Location: Financial and Commercial Centre, Volta S.A. Factory (Ivry-sur-Seine).
Date: May 1992.
Point of view: Omniscient (Focus on Karim Belkacem).
On the second floor of the Ivry-sur-Seine factory, far from the radiant heat of the wave-soldering machines and the clinical silence of the sub-micron clean rooms, the financial and commercial centre of Volta S.A. resembled a Wall Street trading room that had never known a closing bell. But here, under the harsh fluorescent light, one did not speculate on volatile stocks or virtual commodities. Real money was counted. Hard, cascading, massive cash.
Karim Belkacem stood with his hands clasped behind his back, behind the long soundproofed bay window separating his director's office from the humming open plan below. The young interim CEO — his features still bearing the crushing weight of running the empire in Lazare's absence, and the still-raw grief for Alexandre de Vigan — watched the large digital dashboards scrolling across the room's walls.
The columns of numbers lined up in bright green, updated in real time. The Institutional Security division, which supplied IMPERATOR servers to the French Defence Procurement Agency and European ministries, generated colossal revenue, more than justifying the company's status as the jewel in the Republic's technological crown. But it was not that division holding Karim's mesmerised attention that morning.
His dark gaze was locked to the counter of the Entertainment division. The original cash cow. The Trojan horse the Ogre of Ivry had launched years earlier against the Japanese archipelago, in total defiance of American computing orthodoxy.
Lazare's first wager had paid off a hundredfold, with an economic violence that even the financial analysts of the Bank of Tokyo had not dared to model in their wildest scenarios. The first iteration of their graphics chip — the SONG 0.5 — coupled with their universal VESLA hardware architecture, had totally, brutally, and definitively disrupted the Japanese arcade industry.
Before the arrival of Volta, the business model of entertainment giants such as Sega, Namco, or Capcom had been frighteningly cumbersome. Whenever a major development studio conceived an ambitious new arcade game, a bespoke motherboard had to be designed to house the code. Specific chips had to be fabricated, complex PCBs created, industrial moulds paid for, everything assembled by hand. The cost of producing hardware literally devoured the profitability of software. Innovation was strangled by the dictatorship of tin and copper.
Lazare had eliminated this constraint with a single stroke of his axe. He had given them the universal Volta board — a slab of black silicon and epoxy, standardised and overpowered, that any arcade machine from Tokyo to Los Angeles could accommodate. He had killed the need for bespoke hardware.
The result in the dark game centres of Shibuya, Osaka, and Akihabara had been seismic. Freed from the ruinous shackles of hardware design, Japanese developers had rushed to Karim's software API — an elegant code library that made it simple to communicate with the SONG 0.5 chip without assembly language barriers.
Within two years, they had ignited the craze for simplistic 3D games — racing cars and space fighters built from flat, untextured polygons, but animated with an insane speed and depth of field that left Western players slack-jawed. And for classic 2D fighting or platform games, the French architecture offered impeccable fluidity, capable of animating dozens of large sprites without a single slowdown or screen tear.
It was a global aesthetic revolution. But for Karim, in the silence of his office, it was above all a masterpiece of accounting vampirism.
"Bring me the figures for this quarter's software licence revenue, Édouard," Karim said, his voice gravelly, without turning from the window.
Volta's chief financial officer, Édouard Renault-Tessier, stood behind him, a heavy red-covered file in his hand. The former investment banker no longer bothered to conceal his near-religious admiration for the extraction mechanism the two young engineers had built.
"The numbers are obscene, Karim. Absolutely obscene," Édouard said, in a voice betraying a delight rare in a man of his customary phlegm.
He opened the file, the crack of the paper breaking the silence.
"We earn almost nothing on the physical sale of the universal motherboard. In keeping with Alexandre's strategy, we are selling it to the Japanese at cost to flood their arcades. But the profitability of the infrastructure is extraordinary. The Huabei factory produces the same standardised circuit in runs of hundreds of thousands of units. We have zero renewal R&D cost on this production line."
The financier moved forward and pointed to the exponential curve spiking off the top of one of the wall screens.
"The invisible gold is in the licence, Karim. On every arcade machine sold by Sega or Namco equipped with our system — and especially on every game cartridge developed using your graphics API — Volta charges a fixed fee of twelve percent of gross revenue. A toll on creativity."
Karim smiled coldly, his eyes fixed on the curve.
"A toll on the entertainment of the masses, as Lazare would say."
"Exactly," Édouard agreed, closing the file with a sharp snap. "The Japanese design the games, absorb the colossal risks of marketing, pay their armies of developers, manufacture the heavy wooden and plastic cabinets, and handle the logistics of global distribution. And we, sitting comfortably in Paris, collect twelve percent of their gross revenue — simply because they use our mathematical equations to draw their pixels. The Entertainment division alone generates a net cash flow of more than eighty million francs per month. And this figure is growing."
Eighty million francs of pure profit, every month.
Karim felt the dizzying weight of the figure settle on his shoulders. It was the absolute cash cow of the empire. A river of liquid, inexhaustible gold, invisible to Washington's eyes, fed by the hundred-yen coins inserted by millions of teenagers across the world.
This cash flow was not destined to pay extravagant dividends to shareholders, as was the custom in the timid boardrooms of Silicon Valley. It was the financial lung of the total war they were waging. It was this money — generated in the blind spot of video games — that financed the ultra-secret projects of Level 4. It was this money that made it possible to corrupt global logistics networks to keep Operation Scavenger alive, to recruit elite Soviet engineers, and to set aside the billions of francs needed for the imminent construction of the future European MegaFab, all without having to beg for toxic lines of credit from American or British investment banks.
"We must protect this source of income at the cost of our lives, Édouard," Karim murmured, his gaze hardening. "If the Entertainment division runs dry, our entire geopolitical war effort collapses. The shield of French Defence and Mitterrand's subsidies will not be enough to fund the foundry and our engineers. If we lose Japan, we lose the war."
"It will not run dry as long as the Japanese continue to push boundaries," the financier countered with the pragmatism of a man who trusts in the durability of exponential curves. "And they currently have no viable hardware alternative to our motherboard."
"Arcade machines are merely a transitional stage," the software architect corrected, turning at last from the bay window to face Édouard. "The arcade is the technological showcase — the laboratory. But the real volume, the true mass market of hundreds of millions of households, is in the living room. Beneath the television. In the 16-bit and 32-bit home consoles that the Japanese giants are designing in total secrecy."
Karim walked to his desk and placed both fists firmly on the solid wood.
"The human eye grows accustomed to miracles very quickly, Édouard. The flat polygons and fast 2D display of the SONG 0.5 astonished them for two years. But in computing, two years is a century. That chip is already archaic. From next year, players will find this simplistic 3D hideous. They will want more. And if we do not give them more, Sega, Namco, or that new rebel Sony will go to Silicon Graphics or American subcontractors to equip their next living room machines."
The technical director adjusted the collar of his dark jacket. The feverish concentration of the blacksmith swept away the traces of his sleepless nights.
"The cash cow needs new genetics, Édouard. Go back to counting your billions and securing the payments from Tokyo. I am heading down to the lab. I have an illusion of the century to create."
❖
Location: Hardware R&D Laboratory, Volta S.A. Factory (Ivry-sur-Seine).
Date: Early May 1992.
Point of view: Omniscient (Focus on Karim Belkacem and his engineers).
The Hardware R&D Laboratory, occupying a level wedged between the deafening frenzy of the assembly lines and the tamper-proof sanctuary of Level 4, was a clinical space. Saturated by the harsh white light of halogen lamps, the air carried a permanent smell of ozone, thermal paste, and white-hot silicon.
It was the domain of the wizards of matter — the place where the abstract concepts of software engineering took on flesh in metal.
Karim Belkacem pushed open the heavy door secured by a keypad lock. It had been exactly one month since he had assumed absolute command of the Volta empire, following the bloody ambush at Eindhoven that had struck Lazare down. In early April, when de Vigan's blood was barely dry and the Ogre was fighting for his life at Val-de-Grâce, Karim had not merely managed the logistical emergency — he had opened a new development front, building on the secret labours Lazare had left behind. He had come today to audit its progress.
Around the enormous antistatic central bench — covered with UNIX CAD workstations and Tektronix oscilloscopes — Volta's five best hardware engineers were waiting for him. Men and women of atypical profile: mask designers poached from the aerospace industry, lithography experts, and Julien, the lead mathematician specialising in spatial geometry. The Forge elite.
Karim dispensed with social formalities. The young technical director — his features still drawn by insomnia but his eyes burning with fresh authority — walked up to the bench and placed both hands flat on the spread drawings.
"One month, gentlemen," Karim announced, his voice cutting through the dry acoustics of the room. "At the start of April, I gave you the order to derive the SONG II graphics coprocessor architecture and produce a revolutionary unified 2D/3D chip. The SONG-III. Show me what you have forged since I gave you Lazare's specifications."
Julien, pushing thick glasses up his nose, stepped forward. He pulled a heavy black binder — the preliminary specifications — toward him.
"We have not been idle, boss," the mathematician began. "The plans Lazare left in March 1991 for this Revision A were absolutely audacious, but... they hold. We have a 132-square-millimetre die integrating 520,000 transistors. The density of the previous generation is almost doubled, thanks to VLSI Technology's 0.6-micron CMOS process."
Julien opened the binder to the internal architecture page.
"The SONG-III is no longer a simple accelerator. It is the first unified graphics processor. It merges professional 2D acceleration for our VoltaOS 3.0 operating system with a full real-time 3D rendering pipeline — all on a single chip, clocked at 66 MHz."
Karim nodded, scanning the schematics.
"That was the goal. Offload the CPU. If the CPU has to draw windows and calculate shadows, we die. The GPU must do the heavy lifting."
The chief lithography engineer — a grey-haired man recruited from Thomson — took over, pointing to the memory controller schematic.
"The real power, Karim, is in the memory addressing. It is paired with 4 MB of standard VRAM on EDO DRAM, with a 64-bit bus. That gives us an insane bandwidth of 528 megabytes per second. And above all, it is not a chip soldered blindly onto the motherboard. Lazare designed it as a daughter card in BBI-96 format — inserted directly into the expansion connector of the Bonaparte Bus Interface."
"Modularity," Karim agreed, with a brief smile. "We can upgrade the GPU in our professional clients' machines without requiring them to replace the whole unit. An approach American industry has not even begun to conceptualise. But what interests me this morning is not the office automation of our state servers. It is the Japanese target. Entertainment."
Karim leaned against the bench, his face closing. The challenge of this meeting was not simply to validate a piece of hardware for their own office machines.
"The SONG 0.5 and the universal arcade board whetted the Japanese appetite for 3D," Karim recalled. "But the human eye is insatiable. Flat cubes are no longer enough. Our Asian clients demand the next step. They want complex textures, dynamic lighting, transparency. And if they cannot find that power here, they will go to Silicon Valley for it. They will sign with the Americans for their next home consoles."
He pointed an authoritative finger at the plans spread across the table.
"The true purpose of this SONG-III is to sit at the heart of the next generation of Japanese consoles. In particular, the secret CD-ROM machine that Sony engineers, led by Kenji Takeda, are developing in Minato in rebellion against Nintendo. By delivering them the perfect, inexpensive, overpowered component, Volta ensures that Japanese developers continue to use our programming language. This is what will lock in our global royalty collection for the decade to come. So tell me — is this chip a weapon of mass seduction?"
The team's mathematician, crossing her arms, took a slow breath. It was the moment of truth.
"It is a monster, boss. But a monster we had to tame."
She drew the section detailing the 3D pipeline toward her.
"The SONG-III 3D pipeline is a rasteriser with six stages: Setup, Raster, Texel, Filter, Z-test, and Blend. On paper, it is magnificent. But we ran into the same problem as with the military version: texture mapping. Placing 2D images on 3D polygons — accounting for perspective — consumes a colossal amount of floating-point computing power. If we let the chip perform exact calculations every cycle, it becomes too expensive to produce for a consumer console sold at cost."
Karim remembered the instructions he had given a month earlier.
"I told you to find a shortcut. The eye is fooled by the illusion."
"And that is precisely what we did, Karim. But Lazare had left very precise instructions to go further than a simple shortcut," Julien corrected, a gleam of admiration in his eyes. "He demanded hardware-accurate perspective correction. We could not ignore this if we wanted to prevent textures from warping the way they do on cheap low-end hardware."
Julien pointed to the diagram of the Z-TEST block.
"The clever trick does not lie in the mathematical shortcut on textures, but in memory management. Lazare invented the Hi-Z — anticipatory depth rejection."
Karim frowned, intrigued.
"Explain."
"A classic 24-bit Z-buffer calculates the depth of every pixel to determine whether it should be displayed or is hidden behind a wall," the engineer said. "It is expensive. Lazare's Hi-Z rejects entire 8-by-8 pixel blocks before even reading the detailed Z-buffer, by comparing a single aggregate value for the block. This saves us almost thirty percent of VRAM bandwidth — bandwidth we can redirect to perform perspective-correct texture mapping without blowing up the chip's production cost."
The mathematician resumed, scientific excitement overtaking her caution.
"And with the freed bandwidth, we were able to implement what you wanted to make the image truly beautiful. The FILTER block applies hardware bilinear filtering to textures — a weighted average of the four neighbouring texels. No more jagged, hideous square pixels when the camera approaches a wall. The result is smooth, fluid. And the BLEND block handles per-pixel alpha blending across 256 levels. Transparent windows, fog effects, drop shadows — all in hardware."
"Give me the numbers, Julien," Karim interrupted, the commercial mind overriding the engineer. "Raw performance."
Julien flipped quickly to the performance charts.
"In pure 3D without texture, two million polygons per second. In full 3D with perspective-correct texture mapping, bilinear filtering, and active Z-test active... we hold 800,000 triangles per second."
Karim was silent for a moment. 800,000 textured and filtered polygons per second. In 1992, most arcade machines were capped at a few tens of thousands of raw polygons. American workstations costing 100,000 dollars sweated to reach those numbers. The SONG-III was about to miniaturise this power and place it beneath a television screen. It was a technological rout foretold.
"The industry will not offer consumer chips with bilinear filtering and alpha blending for another five or six years," Julien concluded, almost unnerved by the anachronism of the component they had just completed. "We are five years ahead of the state of the art."
"There is a catch," Karim said, his eyes sharpening. "Lazare's equation is too elegant. This density on 0.6 microns, at this level of complexity... Where is the problem?"
The engineers exchanged embarrassed glances. The mathematician sighed and opened the Errata section of the document.
"You are right, boss. Revision A is not perfect. There are bugs related to the hardware divider."
She pointed to Errata S3E-001.
"Texture artefacts on sub-pixel triangles. Polygons with at least one vertex within half a pixel of the screen edge produce precision errors in the perspective divider. It causes edge pixels to flicker. Developers will need to work around this in software by clipping triangles one pixel inside the edge."
"And the Hi-Z misfires on high-velocity scenes," the chief engineer added — Errata S3E-003. "It does not invalidate correctly after a Z-buffer clear, which can cause incorrect pixel rejection on the first frame after a clear."
Karim listened to the list of flaws. Far from being troubled, he smiled coldly. This was the reality of the silicon industry. Perfection did not exist in a first revision. What mattered was reaching the market first, with a lead so commanding that developers would forgive the errata.
"These are manageable bugs for a launch," the technical director said. "Fix the divider precision and the Hi-Z invalidation on Revision B, scheduled for the end of the year. For now, I want this chip taped out."
Karim stepped to the table, placing both fists firmly on the SONG-III schematics.
"The architecture is validated. This GPU is our weapon of mass seduction. We are not only going to offer it to Sony for their future console. We will integrate it into the next generation of Volta computers for the consumer market. The SONG-III will transform the average person's workstation into a 3D rendering machine capable of competing with Silicon Graphics' professional stations."
He straightened up, scanning the room with an imperious gaze.
"The Americans are trying to suffocate us by cutting off our supplies. We will answer them by creating the digital retina of the world. Prepare the lithography masks for despatch to the foundry. I want physical prototypes in two months to send to Kenji Takeda in Tokyo. Lazare is wounded, but his silicon twin has just been born. Get to work."
Karim left the laboratory. In the ozone-saturated air, the five engineers looked at one another. The pressure was immense, but they had just understood that the future of global visual computing was at the tip of their styluses. The silicon war was intensifying — and the Ivry forge had just produced its masterpiece.