The Story of Rendition: A Pioneer in 3D Graphics

Today’s PC gamers can consider themselves lucky when it comes to choosing a graphics card (relatively, of course). While there are only three GPU manufacturers, every model they produce is almost guaranteed to run any game you like, albeit at varying speeds. Thirty-plus years ago, the landscape was very different, with more than ten GPU companies, each offering its own approach to rendering graphics. Some were highly successful, while others were less so.

But none were quite like Rendition, a pioneer of the 3D gaming industry. It had a great first product, followed by two underwhelming ones, before fading into obscurity – all within five years. Now that’s a story worth telling!

A messy start to the decade

PC gaming in the early 1990s was exciting and frustrating in equal measure – getting hardware to work properly required endless fiddling about with BIOS and driver settings, and even then, games at the time were notoriously picky about it all. But as the years rolled on, PCs became increasingly popular purchases, and millions of people around the globe proudly owned machines from Compaq, Dell, Gateway, and Packard Bell.

Not that these were aimed at gaming. Spending $3,000 would get you a nice Intel 486DX2 machine, running at 66 MHz, and boasting 4MB of RAM and a 240 MB hard drive. But the graphics card? If you were lucky, it might have sported an ATI Graphics Ultra Pro 2D accelerator, replete with 1 MB of VRAM and the Mach 32 graphics chip. But if lady luck wasn’t on your side, then you’d be cursed with something ultra cheap from SiS or Trident.

8-bit color was typical for a cheap 1991 graphics card.

Compared to the visual output of consoles like Sega’s Mega Drive/Genesis or the Super NES, even the best PC graphics were basic but serviceable. However, there were no games on that platform pushing the hardware to its limits.

The best-selling PC titles throughout the formative years of the 90s were point-and-click adventure games, such as Myst, or other interactive titles. None of these required powerful graphics cards to run – just those that supported 8- or 16-bit color.

Change was on the horizon. The direction games were heading was clear as 3D graphics were becoming the norm in arcade machines, though they used specialized hardware to achieve this. PC games that featured 3D graphics, like Doom, handled all the rendering through the CPU, with the system’s graphics card merely turning the frame into something that could be displayed on a monitor.

Doom didn’t need a 3D graphics accelerator

To emulate the graphics seen in games like Namco’s Ridge Racer, for example, a home PC would need hardware with similar capabilities, but at a fraction of the arcade’s retail price. Interest from the old group of graphics companies (ATI, Matrix, S3, etc.) was slow to develop, leaving plenty of room for fresh blood to enter the field.

Numerous startups emerged, designing new graphics adapters that could take over 3D rendering from the CPU. Enter stage left, Rendition Inc, co-founded by Jay Eisenlohr and Mike Boich, in 1993.

Their ambitions were simple yet bold – create a chipset for an add-in card would handle both 2D and 3D graphics acceleration, then sell it to both gaming and professional industries. However, they weren’t alone in this endeavor and faced stiff competition from 4 other new companies, as well as those that had been around for many years already.

First blood is drawn

In 1995, Rendition announced its first product, the Vérité V1000-E. As a fabless company, Rendition relied on selling licenses for others to manufacture and use its design. Four OEM vendors took up the gauntlet, and a year later, Canopus, Creative Labs, Intergraph, and Sierra all launched cards using the new chipset.

Today, there’s relatively little difference between the architectures of various graphics cards, but the early iterations of 3D accelerators were remarkably disparate. Rendition took an unusual approach with the V1000, designing the central chip essentially as a RISC CPU, similar to MIPS, which acted as a front-end to the pixel pipeline.

Image credit: vgamuseum

Running at 25 MHz, the chip could perform a single INT32 multiplication in one clock cycle. However, standard rendering tasks such as texture filtering and depth testing all took multiple cycles to carry out.

Theoretically, the V1000 could output 25 Mpixels per second (known as the fill rate of a graphics card), under very specific conditions. For typical 3D games, the chip took at least two clock cycles to output a single textured pixel, effectively halving the fill rate.

But Rendition had a few aces up its sleeve with the first Vérité model. It handled all the triangle setup routines in hardware, whereas every other graphics card required the CPU to handle these tasks. Since the PC’s central processor was also responsible for all vertex processing in a 3D game, this feature of the V1000 gave the CPU some much-needed relief.

ISA VLB slots used before PCI became commonplace.

All V1000 cards used the PCI bus to connect to the host computer, which was still a relatively new standard in the industry. They could take advantage of features like bus mastering and direct memory access (DMA) for extra performance.

Rendition’s graphics card was also fully programmable (technically, it was the first ever consumer-grade GPGPU), and the engineers developed multiple hardware abstraction layers (HALs) for Windows and DOS. These HALs converted instructions from various APIs into code for the chipset. In theory, this made the Vérité graphics card the most widely supported by software at the time.

Bringing in the big guns

The pièce de résistance came in the form of a rather famous game – Quake.

After the rampant successes of Doom and its sequel, id Software began work on a new title, one that would take place in a fully 3D world (unlike the pseudo-3D nature of Doom). Quake was released in June 1996, and six months later, id Software offered a port optimized for the Vérité chipset, called VQuake.

In the original version of the game, all 3D rendering was handled by the CPU. However, programmers John Carmack and Michael Abrash rewrote large parts of the code to take advantage of the Vérité’s capabilities.

At that time, processors like Intel’s Pentium 166 could run Quake at around 30 fps at a resolution of 320 x 200. With VQuake and Rendition’s graphics card, this increased to over 40 fps, and the game included proper bilinear texture filtering and even anti-aliasing (via a system developed by Rendition and patented a few years later).

That might not sound like a huge improvement, but for a fast-paced game like Quake, every extra frame per second and graphical enhancement made a difference. Still, the question remains: why wasn’t the performance even better? Part of the answer lies in the V1000’s lack of hardware support for z-buffers, so programmers still had to rely on the CPU for depth calculations, and even then, do them as little as possible.

For a first attempt at a graphics accelerator, the Rendition Vérité V1000 was impressive. It was the only product on the market that handled both 2D and 3D graphics, and had lots of performance available, as long as the game avoided its weaknesses – such as using DirectDraw or one of its proprietary 3D HALs, it was very fast.

However, V1000 was also notoriously buggy (especially in motherboards that didn’t support DMA), very slow when operating in legacy VGA modes, and OpenGL support was very poor. In the end, it was this weakness in OpenGL that ultimately defined the product’s legacy.

Taking second place to the star performance

Another startup with ambitions in the world of 3D was 3Dfx Interactive, founded a year after Rendition, but releasing its first product, the Voodoo Graphics, well before the V1000. Initially sold to the professional market, the 3D-only accelerator made its way into home PCs by 1996/1997, thanks to a significant drop in DRAM prices.

Boasting a 50 MHz clock speed, this graphics chipset (codenamed SST1) could render a single pixel, with a bilinear filtered texture applied, once per clock cycle – considerably faster than anything else on the general market.

Image: VGA Legacy MKIII

Like Rendition, 3Dfx developed its own software, called Glide, to program the accelerator. However, by the time it appeared in a wide array of OEM models (from the same companies that used the Vérité V1000), another HAL was introduced – MiniGL. Essentially, MiniGL was a highly cut-down version of OpenGL (an API typically used in the professional market) and came about entirely because of another version of Quake that id Software released in early 1997.

GLQuake was developed because John Carmack didn’t enjoy working with proprietary software. This version used a standardized, open-source API for rendering, offering a significant boost in performance and graphical quality for supported cards. 3Dfx’s MiniGL driver converted OpenGL instructions into Glide ones, allowing the Voodoo Graphics chipset to fully support GLQuake.

A Canopus graphics card using the faster V1000L-P chip. Image: vgamuseum

With a 50 MHz clock speed and hardware support for z-buffers, cards using Voodoo Graphics were notably faster than any using Rendition’s V1000. 3Dfx’s first model was limited to a maximum resolution of 640 x 480 and always in 16-bit color, but even so, the best Vérité cards were only half as fast as any Voodoo model.

Attempts were made to fight back with the release of an updated chip (the V1000L or L-P), which ran at a lower voltage, allowing it to be clocked up to 30 MHz and paired with faster RAM. However, it still lagged behind 3Dfx’s offering. In the competitive semiconductor industry of the 1990s, much like in sports, history tends to forget those who come in second.

A faltering second act

Rendition began work on a successor to the V1000 almost immediately, aiming to build on the first model’s strengths while addressing its weaknesses. With a target launch of summer 1997, the engineers reworked the chip designs – improving the clock speed and feature set of the RISC processor (up to three instructions per cycle), expanding the capabilities of the pixel engine (adding hardware support for z-buffers and single cycle texture filtering, for example), and upgrading from slow EDO DRAM to faster SGRAM.

Chip binning was starting to become more noticeable in the graphics chip industry

Chip binning, a practice becoming more common in the graphics chip industry, allowed manufacturers to sell different versions of the same chip at various price points. Rendition took advantage of this by releasing two versions in September 1997 – the V2100, clocked at around 45 MHz, and the V2200, which ran 10 to 15 MHz faster. Aside from clock speed, there was no difference between the two chips, but most vendors opted for the faster model.

The reason was simple – they could charge a lot more for it. For instance, Diamond Multimedia initially sold the V2100-powered Stealth II S220 for $99 (later reducing it to half price due to poor sales), while the Hercules Thriller 3D, using a 63 MHz V2200, was priced at $129 for the 4 MB version and $240 for the 8 MB version.

Image: VGA Legacy MKIII

As successors to the V1000, these new chips were clear, if understated, improvements. Performance was definitely better, and the additional features improved compatibility with the growing number of 3D games hitting the market. However, not everything was going smoothly at Rendition.

Development of the V2000 series had been plagued by numerous glitches, partly due to the company’s design process. Lacking its own fabrication plants, most of Rendition’s processor testing was done via software. Unfortunately, what works well in simulations often doesn’t transfer properly once fleshed out in silicon, and Rendition missed their target launch by a number of months.

The competition wasn’t faring much better (3Dfx’s second offering, the Voodoo Rush, was actually slower than the original Voodoo Graphics), but the V2000 series simply wasn’t a big enough step forward compared to what ATI, Nvidia, and PowerVR were releasing.

3dfx’s Voodoo Rush used one chip for 2D (left) and two chips for 3D (right). Image : vgamuseum

While the new hardware finally introduced mip-mapping, it could only apply it per triangle, whereas competitors could do so per pixel. Rendition also continued to push its proprietary software, perhaps at the expense of offering better support for OpenGL and Direct3D, though both APIs were now properly supported.

Despite these challenges, Rendition’s engineering efforts garnered enough interest to attract further investment from other companies.

Hopes and failed ambitions

In 1998, while Rendition was still developing the third iteration of its Vérité graphics chipset, the company – along with its IP and staff – was sold to Micron Technology, an American DRAM manufacturer that was acquiring multiple businesses at the time. The V3000 series was expected to be more of the same, albeit with higher clock speeds (IBM was slated to produce the chip) and a significantly improved pixel engine.

However, despite having access to its own fabrication facilities and far more resources, the project failed to progress quickly enough to remain competitive with products from 3Dfx, ATI, and Nvidia. Micron’s bosses decided to pull the plug, and the V3000 was abandoned.

Instead, a new direction was charted, and the V4000 project was initiated, with a planned launch in 2000. This chipset was designed to incorporate a host of new features, the most impressive being at least 4 MB of embedded DRAM (eDRAM). The graphics chip in Sony’s PlayStation 2 (launched in March 2000) also featured eDRAM, so this wasn’t an overly ambitious goal. However, Micron wasn’t targeting consoles or discrete graphics cards.

The PS2’s basic GPU surrounded by 4MB of eDRAM.

To the management, the motherboard chipset market seemed to be more profitable, as 3D accelerators were only of interest to PC gamers. By making a single processor that could take over the role of the traditional (and separate) Northbridge and Southbridge chips, Micron was confident that they had the money and people to beat the likes of Intel, S3, and VIA at the same game.

Ultimately, it was not to be. The project was abandoned over concerns that the single chip would have been far too large – over 125 million transistors. For comparison, AMD’s Athlon 1200 CPU from the same period comprised just 37 million transistors. Micron later attempted to develop a chipset for the Athlon processor, with some eDRAM serving as L3 cache, but this effort didn’t go far, and the company soon exited the chipset market altogether.

An ignominious end

As for Rendition and its graphics processors, it was all over. Micron never did anything with the IP, though it briefly used the Rendition name for a budget range of Crucial memory products, before swapping it for something new.

Cards with Vérité chips were only on shelves for a handful of years, though for a brief moment if you wanted the best performance and graphics in Quake, it was the name to have inside your PC. Time has pruned away nearly all the other graphics firms and the market is now dominated by Nvidia, which purchased 3Dfx over 20 years ago. ATI was acquired by AMD in 2006, Imagination Technologies stopped making PowerVR cards but their graphics IP is used in modern chips made by Apple and many others, while Matrox abandoned the gaming sector to serve a niche professional market.

Today, Rendition is a small footnote in the history of the graphics processor – a reminder of the days when 3D graphics was the next big thing, and chip designers pursued radically different approaches. Though long gone, Rendition it’s not forgotten.