3Dfx’s supremacy. Intel (the CPU
manufacturer) entered the fray with its i740 chipset (which was based on the Lockheed
Martin chipsets), which became a popular budget video card, but never truly found
acceptance amongst hardcore gamers. A company called nVidia followed 3Dfx’s example
of developing only the chipset, the Riva 128, and selling the chipset to other companies
to produce the actual card. As it turns out, nVidia will become a key player later on, but
even the Riva 128 (and its later, enhanced variant, the Riva 128ZX), posted encouraging
sales figures. Even though they had minor graphical display issues, the value for the
dollar was excellent. The Matrox G100 came and went, and the ATI Rage Pro was released,
but in games only performed slightly better than the older Voodoo 1 chipset.
We’ve now made our way to the middle of 1998. Matrox followed up its G100 with the
G200, in several models. The Millennium G200, the Mystique G200 (which offered TV out, but
was slightly slower than the Millennium), and the Marvel G200 (Video-in, TV-Out, and
obscenely expensive). TV out is a feature that allows a user to connect their computer to
their television and use the TV as a monitor. Conversely, TV-in is video capture, allowing
the user to digitally record video onto their computer. While its speed was good for the
time, it was surpassed in the coming months by others, but what set it apart was the
beautiful image quality which is even today, unmatched. Hindered by the lack of an OpenGL
ICD (Installable Client Driver, which allows the hardware to interface with the OpenGL
API. However, Matrox has yet to be release the final version of their ICD), and middling
3D performance, Matrox realized they had made some excellent advances, but still lagged
behind the competition. Next on the scene, almost simultaneously, were 3Dfx’s and
nVidia’s next generation cards. The Voodoo Banshee was 3Dfx’s second attempt at
a combination 2D/3D card, and it nearly succeeded. It benchmarked fairly close to a single
Voodoo 2, but did not gain widespread acceptance for several reasons. First, it was based
on the Voodoo 2 core, which is limited to 16 bit rendering (65,535 colors), versus the 32
bit rendering which its competitors were implementing. Secondly, it only had one texture
unit, as opposed to two on the Voodoo 2, which slowed it down in any game that implemented
multitexturing (that is, rendering more than one texture over a single polygon, like a
texture and a light map), such as Quake 2. Finally, all other current generation cards
could be found in both the traditional PCI (Peripheral Component Interconnect) interface,
as well as the newer AGP (Accelerated Graphics Port), which was both faster, and was able
to utilize system RAM to store textures, making games with large textures significantly
faster. The Banshee came in an AGP flavor, however it was not a full AGP implementation,
and could not take advantage of features like AGP texturing, essentially defeating the
purpose of AGP entirely. After nVidia's success with the Riva 128 it seemed inevitable
that their next generation chipset would seriously challenge 3Dfx's dominance. Their Riva
TNT chipset has completely dominated the 3D market since its release in mid-1998. It
was everything the banshee was not, it had 32 bit rendering, a full AGP implementation,
better image quality, and was noticeably faster, yet it cost only slightly more. The only
card that looked better were G200 based cards, and the only thing faster than it was a
pair of Voodoo 2’s in SLI, definitely a tasty combination.
Finally, we come to a breakdown of what is available now, and what is to come.
Matrox G200
Pro’s: The best image quality currently available; TV out is also
far superior to any of its competitors; The Marvel G200 with its video capture mode as
well as TV out is interesting for someone in the market for a card that can do it all; 32
bit color
Con’s: Still no final version of the OpenGL ICD (rendering it
useless to 3D modelers); Slowest of all current generation 3D accelerators.
Who should buy it?: Stick to either end of the model spectrum, for a
business workstation on a budget, the Matrox Millennium G200 LE is an attractive choice
(it has the speedy 250mhz RAMDAC, but is not upgradable, either with additional RAM or the
Video-in/TV-out add-ons Matrox offers). If you have money to burn and would like to play
with video capture, but also require solid 2D, and possibly a little 3D gaming on the
side, consider the Marvel G200.
Price: Millennium G200 LE 8meg AGP - $100
Marvel G200 8meg AGP or PCI - $375
ATI Rage 128
Pro’s: Solid 3D performance; full OpenGL ICD; 32-bit color
Con’s: ATI has never had true success in the retail market, the
majority of their cards are bundled with systems; the Rage 128 chipset runs VERY hot (up
to 91 degrees Celsius)
Who should buy It?: I would not recommend it to anyone, rather, it will
be nice for people buying packaged systems, and will finally have a real 3D accelerator
inside the box.
Price: Xpert 99 Rage 128 AGP 8 meg - $125
S3 Savage 3D
Pro’s: S3TC (Texture Compression) compresses textures leading to
faster framerates with little to no quality loss; above average 3d performance
Con’s: Some driver issues still remain; S3 has never had retail
success with its products, they have always been cheap, low end parts found in bundled
systems
Who should buy it?: No one. If you really want S3TC, wait for Savage4.
Price: Acer PA70 Savage 3D 8 meg AGP - $140
TNT
Pro’s: Excellent speed ; Very good image quality; 32-bit rendering;
full OpenGL ICD
Con’s: It’s already been out for over 6 months, an eternity in
the computer world, holding out for a TNT2 might be advisable.
Who should buy it?: Anyone who wants the best value for their dollar and
absolutely cannot wait for the next generation of cards
Price: Creative Labs Graphics Blaster Riva TNT 16meg AGP - $135
Asus V3400 Riva TNT 16meg AGP with TV-out and Video-In - $230
Voodoo 2
Pro’s: In SLI configuration, its speed is presently unmatched; Glide
API
Con’s: Since it is an add-on card, it can only perform 3D operations
in full screen; only 16 bit rendering; consumes extra PCI slots as you also require a 2D
card (and potentially another PCI slot for SLI)
Who should buy it?: Someone who already has one Voodoo 2 and a decent 2D
card and isn’t ready to trade them in yet.
Price: Maxi Gamer 3D II 12 meg Voodoo II - $200
Generic 12 meg Voodoo II - $150
Voodoo Banshee
Pro’s: Glide compatibility; Reasonably fast; good 2D performance
Con’s: 16bit rendering; slow multitexturing; already being phased
out
Who should buy it?: The only situation where getting one is justified
(rather than a TNT) is when upgrading a computer slower than a Pentium 233MMX, where a TNT
(which is CPU limited) would be wasted. However, with the marginal price difference,
purchasing one anyway and using it in a future upgrade might be a better solution
Price: Creative Labs 3D Blaster Banshee 16 meg AGP- $140
What’s Coming…
Matrox G400
The G400 has two features that will set it apart from the competition. The first is
true hardware bump mapping, which can make a huge difference in an image’s realism. A
technical demo for the G400 showed a lake with rippling waves, rendered using only 4
polygons thanks to bump mapping (a bump map, for example, might be used on a flat wall
painted with a brick texture, where using a bump map to show the depressed cement around
the edges of the bricks can add to the realism). The second feature is far more innovative
and is bound to sell cards based on this feature alone. It is known as Dual Head, and will
allow you to hook two monitors, a monitor and a TV, a monitor and an LCD monitor, or a
monitor and an HDTV, up simultaneously, and run them at independent resolutions and
refresh rates. Imagine, outputting to a TV without reducing the refresh rate on your
monitor to 50hz to compensate, or designing a web site under 1024x768, but having a second
monitor running at 640x480 to test the lowest common denominator. The best part is, Dual
Head will be implemented in all G400 products, and are expected to cost less than USD$150.
If Matrox is correct in their promises that the shipping G400 will be 3 to 4 times faster
(possibly twice as fast in terms of actual framerate), and a final OpenGL ICD ships, this
card is poised to turn a lot of heads.
Voodoo 3
Presently, it appears that this card will be a miserable failure. It will come in 3
versions, each faster than the other, and costing correspondingly more, with the high end,
the Voodoo 3 3500, costing an astronomical USD$250! However, it barely outperforms two
Voodoo 2’s in SLI, and is still restricted to 16-bit rendering and the inefficient
implementation of AGP. 3Dfx is claiming that framerate is more important than the speed
sacrifice that must be made for 32 bit rendering (sounds like the m3D, doesn’t it?),
but preliminary tests are showing the nVidia TNT2 running in 32 bit color outperforming
or performing equal to the Voodoo 3 in 16bit color, and costing USD$50 less! It seems
that 3Dfx has truly dropped the ball, and it is no longer clear what place this card will
take in the marketplace, where a cheaper, faster, and better looking alternative exists.
Also worth noting is that 3Dfx recently purchased STB, a video card manufacturer who had
previously sold boards based on 3Dfx chipsets. It was 3Dfx's intention to make inroads
into the lucrative OEM (Original Equipment Manufacturer, in other words, prepackaged
systems like Compaq, Dell, etc.) market. However, in doing so they severed all ties with
companies like Creative Labs and Diamond Multimedia, who had been staunch supporters of
3Dfx chipsets up until that point. Essentiallty 3Dfx forced two companies with loyal user
bases even closer to the Riva TNT camp.
TNT2
nVidia also plans on releasing this card with multiple model lines, a low end and a
high end, to appeal to both budget minded, and power gamers. In terms of pure performance,
the TNT2 will be the card to beat in 1999, so unless one of its competitors can
offer significantly faster speeds (possibly Savage4?), cost noticeably less, or offer a
larger feature set (Matrox G400), then it seems nVidia’s reign will continue for
another year. Not that that is a bad thing, by any means.
And that wraps up this months article. We saw what was, what is, and what’s
coming, so hopefully you will be at least a little more informed when it next comes time
to upgrade your computer.
Example of hardware bump mapping
|
