Microsoft and Proprietary Chips

April 10, 2009

Stacey Higginbotham’s “Is Microsoft Turning Away from Commodity Server?” here reminded me of a client study I did five or six years ago. The Sony PS3 was working on a proprietary chip. IBM was involved, and I documented the graphics method which built upon IBM technology. In short order, Microsoft and Nintendo signed up with IBM to use its generic chip design for their next generation game devices. Sony ran into three problems. First, costs went through the roof. Sony did not have a core competency in chip design and fabrication, and it was evident even in the sketchy technical information my Overflight service dug out.

Second, the yield on chips is a tricky issue. Without getting into why a yield goes wrong, I focused on the two key factors: time and cost overruns. The costs were brutal, eventually forcing Sony to change its fabrication plans. The time is a matter of public record. Microsoft beat the PS3 to market, and Sony is starting to recover now. We’re talking years of lost revenue, not days or weeks or months.

Third, the developers were stuck in limbo. With new chips, new programming tools and procedures were needed. Without a flow of chips, developers were flying blind. The problem became critical and when the PS3 launched, the grousing of developers about the complexity of programming the new chip joined with complaints from fanboys that games were in short supply.

Compatibility, availability, and affordability joined the chorus.

Ms. Higginbotham’s article summarized what is known about Microsoft’s alleged interest in creating its own chips for its own servers. The motivator for Microsoft, if I read Ms. Higginbotham’s article correctly, is related to performance. One way to get performance is to get zippier hardware. With faster CPUs and maybe other custom chips, the performance of Microsoft software would improve more than it would by using Intel or AMD CPUs. (Google uses both.)

For me, the most interesting point in her write up was:

The issue of getting software performance to scale linearly with the addition of more cores has become a vexing problem. Plus, as data center operators look for better application performance without expending as many watts, they are experimenting with different kinds of processors that may be better-suited to a particular task, such as using graphics processors for Monte Carlo simulations.

She did not draw any parallels with the Sony chip play. I will:

  1. The Sony Ken Kutaragi chip play provides a good lesson about the risks of rolling your own chips. Without a core competency across multiple disciplines, I think the chance for a misstep is high. Maybe Microsoft is just researching this topic? That’s prudent. Jumping into a proprietary chip may come, but some ramp up may be needed.
  2. Google does many proprietary things. The performance of Google’s system is not the result of a crash project. Time is of the essence because the GOOG is gaining momentum, not losing it. Therefore, the Sony “time problem” with regard to the Xbox may translate into years of lost opportunity. Chip designs are running into fundamental laws of physics, so software solutions may reduce the development time.
  3. The performance problem will not be resolved by faster hardware. Multiple changes are needed across the computing system. There are programming slow downs because tools have to generate zippy code for high speed devices. Most of the slow downs are not caused by calculations. Moving data is the problem. Inefficient designs and code combine with known bottlenecks to choke high performance systems, including those at Google. As the volume of data increases, the plumbing has to be scalable, stable, and dirt cheap. Performance problems are complex and expensive to resolves. Fixes often don’t work which makes the system slower. Nice, right? Need more data? Ask a SharePoint administrator about the cost and payoff of her last SharePoint scaling exercise.

My view is that one hire does not a chip fab make. Microsoft’s analysts have ample data to understand the costs of custom chip design and fabrication. Google requires immediate attention and rapid, purposeful progress on the part of Microsoft’s engineers. Time is the real enemy now. Without a meaningful competitor, Google seems to enjoy large degrees of freedom.

Stephen Arnold, April 10, 2009


Comments are closed.

  • Archives

  • Recent Posts

  • Meta