The Voodoo's Rise and the Golden Sample
When the reviewers of Ultima Online publicly released their individually measured benchmark data, a massive backlash erupted among gamers. The reason was the absurdly overwhelming performance of the Voodoo graphics card.
Before Voodoo arrived, the card boasting the highest performance was Nvidia’s Riva TNT. The name TNT stood for Twin Texel, a term created to highlight its ability to process two texels per clock cycle. The newly released Voodoo also processed two texels per clock, but its key difference was hardware-level support for multitexturing.
A simple way to elevate graphics quality in 3D games was to use multiple textures. Realistic reflection effects or explosive visuals could be implemented far more easily and dramatically with multitexturing. The downside was that it consumed a fair amount of texture processing power, so excessive use would cause frame rates to plummet.
Voodoo dramatically improved multitexturing performance by refining its graphics processing engine. Thanks to this, frame rates in existing games rose sharply. In Doom 2, which had become the standard for 3D games, it outperformed the previous champion, the Riva TNT, by more than forty percent.
Yet 3DFX wasn’t satisfied. They had prepared two more trump cards.
“High-Octane? What kind of naming sense is that?”
The first secret weapon was their proprietary library, named High-Octane. Using 3DFX’s exclusive High-Octane library would guarantee at least a fifty percent performance boost over competitors, they claimed, with the potential to deliver over one hundred percent in some cases. To demonstrate, they released a demo of the racing game Need for Speed.
At SVGA’s standard resolution of 640×480, the Riva TNT couldn’t even reach thirty frames per second. In contrast, Voodoo delivered a staggering performance close to fifty frames. This was using the Android OS’s Glide X library. When the proprietary High-Octane library was applied on top of that, it could stably maintain over sixty frames per second.
It was all talk. No matter how excellent a library was, if the game developers’ programming skills were poor and optimization failed, the result would be worthless. On the other hand, someone with John Carmack’s level of talent could squeeze incredible performance even from an ordinary library.
“At least SLI sounds somewhat reasonable.”
The second weapon was a technology called SLI—Scalable Link Interface. By connecting two or more cards together, it maximized 3D processing power. Typically, two cards were linked with a cable, and this setup could stably deliver over sixty frames per second in Doom 2 at SVGA’s highest resolution of 1024×768.
Since a single Voodoo card cost $399, buying two would set someone back $800. On top of that, Voodoo lacked 2D graphics capabilities, requiring a separate VGA card for 2D processing. If one chose a high-end 2D card from Matrox or ATI for better color and speed, the total cost for graphics alone could easily reach a thousand dollars.
Nevertheless, gamers and netizens went wild for Voodoo. Internet multiplayer enthusiasts were especially enthusiastic. Higher frame rates meant faster reactions than opponents, which directly translated to higher win rates.
“Seriously, a proprietary library?”
Yoo Jae-won grumbled as he stared at the 3DFX Voodoo cards sitting on his desk. It was only after the NDA had been lifted and reviewers began distributing their benchmark data that he finally received his own Voodoo units. The blatant cold-shoulder treatment had left him thoroughly irritated.
Since ID Group had established itself as the dominant force in the IT industry, 3DFX was the first company to treat Yoo Jae-won this way. He now recalled how, during the creation of the original Glide X 3.0, he had stepped in to support 3DFX when other VGA manufacturers tried to veto their new technology. The memory only doubled his sense of betrayal.
“Their management has always been shortsighted, but I didn’t expect them to go this far.”
In his heart, he wanted to call 3DFX’s president and demand an explanation. But he didn’t act on it. Anyone capable of thinking that far ahead wouldn’t have chosen to openly antagonize ID Group like this.
Suppose High-Octane became the standard. The benefits they would gain were almost nonexistent. Even if they were extraordinarily lucky and High-Octane somehow replaced Glide X, they still couldn’t monetize the library. If they ever tried tampering with it—inserting code that deliberately degraded performance on competitors’ cards like Nvidia’s—they would end up like Microsoft in an instant.
All they could realistically hope for was a significant boost in brand recognition and increased Voodoo sales under the title of “technology originator.” As a bonus, their development costs would also skyrocket. Just looking at Glide X, the latest version—X3—had grown the library’s size more than eightfold. That increase came with a mountain of new technology, all of which required time and money. The budget invested in development alone easily exceeded tens of millions.
3DFX operated with a development team less than one-hundredth the size of ID Group’s. As computer technology continued to advance, graphics would only grow more sophisticated. A small team like theirs would never be able to keep up.
“They went under like this before, too.”
In the previous timeline, Voodoo had collapsed almost overnight. The reason was precisely these shortsighted decisions. The typical business model for companies making 3D-only chips was to produce only the accelerator chip themselves while letting established VGA manufacturers handle card assembly and distribution. 3DFX, apparently unwilling to share the pie, switched to an exclusive production system. They manufactured entire cards and then supplied them to other companies.
It was doomed to fail. They barely had time for technology development, yet they were trying to handle manufacturing and distribution as well. 3DFX’s resources became dangerously stretched. Meanwhile, their rival Nvidia stuck to the original model and focused purely on performance improvements. The performance gap widened rapidly, and 3DFX eventually ended up being acquired by Nvidia.
“Will it be any different this time?”
To assess 3DFX’s capabilities, Yoo Jae-won began examining the High-Octane library.
“Just as I thought.”
The name High-Octane sounded impressive—like something that would ignite with explosive heat. In reality, it was little more than a stripped-down version of the OpenGL library. He hadn’t even needed to perform reverse engineering. A quick visit to 3DFX’s developer page, downloading their library and sample code, and a cursory glance had been enough to see it clearly.
They had simply extracted commonly used functions from the vast OpenGL library and tuned them. Most of the functions were downgraded. While some compromises were inevitable for hardware acceleration, many changes sacrificed graphics quality. Moreover, the once-intuitive and systematic command structure had been altered so drastically that programmers would now have to relearn everything.
Glide X had faced similar criticism in its early days. But with consistent support from ID Group and the runaway success of the Doom 2 engine, the situation had changed completely. Even developers with mediocre programming skills could create everything from single-player to multiplayer games with ease using the Doom 2 engine. Although the graphics ended up looking somewhat similar, the recent surge of original titles had cemented Glide X as the undisputed standard.
The upcoming Return to Castle Wolfenstein and Quake would eventually be sold as standalone engines as well. In terms of development environment, there was simply no comparison.
“They’re going to collapse even if I leave them alone.”
The conclusion was clear. Yoo Jae-won felt confident enough to bet his entire fortune that 3DFX would go under far faster than in the previous timeline.
“Still, I can’t just sit back and watch.”
Even though their shortsighted strategy was obvious, Voodoo was currently the best 3D card on the market. If too many developers, seduced by its performance, adopted High-Octane, it could derail the grand picture Yoo Jae-won was trying to paint.
The answer, as always, was a technological showdown. 3DFX might win the current graphics card war, but if Glide X’s camp could reclaim the top spot in the next battle, that would be enough.
Yoo Jae-won decided he would need to introduce bold new technology in the next-generation graphics card—something that would clearly separate allies from enemies.
“Simply switching the graphics processing method to a shader system should be enough to break Voodoo.”
Shaders were a technology that handled the most basic elements of graphics—pixels and polygons. At the moment, hardware was still being built to support various techniques because overall capability was limited. If every technique had to run on a single screen, it inevitably meant some parts of the hardware were sitting idle.
With a shader system, the 3D graphics card’s resources could be dynamically allocated according to each situation, maximizing performance. The more efficiently shaders were designed and the more of them that could be integrated onto a single chip, the greater the leap in graphics performance. At the same time, the difference in performance depending on a programmer’s skill would become enormous.
The same card could produce breathtakingly smooth and beautiful graphics in one game, while another title might look mediocre and stutter badly. This was exactly why Glide X—with its efficient libraries and superior game engines—would hold a massive advantage.
“Hmm. That means this will have to go into the next Android OS version.”
The burden on the tentatively named Android 3.0 suddenly felt much heavier. The version already carried a heavy workload, and now it would also have to shoulder the graphics card war.
Moreover, there weren’t many developers capable of programming exactly according to the vision Yoo Jae-won had in mind. Only a handful of elite programmers—the Alpha Team at Android Corporation and John Carmack among them—could keep pace with him. And they already had more than enough on their plates.
Even Yoo Jae-won himself had too many responsibilities. He was overseeing the next Android project, the ongoing development of iWorks, and the mass production issues for the Tiffany Phone, which would soon launch alongside the commercialization of 2G mobile communication.
His regular duties alone were overwhelming, yet there were even more matters demanding his attention. The North American summit was coming up in early May, labor law reforms needed to be addressed, and internet-related legislation was piling up.
“Let’s die trying!”
None of it was something he could easily abandon. The ambitious Yoo Jae-won had no intention of giving up on any of it. With renewed determination, he immediately turned his focus to iWorks.
Since 3DFX had clearly declared its independent path, the video card for iWorks was decided to be Nvidia’s Riva TNT. Once the decision was made, Yoo Jae-won promptly sent a message via ID Talk to Nvidia’s CEO, Jensen Huang.
Jensen Huang, who had been brimming with confidence after launching the Riva TNT, had been in a state of panic ever since Voodoo’s overwhelming performance was revealed. Nvidia had already placed massive orders for wafers based on their expected performance. With Voodoo’s monstrous results, all that inventory risked becoming dead stock.
But now that Yoo Jae-won had chosen the Riva TNT for ID Technology’s first workstation, Jensen could finally breathe again.
“However, I have a small request.”
“What is it? We’ll do our best!”
“Would yield sorting be possible?”
Before creating a brand-new 3D graphics architecture, Yoo Jae-won’s temporary countermeasure against Voodoo was yield sorting. To a manufacturer, yield usually meant whether a chip worked or not. What Yoo Jae-won meant, however, was chips that operated exactly as designed—and at the highest possible speeds.
The lithography process used to etch circuits onto wafers relied on heat and chemical reactions. If the circuit wasn’t drawn properly, the chip was a complete defect. Conversely, some chips reacted exceptionally well and operated at speeds far beyond the original design. These were known as golden samples.
“Huk! That might be difficult, Chairman Yoo. As you know, that would require additional testing during the packaging process. We also have to consider fairness with other companies. It’s not something a small firm like ours can easily handle.”
“Of course, I understand. I’m not asking for golden samples at regular chip prices. Please listen to my proposal and think it over carefully.”
Yoo Jae-won wasn’t simply asking them to pull golden samples. His idea was to name the chips according to their operating speeds. The current Riva TNT ran at 200 MHz. In the previous timeline, the model had only reached 90 MHz, but with foundries switching to HPC processes and design teams adjusting accordingly, clock speeds had more than doubled.
He suggested naming the 250 MHz version Riva TNT Pro and the 300 MHz version Ultra, allowing them to charge higher prices accordingly.
“Of course, the prices shouldn’t be absurd.”
“Absolutely! That makes perfect sense.”
Jensen Huang’s attitude had completely flipped from his initial reluctance. Differentiating prices by clock speed was common practice in the CPU industry. In the VGA sector, however, factors like Glide X compatibility, video memory capacity, and TV output capability had been the main price determinants. Now that chip operating speed had become a major performance factor, tiering accelerator chips by speed—just like CPUs—was inevitable.
Since it was going to happen anyway, giving them advance notice in exchange for golden samples was more than reasonable. Jensen Huang immediately accepted the proposal. As the sole decision-maker at Nvidia, he didn’t even need to hold a meeting. Moreover, the deal meant selling golden samples that would normally be sold at regular prices at a premium—while supplying them to ID Group’s next-generation PC. It was pure profit.
“I’ll send the samples right away.”
In high spirits, Jensen Huang opened his treasure vault. Like most semiconductor companies, Nvidia kept specially performing chips aside for yield research. He decided to send some of them to Yoo Jae-won.
For Yoo Jae-won, who had been delayed in testing the iWorks prototype because of Voodoo, this was an enormous help.
That afternoon, Jensen Huang kept his word. An Nvidia employee personally delivered the golden samples, completing the iWorks prototype’s component list.
Yoo Jae-won immediately began assembly. Although the system was ultimately going to be manufactured by TG Computer like the Egg series, he couldn’t wait for the finished product to arrive by plane. Besides, assembling it himself would let him discover areas for improvement through trial and error. So once again, he did it personally.
Most computer components followed strict standard specifications, making assembly as simple as putting together Lego blocks. The difficult part was integrating newly designed components without established standards—such as the internal LED lighting and cooling parts.
The LEDs weren’t just left to shine raw light. They were mounted along the case’s edges with slightly translucent plastic to create atmosphere, which made secure, clean installation quite challenging. Fortunately, the cooling fans had been commissioned from a specialized manufacturer and arrived as finished green-illuminated units.
Once assembly was complete, the most nerve-wracking moment arrived—the first power-on.
“Alright. Let’s turn it on.”
“Yeah! Hurry and turn it on!”
With Tiffany standing right beside him, the tension multiplied several times over. After taking a short breath, Yoo Jae-won bravely pressed the power button.