I’ve finished reading a remarkable new geek book, “The Race for a New Game Machine: Creating the chips inside the Xbox 360 and the PlayStation 3” by former IBM engineers David Shippy and Mickie Phipps, released last month by Citadel. I’ve also interviewed Shippy, the chief architect of the PowerPC core of the Cell chip, to get answers to some of my own nagging questions that have lingered from writing two books on Microsoft’s Xbox business (Opening the Xbox and The Xbox 360 Uncloaked). These co-authors have left IBM and so are somewhat free to speak their minds about what happened on one of the most interesting chip design competitions of the modern age. The Wall Street Journal wrote a sensational article that made fun of Sony for playing the fool and financing the development of Microsoft’s game processor. I took my time with this one, because I wanted to use my interview with Shippy and my own research to put together a more complete sense of what happened as IBM designed chips for both Sony and Microsoft using some of the same core engineers, who, at times, were separated only by one floor of a building in Austin, Texas.
The book is fascinating because it’s written by one of the few people who had insight both into the creation of the groundbreaking Sony-Toshiba-IBM Cell chip for the PlayStation 3 and the code-named Waternoose microprocessor that IBM created for Microsoft’s Xbox 360. When I tried to get such an interview in the past, IBM turned me down flat and gave me a bland interview with someone several layers above Shippy. I got a sanitized version of the story, but Shippy gives us the contaminated version. For contaminated is exactly how he felt when he was essentially asked to be a double agent in the chip design process.
Sony scared the bejeezus out of Microsoft in March, 2001, when it announced it would spend $400 million on an alliance with Toshiba and IBM. Microsoft hadn’t even shipped the Xbox yet, and here Sony was working on the successor to the market-leading PlayStation 2.
Recruited by IBM fellow Jim Kahle to work on the Cell microprocessor on the joint venture with the two Japanese companies in Austin, Shippy was a key architect of the Cell’s core PowerPC sub processor. Shippy had been a part of IBM during the days of the Somerset Design Center. It brought together the engineers of Apple, IBM and Motorola in their PowerPC alliance to unseat Intel and Microsoft. That ended in failure, but Shippy was up for another alliance that sought to run circles around Intel and Microsoft as well. Sony was creating something from scratch, and for an engineer, that was a chance of a lifetime.
In 2001 the team commissioned engineer Peter Hofstee to calculate where they had to be in order to make sure that their chips were competitive against the best that Intel could do in 2005. Hofstee told the team it had to hit 4 gigahertz in order to keep competitive. At the time, the fastest chips ran at 1.5 gigahertz. On top of that, the power consumption had to be much lower than the current.
Then one day, Shippy’s boss Chekib Akrout, an IBM senior vice president responsible for the IBM Cell team, called Shippy into his office. Akrout told Shippy that another customer wanted to use the core for the super-secret Cell chip. That was always the plan, as a mass-produced Cell chip would be much cheaper for Sony if there were more customers for it. IBM planned to use it in servers and Toshiba would use it in TVs. (Toshiba finally showed off a Cell-based TV at this year’s International Consumer Electronics Show by the way). But Akrout then dropped a bomb. The other company was Microsoft, which would use the PowerPC core in its Xbox 360 game console.
Microsoft was two years behind Sony and was in a panic. It had tried to go to rivals such as Intel and failed to secure an agreement for a low-power but high-performance processor. Intel just didn’t make such chips in those days, as its chips were power-hungry. But Akrout said Bill Gates of Microsoft called up IBM CEO Sam Palmisano and essentially offered to pay a price that Palmisano couldn’t refuse. More than a billion dollars was at stake.
Microsoft looked at all of the processor cores IBM had to offer, but it didn’t like any of them, at least until an IBM engineer named Adam Bennett revealed a core that was still in development in early 2003. It was the same one that IBM was developing for use in Sony’s Cell processor. Akrout, IBM executive John Kelly, and others decided to offer the core the Microsoft with modifications. But Kahle felt IBM was being reckless, showing a design that was destined for its partners to one of their rivals.
The problem was that Microsoft wanted to get to market at exactly the same time as Sony, in the fall of 2005. This was 2003, and chips normally took three years to get to market from scratch. IBM was going to cut it close because Microsoft was in a panic. Shippy recounts the history of Microsoft’s thinking, which I told in my own books. (I was irked that the book doesn’t footnote my own, and Shippy said he would fix that if the book has a paperback edition).
Shippy’s reaction was one of emotional dismay. He was asked to be a traitor to his partners at Toshiba and Sony, whom he had worked with every day for a couple of years. But this was the chance for IBM to sew up the video game industry –- making chips for Nintendo, Microsoft and Sony –- and put Intel into a state of misery. Jim Kahle, who recruited Shippy, stuck with the PS 3 chip project and resented how his friend took up work for the enemy. Other team members felt the same way. But Shippy took the view that this was good for IBM, that the company could keep the walls between the projects separated, and that it made perfect sense. He said, for instance, that Intel probably never thought twice about selling the same chip to Dell as it did to HP.
These emotional issues make the book more than a dry recitation of the process of designing modern chips. There is a fair amount of chip-design talk, but the authors paint the picture with analogies that most folks should be able to understand. I reported a story years ago for the San Jose Mercury News that Sony had gotten some patents for the Cell microprocessor for a future gaming machine. Inside IBM, Shippy says that the news stunned Kahle and others on the IBM team. He went to the Sony engineers and asked why they had patented work that had also been done by IBM without telling IBM. He threatened a lawsuit, and Sony amended the filings to name IBM engineers on the patents. Those patents gave Microsoft engineers clues to what they should do and what they should avoid in order to keep out of legal trouble with Sony.
Shippy and Phipps do a good job of painting the drama in the chip work. IBM vice presidents bickered over why solutions from their own divisions weren’t used in the two projects. As the vice presidents hovered over Shippy, he felt like he was on the witness stand in a “torture room” as they and others lobbed verbal grenades about the progress or lack thereof on the project. Shippy had numerous run-ins with Albert Randall, an IBM executive who allegedly spewed hostility in meetings because Randall felt the Austin team favored the Sony project over Microsoft’s. Shippy was virtually one of the only engineering managers who straddled both projects and so his loyalty was often tugged in two directions.
Given how closely Shippy worked with Microsoft and Sony, you’d think the book would be perfect grounds for Sony to file a lawsuit against IBM. IBM says it kept the boundaries clear, but Shippy confesses that he felt contaminated in meetings with Microsoft. Rumors commonly crossed the boundaries on the companies that were working on different projects. In one hilarious moment, Shippy had to quickly relocate a meeting near a cafeteria to a more remote room because Microsoft’s engineers were coming to visit, and Shippy had to make sure they didn’t run into any Toshiba or Sony engineers in the common areas.
In some ways, I’m stunned that IBM would allow Shippy to write this book. When I asked him about this, Shippy said he was careful not to reveal any trade secrets. He also says that IBM has not screened the book or pressured him about it in any way. He wanted to capture the picture of the interrelationships between characters, to talk with pride about IBM’s great achievement, and to inspire other young people to go into engineering.
There are moments where I can tell Shippy’s holding back. He acknowledges that the conversations in the book are reconstructed to the best of his and his co-author’s memories. The lines that flow from the different antagonists in the engineering discussions are just too grammatically correct. Shippy says he did not keep a diary that helped him reconstruct the conversations. He is plenty critical of certain IBM people in the book. “We all felt that IBM had violated many of its core business practices in jockeying both horses in this particular race.” But in the end, Shippy comes off as a patriot with Big Blue blood in his veins. He felt like he had the weight of an $80 billion corporation’s future on his shoulders.
In a way, IBM was betraying a partner. On the other hand, it was exercising its legal right to resell a core that could be customized into something else. Shippy decided that he had to get over the ethical problem and just try to make the best possible technical achievement that he could. The rest of his team grumbled, but fell into line.
The great failure of Ken Kutaragi, the head of Sony’s game business, and his lawyers was that they failed to foresee how their competitor could catch them. IBM failed many years back to acquire all rights to Microsoft’s MS-DOS operating system. Bill Gates turned around and licensed it to all PC makers, creating an IBM PC clone industry that eventually allowed Microsoft and Intel to usurp all power from IBM. In this case, Kutaragi could have asked for exclusivity in the video game business for the PowerPC core, but he was blindsided. Sony’s failure to bar IBM from selling its own chip technology to Microsoft — which launched before Sony could use the chip technology itself — is astounding. This will go down as one of the worst business decisions of all time.
But as Microsoft’s engineers told me for my own book, they asked for a lot of modifications to that chip. They were aware of Sony’s own patents for the Cell processor and knew what they wanted to do that would differentiate the core. So Microsoft’s head engineer, Jeff Andrews, asked for changes and wound up with a very different microprocessor. The core was the same, but there were three cores on Microsoft’s chip and just one on Sony’s. And Microsoft had asked ATI Technologies, now part of Advanced Micro Devices, to fashion its own graphics chip. Microsoft engineers didn’t want anything to do with the Cell architecture or the vector units. They wanted much more powerful vector processing on the main cores. In the end, the Microsoft Waternoose chip was a custom microprocessor.
There was one IBM engineer who had his name both on Cell microprocessor patents as well as Xbox 360 microprocessor patents. But that didn’t mean that Microsoft completely copied Sony’s chips. Microsoft’s final chip had 165 million transistors, while Sony’s Cell had 234 million. You could say that Microsoft stole Sony’s chip, as some superficial book reviews say, but it’s just not true. Shippy agrees that angle is “overplayed” in the press (such as the Wall Street Journal story).
“That wasn’t the reason we wrote the book,” Shippy said. “IBM’s own core technology started even before Cell. We developed high-frequency, circuit design techniques at low power. That enabled us to create these supercomputers on a chip. It’s over sensationalized that Sony funded the Xbox 360 chip.”
Having said that, Shippy does a very good job of explaining how it felt to work on both chips at the same time. The tension led to a lot of stress for engineers who were already on a death march schedule. And this is where the writing bears most resemblance to the classic 1981 book, The Soul of a New Machine, by Tracy Kidder, which won the Pulitzer Prize for its detailed description of the making of a Data General minicomputer. Shippy and Phipps say that book was their inspiration.
Jim Kahle, the manager in charge of the Cell chip, had to make tough calls. He decided, against Shippy’s advice to make changes to the design that delayed the schedule. Early on, Kahle ran into Kutaragi’s intransigence. Kutaragi ruled Sony’s game division with an iron fist, says Shippy, while Microsoft clearly delegated decision making. Kahle wanted the team to put six synergistic cores, or vector processing units that could handle math operations, into the Cell. But Kutaragi insisted on eight, Kahle related to his team, because “eight is beautiful.”
Kahle also had to decide to dispense with a feature dubbed “out of order processing.” This is a more complex way of handling computation. It makes for better performance but comes at a steep price in cost and complexity. That led Jon Rubinstein, who was then an executive at Apple, and Bob Mansfield of Apple to scream bloody murder. It meant that Apple would likely still fall behind Intel in microprocessor performance. And it was one of the decisions that led Apple to defect from IBM’s PowerPC architecture to the Intel platform. This caused a huge shift in the bedrock of the computing industry. I saw all of this happening from the outside as IBM jilted Apple in favor of Sony. But it’s interesting to see the names and circumstances under which the decisions were made. In Kahle’s defense, the decision was necessary to keep the Cell chip on track. IBM also was very heavily focused on server chips, rather than serving Apple. In other words, there were other things about the IBM-Apple relationship that led Apple to go to Intel.
Kahle and Shippy also had to deal with the derision of IBM’s server chip designers, who insisted their chips should be used in the game consoles. But Shippy laughed that off since he needed extremely high performance balanced with power efficiency. The server chips threw off 100 watts of power, while the game chips could tolerate 10 watts. Otherwise, the systems would need all sorts of expensive cooling systems, fans, and would still probably melt down your living room.
One of the enlightening technical details in the book dwells on an important issue for Microsoft. Very late in the game, Redmond’s engineers asked IBM to do what it could to support emulation, or the ability to run older Xbox games designed to run on the original Xbox with an Intel chip. In my own book, I wrote about how Microsoft went back and forth on this problem. Microsoft’s marketers discovered that a huge number of Halo 2 fans were playing the old game and were likely to continue playing the game online for a long time. That meant that the Xbox 360 had to be somehow compatible with the original Xbox games or Microsoft would anger its most loyal customers. IBM had to write some new instructions to support the emulation. Shippy says there were fights about adding this support, with IBM managers calling bullshit on each other. IBM could have added hardware support for emulation in the second round of debugging. But the company’s top ranks eventually quashed the idea as far too risky.
Amazingly, IBM’s engineers finished the Microsoft microprocessor in 11 months. Again, this was partly because Sony had commissioned work on the core years earlier. IBM finished both the Cell and Microsoft chips on time in September, 2004. But that just meant they were ready to go to the factory and could then begin months of revision and prototyping. IBM had to debug both chips at the same time. Some engineers didn’t realize that some of their bug fixes would actually benefit their rivals. The chips were similar enough that IBM could use a single team to debug both chips, though the engineers had to be careful how they applied fixes to each chip and be wary about spilling secrets from one chip to the other.
Sony’s game developers were at a huge disadvantage. Since there was nothing like the Cell out there, Sony couldn’t get the game developers started on making games until it had chip prototypes ready. Jim Kahle, on the other hand, suggested that Microsoft use the current PowerPC 970 microprocessors that were being designed for Apple as the prototypes for Microsoft’s game console. That’s why Apple wound up shipping a lot of its best desktop computers to Microsoft, which gave them to the game developers to use as development machines for Xbox 360 games. So Microsoft game developers had a year’s head start on the Sony game developers.
IBM was actually surprised that Microsoft exercised its contractual right of lining up a second source of microprocessors. It hired Singapore’s Chartered Semiconductor to make the IBM-designed chip and delivered chips to Microsoft a month earlier than IBM. Sony, meanwhile, had no such luck. It ran into delays for its chip at the factory and, as a result, Microsoft received working chips well before Sony did.
When IBM received the chips in January, 2005, a group of engineers gathered. The chip “had a heartbeat,” Shippy said, as if it were his own baby. When the engineers tried to boot the Windows kernel, or core software of the operating system, it hung. They called Dinarte Morais, a Microsoft engineer who almost single-handedly wrote the operating system for the Xbox 360 and debugged it over the phone as Morais was watching the Super Bowl. Then the chip booted properly. Shippy got to play the first game on the chip.
Sony’s first chips didn’t arrive until late February. In the middle of the night, the engineers got the Cell chip to boot and they celebrated with champagne. The PS 3 chips ran at over 5 gigahertz. The next day, as he stepped out of the elevator, Shippy realized he was in the elevator with representatives of Sony, Toshiba, Microsoft and IBM. A sign said, “Do not discuss confidential information in this area.”
But while the Sony-Toshiba-IBM Cell chip was done on time, a lot of debugging still had to be done. While some chips could run faster, others could not. So IBM scaled the speed of both the Xbox 360 and Sony chips to 3.2 gigahertz, rather than 4 gigahertz, to ensure that high quantities of acceptable chips could be made. On top of that, Sony made a conservative decision to shut off one of the eight vector cores on the Cell chip. That made the chips slower, but it ensured that the chip would be more manufacturable.
The Wall Street Journal article termed the Cell chip a disaster. But that just wasn’t true. Sony messed up for other reasons. The Cell chip came out on time and was almost ready for a launch. But it turned out other factors, such as Blu-ray and the graphics chip, held up Sony. The lesson wasn’t that Cell was a bad idea. After all, Cell is being used in more new products, from medical imaging systems to IBM supercomputers. Rather, Sony just tried to be an innovator on too many fronts and went a bridge too far. The PlayStation 3 was expensive not because of its expensive Cell chip, but because of the Blu-ray drive and other components that could really only be appreciated by audio visual freaks.
Nvidia had to come in at the last minute and create a graphics chip for the machine, because Sony’s engineers had failed to create a companion graphics chip. Shippy said that delayed the system. The other cause of the delay has been well documented: the specifications for making Blu-ray drives had been held up amid bickering with a variety of parties. Lastly, the software development tools were late, so game makers needed more time to make the launch games. Shippy and Phipps briefly discussed this at the end of the book. It was a huge blow to the chip teams when Sony announced a delay. As a result, the Sony PS 3 slipped by a year.
True to its history of risk taking, Microsoft authorized a manufacturing launch against the advice of IBM. It authorized the beginning of chip building even though the design wasn’t fully debugged in the labs. Shippy says it was typical of Microsoft’s attitude on shipping Windows, to get it to market first and worry about the bugs later. Akrout, the IBM senior vice president, pleaded for more time. But the holiday launch was at risk. The first chips came back in May, 2005. The gamble paid off, as the IBM chips coming out of the factory worked. Akrout, one of the stalwart executives on the projects, resigned in April, 2005, mainly due to politics. That shook Shippy to the core. But Akrout landed safely at Advanced Micro Devices.
The Cell and Microsoft microprocessors were huge undertakings that generated more than 500 patents along the way. Sony’s mistakes have cost it huge market share losses in games. Nintendo has taken the No. 1 spot thanks to its unusual motion-sensor-based controller design for the Wii, but Microsoft has moved into the No. 2 spot and Sony is a distant third. Sony has had to scale down its business, cut a lot of game developers, and it sold off its chip manufacturing operations to Toshiba. Kutaragi lost his chance to be CEO of Sony because of the turnabout, and he retired after 30 years at Sony.
Phipps left IBM first, and then Shippy left in early 2006, after all the work was done, to rejoin the chip start-up Intrinsity. Their work was done. Shippy said he wanted a smaller design team, less bureaucracy, shorter schedules and a smaller company environment.
Meanwhile, Microsoft’s culture of risk-taking ran aground as defects known as “the Red Ring of Death” brought down one Xbox 360 after another. (See our series on the Xbox 360 defects here.) The problem was mostly in the hot graphics chip that ATI made and how it was attached to the main circuit board. But Shippy wasn’t enlisted to help fight that fire. So much so that he declined to comment on it when I asked.
Looking forward, Shippy believes it is inevitable that both Sony and Microsoft will try to reduce costs by combining the two critical chips -– microprocessor and graphics -– into one. That’s what Sony eventually did with the PlayStation 2. But that’s not an easy task, and it means that warring companies will have to work together. IBM and Nvidia, for instance, would have to work together to combine the Cell chip and the graphics processor for the PS 3 into one chip. For the next generation, Shippy believes that by 2011 or so the companies will likely launch a new round of game consoles. Those machines are likely to combine microprocessors and graphics chips in the same machine right off the bat. The chips in those machines will have somewhat faster frequencies, but the big performance boost will come from having lots of cores working in parallel on the same chip. AMD says it is doing just that with its Fusion chip project, and I have no doubt that someone like Akrout would try to get it into the PlayStation 4 or the next Xbox machine. It’s going to be a lot of fun to watch. In the meantime, it would be nice if Shippy’s confessional book triggered similar stories from the other players.
For more on the games business, check out VentureBeat’s upcoming GamesBeat 09 game conference.