Introduction

ATI goes over a billion while Real3D, and Raycer prove that technology is not enough to succeed in graphics.

It's been a hell of a week. ATI breaks through the $1 billion barrier, and that's so amazing no one talks about it. I guess, everyone is dumbfounded. We've never had a graphics chip company that big, and I think it further strengthens my arguments about the industry from last week's column, in which I took a dig at 3dfx, in jest, although I received a fair number of flames. They were lower in numbers than in the past, and a bit half-hearted, but what can you do. So, I stick by my assertions that the graphics industry basically belongs to ATI, S3, and Nvidia. They can all rack up billion dollar revenues, and stay above the menace of Intel, because that's what they're going to have to do, as you'll read later.

I am very disheartened to see that the extent of coverage given to ATI's results was based on two things, the company missing profit expectations, and references to comments about the effects of the Taiwanese earthquake. I don't want to praise ATI because, I just don't want to praise ATI, but I will say this, One Billion Two Hundred and Thirty One Thousands Dollars! That's a worth a big bunch of celebrations. Everyone who holds stock in a graphics company should be sending letters of thanks to ATI for pointing out that there is big growth in the market for some, and it doesn't have to come from glitzy television marketing campaigns, or the fastest performance products.

Big Fly in the Ointment

Having said all of that there is a big fly in the ointment for the whole industry, one that keeps getting fatter. Intel now holds one of the most enviable graphics patent portfolios in the business. Only Evans and Sutherland can boast its equal, and they haven't done much with theirs in the PC industry lately. No such luck for Intel's competitors. Not only does Intel have cross-licensing agreements with SGI, and S3, but now it has a bag full of Lockheed Martin/Martin Marietta/Real3D intellectual property. See the story below on the collapse of Real3D, culled from the Wave Report:

"***Real3D Collapses

Burdened by a failure to execute and a continually shifting business model Real3D has collapsed of the weight of its aerospace legacy. On 10/14 Intel purchased the Lockheed Martin stockholdings for an undisclosed sum. Then Intel let go of the staff. But on Friday, 10/15, many were re-hired by Intel as contractors. Jobs are open for evaluation. Lockheed Martin is also paying the Real3D staff until January 1, 2000 and with a 1 week severance for each year of employment. Intel gains by getting Real3D's extensive patent portfolio and potentially is better able to compete in the 3D market. Yet, the WAVE Report has picked up that ATI has opened an office in Orlando ostensibly staffed by former designers at Real3D. In the last response to those who stayed on to the end, the Real3D options were not converted to Intel stock or options. A sad tribute to those who tried in a company that could not last 2 years."

In simple terms, Intel has many, many years worth of graphics technology that it can simply squeeze into its components and chipsets, and it doesn't have to go out and re-invent the wheel to do it. In fact, Intel can scare off any graphics company that tries to close in on it by just waiving a handful of patents, and that's that. The rich get richer, and everyone else just has to suck up to them. Heck, what else does it mean? The Intel PC universe has expanded to the point that it is in danger of collapsing in on itself (that's a high-brow, cosmic analogy of sorts, people). All Intel can do is prop it up by taking over anything and everything that might impact the price of a CPU. AMD can beat them on price, and on performance even, but it can't prime PC OEMs, and systems integrators with a packaged solution that requires them to do nothing more than stick a box around Intel's components, and slap their own label on it. No pesky graphics people to worry about, and if the whole RDRAM thing gets going, even the memory is kind of in Intel's hands.

Heck, what else should Intel do?

Another News

Anyway, on with another interesting piece of news. John Latta at Fourth Wave, the publisher of the Wave Report, also scooped up this story:

"***Raycer Graphics Runs out of Cash - Apple Steps In

Raycer Graphics ground to a halt before it could complete its 3 chip set for a high-end 3D accelerator. Having used up a rumored $30m the company could not secure any more funds from its investors, Vulcan Ventures (Paul Allen) and Sevin Rosen Funds, and the VC community also said no. The latest word is that Apple Computer has stepped in and picked up the engineering team and IP. The rest has been let go. Raycer has accumulated top 3D talent, including those from SGI. Thus, the most valuable asset to protect was the team before they quickly dispersed. Both 3Dfx and nVidia are actively looking for talent. Raycer has an innovative architecture but a very complex one to implement. We might still see the fruits of this work under an Apple logo but very unlikely on a PC."

I have actually heard that Raycer burnt through closer to $50 million, but whatever the case may be, you just have to figure that to do a competitive 3D chip, at any level, is going to require tens of millions of dollars of investment. I go back to what I said last week, again. There's not room for that many players in the industry so, the next eighteen months are more a question of execution than, any nominal gains in performance. I can't see anyone doing a dramatic leapfrog in technology in the mainstream market, and the high-end is going to be a bloody price war because, no one seems to be buying $2,000 OpenGL accelerators. At the prices that graphics boards are going for these days, low, mid, and high-end, you have to really, really, really crave specific performance and features to pay top dollar for graphics. You hear me Quake freaks?

ATI's Sales and Gross Profits

Figure 1: ATI's gross margins have taken a slight dip after steadily rising in the last couple of years, but margins are still impressive in light of the dramatic slope of revenue growth. Information taken from ATI company records.

You know, the saddest thing about all of this is that there's probably a lot of good ideas out there for graphics hardware, and if there was more money flowing into the start-ups, and smaller niche players, the consumer would ultimately benefit. Graphics silicon is so much more interesting than anything else out there, as Sony will attest to. In the PC space, innovation is going to lose out to process. By that I mean, it's all going to be about product positioning, timing and delivery, and hitting a very specific, and obvious feature list. You want to run a graphics chip company then, go sign up for Six Sigma program.

Real3D - an interview with Jon Lenyo in late 1998

I thought this might make an interesting read. If nothing else, it points out the thinking behind much of the contemporary technologies you see in the 3D market. This is timeless stuff.

These are excerpts from an interview I did with Jon Lenyo of Real3D last year which might help shed some background light on the company. There is also an archive article at www.smokezine.com in the graphics section which gives another spin on what follows, and Real3D. The emphasis here is on the arcade systems that Real3D was developing for Sega. As usual, my questions are in italics.

If you can give me something like a brief case study of how you got into the arcade business:

Back in 1990, GE Aerospace directed its operating units to look for ways to leverage aerospace technologies into "commercial adjacent markets". The GE Aerospace unit responsible for simulation at that time was GE Aerospace Simulation & Control Systems Department (GE/SCSD), located in Daytona Beach, Florida. GE/SCSD designed and built advanced simulation systems for military training and research. The systems were multi-million dollar devices which accurately replicated tanks, helicopters, and fixed-wing aircraft. The goal was to put a trainee in a simulator and make it realistic enough so that he was convinced he was sitting in his M1 tank, in the desert in Iraq, at night, and a hostile T-72 Iraqi tank was 3000 meters away swinging its turret towards him. The crew compartment was an exact replica of the inside of an M1 turret-- the gun breach is there, the gunsites are there, the fire control computer functions, the sounds of the engine and turret motors are real, the controls are located in the right place, and everything has the right feel and function.

The illusion is not complete unless the trainee is also given the correct visual cues. A simple and obvious concept, but something that took years to develop. GE/SCSD had perfected the visual imagery required for simulators over the years in its Compu-Scene(r) family of image generators. Compu-Scene devices produced computer generated scenes that were accurate in the visual, radar, and infrared spectrums for use in simulators. Just to put GE's vast experience related to 3D graphics and image generation in context, they actually got into the simulation business as part of their contracts with NASA to support astronaut training for the Apollo Program back in the mid-1960's. The very first Compu-Scene image generator was built to train astronauts how to dock the Apollo Command Module with the Lunar Module when it was discovered that the black & white TV camera/model board system in use was inadequate. This first Compu-Scene system had something like 16 polygons, ran at 60 Hz, and filled up a large room with electronics equipment. GE saw the potential for computer generated imagery to be used to train pilots in the military and started to invest into perfecting and improving the technology.

Compu-Scene systems in 1990 typically sold for over $2 million dollars. The customer base was pretty much limited to the Defense Department and the research departments of major aerospace companies. It was in 1990 that GE started asking the question - "What other markets could we apply simulation and image generation technology to?"

Real-Time 3D Graphics

That question prompted us to start a market research project to look at markets which would find real-time 3D graphics useful. Things like medical imaging, engineering workstations, theme park rides, and entertainment were evaluated. In the area of entertainment, we researched home video games, arcades, locationbased entertainment, and theme park rides. The unanswered question at the time was how to take multi-million dollar technology into these very cost sensitive markets.

Near the end of 1990, we asked our demo group to model the Daytona International Speedway(r), which was conveniently located across the street from our facility. To that geometrically correct model, we added a couple of Formula One race cars with reasonably good vehicle dynamics that you could "drive". This whole thing ran in real-time, at 60 frames per second, with about 6,000 polygons per frame, on 1M pixel displays, on a Compu-Scene PT2000 image generator which sold for about $1.5M. We made a cool looking video tape with music and sound effects and hit the road.

We visited Sega in November of 1990. At that time, all Sega's arcade games were based on sprites. They happened to be developing their very first polygon-based graphics system, which they called Model One. Model One was pretty crude by today's standards, but at the time it was pretty good. By comparison, it looked a lot like the Compu-Scene systems from the Apollo Space Program days, except that it went into a $15K arcade games instead of a multi-million dollar simulator. We showed Sega our video tape which demonstrated real-time, trilinear texture filtering, shading, and a host of other 3D graphics features and they became very interested in the GE technology.

To make a long story shorter, GE/SCSD was interested in adapting its Compu-Scene technology to non-military markets and Sega was very interested in improving the graphics quality of its arcade games. This was the beginning of a relationship that continues to this day. GE/SCSD would adapt its graphics technology to Sega's proprietary arcade system. The result was what Sega calls the Model 2 arcade graphics system. Model 2 brought real-time textured mapped polygon graphics to Sega's arcades. The structure of this relationship was and continues to be very good. GE/SCSD knew real-time image generation and Sega knew how to keep the cost down, the experience fun, and already was number one in the arcade machine marketplace.

The relationship with Sega has continued through the 1990's as GE Aerospace became part of Martin Marietta, who then merged with Lockheed to form Lockheed Martin, who then formed Real 3D to commercialize this vast portfolio of graphics technology. And even Real 3D has undergone a recent change. Real 3D, Inc. is now a new corporation which is owned jointly by Lockheed Martin (80%) and Intel (20%). In addition to custom graphics chips and board designs for Sega, Real 3D is bringing its real-time 3D graphics technology to the PC and workstation markets.

Graphics Features for Sega

How you went about developing the systems for Sega:

The systems we developed for Sega were based on a cooperative development program. We proposed to Sega a list of graphics features we thought would be good to incorporate into their arcade systems. Sega would come back with a design to cost and schedule requirement, and essentially we would work together to achieve the best balance between technology, cost, and schedule. From a technology perspective, we have an unmatched portfolio of graphics intellectual property and experience. When we proposed something to Sega, all we had to do was take a Compu-Scene system and deactivate certain features to show them what Model 2 would look like, in real-time, before we even started designing anything. Remember, we were bringing graphics technology from the very high end down to the arcade. We were very accustomed to military specifications and the performance requirements that came with meeting these specifications. Because we had done it many times before, Sega had great confidence that we could indeed adapt our high-end real-time image generation technology to a lower cost arcade graphics system. At that time, only maybe two other companies in the world could do what GE/SCSD could, and neither of these companies had the patent portfolio or experience that we had.

What were the technical milestones you had to achieve:

Compared to the graphics performance that we were routinely delivering in our Compu-Scene product line, the graphics performance of Model 2 was not a significant technical challenge. The challenge was moving from a mindset of designing multi-million image generators, with a volume of 40 or so systems a year, to designing a graphics system that goes into an arcade game machine that sells for $15K with annual volumes of 65K units.

That is not the case today. The graphics features we designed into Sega's Model 3 arcade graphics systems are very leading edge, especially compared to what the PC market thinks of as leading edge, or "hot" 3D graphics. There is an order of magnitude difference in performance levels, and we think that trend will continue for some time. In other words, while PC graphics will continue to make great strides in image quality and fidelity (thanks in part to Real 3D's efforts with Intel), 3D graphics on the desktop trails the arcade platform by a wide margin in terms of raw performance, level of detail, and image quality. We would anticipate the arcade platform will maintain this advantage for quite some time - even though claims of arcade-like graphics are being made by some today, the arcade-like graphics they speak of equate to Model 2-like graphics. The latest arcade architectures, and certainly the next generation arcade systems in the pipeline right now, deliver graphics performance not possible on the PC platform.

Since this interview, judging by the state of the high-end 3D arcade market, the shift has been towards the PC, and now, next generation consoles. In hindsight, it's easy to say that Real3D's only mistake was not realizing that the PC graphics market is a down and dirty business. Mostly, that has to do with the PC OEMs that dictate the direction the graphics chip suppliers have to go, and frankly, PC OEMs are at the mercy of Intel. So, as a result, the CPU eats up a system's component budget, and every other component vendor suffers as a result. Maybe that's why the iMac is so popular, and is influencing some of the PC OEMs; it's easier to color the box that holds the motherboard then try and do anything out of the norm on the system. Doesn't it all boil down to marketing MHz, and price?

Ultimately, Raycer would have been entering a high-end OpenGL graphics board market in which sub-$1,000 boards rule the day at the expense of heavyweight graphics products from companies such as Intergraph, and E&S.