3DFX: HISTORY AND FALL OF THE VIDEO CARD GIANT
Gordon Campell, Gary Tarolli, Scott Sellers and Ross Smith. These names probably won't say anything to most of you, but what if we told you that this handful of modern engineering Beatles founded a company that revolutionized the video card and 3D graphics landscape?
The 3dfx was officially created on August 24, 1994 by four
former engineers of Silicon Graphics, the same who was responsible for
designing and producing the graphics technology that was the basis of the
Nintendo 64 (in this article we recall what was its flagship game : Mario 64 ).
Silicon Graphics at that time was a hotbed of incredible
technologies, but its target were supercomputers and workstations and did not
see PCs as a possible source of income, despite the fact that the gaming video
card market was a fertile ground, not yet channeled into the current duopoly.
which sees AMD and NVIDIA as the only competitor.
The mid-nineties was a period of continuous innovation, just
think that in 1995 Microsoft bought a company called RenderMorphics, and that
their 3D API called RealityLab were integrated into Windows 95 becoming the
basis of Direct3D, which became the standard of modern gaming. on PC.
Meanwhile, productions like Doom and its debated fake 3D or 2.5D - as some
liked to call it - had shocked gamers all over the world by giving a taste of
what the future could offer, bringing PC gaming into the third dimension.
At that time Id Software was working on Quake and designing
an engine , Id Tech, ( now in its seventh iteration with Doom Eternal ) capable
of generating real 3D graphics - based on polygons - even on a modest Pentium.
Quake turned out to be a real "killer application" that helped arouse
enormous interest in 3D and its performances and unleashed a real "arms
race" on gaming PCs, to snatch a couple of extra fps on the seminal
masterpiece of id Software.
Voodoo magic
3dfx made its market debut during E3 1996, using the first
version of its Voodoo technology on an Arcade cabinet. Their first product was
called Home Run Derby , the aim of the game was to make a "Home Run"
by hitting a virtual baseball with a real bat: the proprietary technology
calculated by infrared time, inclination and speed of the stroke. Surely the Arcades
were a good entry point to present their technologies to the world, but at that
time the market was saturated and 3dfx had very different objectives.
On August 5, 1996, the iconic Diamond Monster 3D was made
available on the shelves of retailers around the world, based on Voodoo
Graphics technology (chip also called Voodoo 1) and produced by Diamond
Multimedia, as at the time 3dfx did not have its own production lines. The
performances were incredible, just think that with its 50 MHz clock and 4MB of
RAM it was able to push Quake to 50 fps at 640x480 with 65,000 colors, against
the 10-20 fps of the Pentium 120 at 320x200 and 256 colors, making it in fact
the most powerful card on the market.
Diamond Monster implemented the rendering of polygons via
hardware with a series of applicable graphic effects including texture mapping
with filters and mip-map management. The maximum supported resolution was
800x600 at 16 bits per pixel without Z-buffer or 640x480 at 16 bits per pixel
with Z-buffer. The Voodoo series did not handle any computation on primitive
geometries, but rasterized the polygons using the data calculated by the CPU.
The first Voodoo chipset-based video card could only handle
3D graphics and had to be used in tandem with a 2D graphics card connected with
a pass-through VGA cable, so many preferred to call it a "graphics
accelerator" rather than a video card. The Voodoo in fact had two VGA
ports, an input and an output: in the 2D rendering phase it acted as a bypass
and took over for the rendering of 3D.
One of the most significant introductions that accompanied
the release of the Voodoo were the Glide API proposed to the developers, a
package of libraries that improved 3D games in several respects, supporting
geometry and texture mapping in formats identical to those used internally by
the architectures, allowing to interface and communicate directly with the
video card like never before.
In 1997 3dfx tried to solve the problem of using 2 video
cards by producing the Voodoo Rush, which unfortunately turned out to be a
commercial flop. The 2D and 3D chipsets shared the same bandwidth and the
performance was reduced by up to 20% compared to the previous model, to the
point that 3dfx stopped producing it a year after its first marketing.
The second Voodoo model was released in 1998, with its 12 MB
of EDO RAM and 90 MHz clock it conquered the market again. The Voodoo 2
supported multitexturing with a dedicated chip, a feature that quadrupled
performance with games using this technology.
This card was the first to introduce SLI or Scan-Line
Interleave rendering , as it was called at the time by 3dfx: this feature
allowed you to install two video cards in series in order to divide the image
rendering load, guaranteeing no only a net performance boost but bringing the
maximum possible resolution to 1024 × 768.
In 1998 3dfx also proposed a card thatit didn't target the
high-end market like Voodoo 1 and 2, with the Voodoo Banshee 3dfx it wanted to
grab the consumer market . Banshee was based on the technology of Voodoo 2 but
integrated a chip to manage 2D graphics, avoiding users having to buy a second
card to manage it. This model proved to be yet another commercial success for
3dfx, which by now was establishing itself as one of the most important players
in the market.
A terrible misstep
At the end of 1998 3dfx announced the acquisition of STB
Systems, one of the major video card manufacturers and from that moment it
stopped supplying its chipsets to third parties, effectively changing its
business model. This choice brought 3dfx into direct competition with video
card manufacturers, first of all ATI and NVIDIA. On balance we can trace the
beginning of the end of 3dfx back to this specific moment.
The first video card designed and built directly by 3dfx was
the Voodoo 3 and represented a real blow on the technical field. In fact, the
NVIDIA cards could take advantage of the color depth of 24 bits per pixel (16
million colors) against the 16 bits (65,000 colors) of the Voodoo. Despite the
generally higher framerate of the Voodoo, the image rendering was better on the
NVIDIA cards which, subsequently, with Riva TNT and TNT2, also filled the gap
in terms of performance.
In November 1999 at the Comdex in Las Vegas - an exhibition
comparable to the modern CES - 3dfx announced the Voodoo 4 4500, Voodoo 5 5000
and 5500 (based on the VSA100, which differed in the number of chips operating
in parallel on the same PCB ).
3dfx's budget offering, Voodoo 4 4500, was immediately
outclassed by NVIDIA with GeForce 2 MX and all the rest of the lineup by
GeForce 2 and GeForce 2 GTS, which had a much more attractive quality / price
ratio. 3dfx suffered further defeat, those who hoped for 3dfx series 5 as a
possible rematch were shocked, by now the market was always one step ahead and
first the GeForce and then the Radeon DDR became too strong opponents to
defeat.
The fall of the giant
3dfx took a final shot of the tail to try to keep up with
Nvidia by announcing the Voodoo5 6000. The card had 4 VSA100 processors
operating in parallel, at 166 MHz, each with 32 Mb of SDRAM, making the 6000
the first card. with 128 MB of RAM in the world. The huge $ 600 price tag, the
external power connector and some really cool specs turned this card into one
of the most anticipated ever. But while fans around the world eagerly awaited
its new little gem, 3dfx was in dire straits.
A millionaire lawsuit by NVIDIA for patent infringement and
a production line not up to supporting its commercialization led 3dfx to cancel
the 6000 in November 2000. Some Voodoo 5 6000 benchmarks that see it above the
GeForce 2 are still available on the net. Ultra and the Radeon 7500, which at
that time dominated the market, and in some tests it was even superior to the
Geforce 3, which would only be released in February 2001.
Despite being in obvious difficulties, there were two
projects at stake that could have revived the fate of the company. Microsoft
was looking for the ideal graphics chip for its brand new gaming machine, the
Xbox (by the way, find our end-of-gen special on Xbox One here.), but in the
end, as we know, Nvidia's proposal was preferred. The second was a project that
3dfx engineers had been working on for years and that could have completely
revolutionized the market: the Rampage chip .
The video cards based on the Rampage chip should have been
three: the Specter 1000, 2000 and 3000 and they saw for the first time the
introduction of hardware support for transformation and lighting , which NVIDIA
had already brought to its GPUs starting from the GeForce 256.
But by now 3dfx was at the end of its life cycle, its
creditors started a bankruptcy process and, with no hope of being victorious,
on December 15, 2000 it announced the cessation of the business through its
website and it was acquired by NVIDIA, until at that time its biggest
competitor. Many of 3dfx's two hundred employees went to work for ATI on the
"CrossFire" series cards and for NVIDIA, creating the "FX"
series.
3dfx Legacy
In a parallel universe could Rampage have changed the cards
on the table? Difficult to say, the only certain thing is that it would have
outclassed any other card on the market at that time, some of the solutions
adopted on that chip were in fact used by NVIDIA for the GeForce 5 series,
which would be released three years later.
It took only two years and a big misjudgment to bring a
company capable of dominating the market to a disastrous fall. 3dfx had the
merit of being able to interpret and convey the market of the mid-90s, the
introduction of specific APIs for 3D support was revolutionary and for the
first time allowed developers to interface directly with graphics chips,
creating a standard , so much so that some of its engineering solutions are
still part of the genetic makeup of modern video cards.