Opening the Door to Keynote at Synopsys Interoperability Forum ... by Tets Maniwa May 20, 2006
Joshi used the Niagara processor as an example of possibilities for changes in design, and his first observation was that "chaos happens." The current trend towards larger SoC designs is causing increases in complexity, power consumption, and costs to achieve greater processing speeds. This trend in design complexity has led to increasingly complex design flows and the nightmare of intercoupled analyses between various effects. Not all bad news, Joshi said. As a result of applying design resources to this chaotic situation, new insights and engineering talent have revealed some measure of order within the morass. Over time, orderliness and innovation result in simplification, he noted. The next-to-last generation of SPARC processor was typical of the general trend. It had 30 pipeline stages and a superscalar architecture, with out-of-order and speculative branches, as well as a very high operating power budget. Sun's Niagara chip, on the other hand, has an all new architecture, only 6 pipeline stages, and no complex branching or speculative execution functions. By simplifying the architecture and design, Joshi said, Sun was able to reduce the design time, costs and power budget for the latest T-1 processors. The new processor combines both multi-processor and multi-threaded functions to improve performance. The chip can run 4 threads through each of 8 processors for a total of 32 simultaneous running threads. Joshi said Sun's software operating analysis indicated that most applications didn't do much floating-point processing, so the chip only has one floating-point processor. As a result of the changes in architecture, the new chip gets about 5x the throughput at one-fifth of the power consumption. And power specification is turning out to be one of the more important metrics, as large users now care a great deal more about power than about other specifications. To demonstrate the magnitude of the issue, Joshi noted that he is responsible for a 12,000 CPU server ranch that consumes megawatts of electricity. A reduction to one-fifth of the power saves his project an amount upwards of $1 million in operating costs per year. Joshi's second observation was that "open" is, in fact, a business model. It grows by reducing variance. The Internet continues to grow by adding about 3 million new users a week, and these users are changing the Internet use model by asking for more participative and interactive experiences, as well as more open-source software. Joshi noted that since its inception, the Internet has changed from a shared academic and military resource to a place to run business, both on-line and legacy institutions. He added, participation increases the number of threads in active use, so the processors handling Web traffic need to have thread-optimized designs. The new Web is built on open technologies and tools, but continued growth will require an active participating community to continuosly contribute innovation to the technology. The Web and any other open-source activities need to have participants and technologies, but also need to have a complete ecosystem. Open-source communities are not about a single person, Joshi said, but about the diversity of a group that has a dynamic need for change and innovation. The community cannot consist of users only, but has to include multiple contributors and a way for the various parties to be financially successful. An open community has to match the desires of the developers - who want to join in, crate new stuff, and get something for free - with those of buyers, who want the latest technologies, a stable environment, and low barriers to adoption and change. Therefore, to be successful, Joshi said, an open business model must add value through services and solutions around an open technology. These types of changes in business models put more pressure on the management of a company, but many software companies are making money supporting a "free" software base like Linux. A new open-source group Joshi referenced in his talk is www.opensparc.org, a Web organization that now holds the technological details of Sun's SPARC processors, including the T-1. This site today has the open-source RTL for the microprocessors, APIs, hardware drivers, and all of the necessary components for a SPARC-based system. The files are under a GPL license; Joshi said he plans to make money on selling the rest of the ecosystem to potential customers. The open-source SPARC makes it possible for developers to run new or existing software on a simulator (Note: Or more likely an emulator ...) and create value limited only by the imagination. The rate of change in markets, Joshi said, is forcing greater reuse and incorporation of existing building blocks into new systems. By having access to the source code of a compute platform, developers can maximize value creation by developing unique IP in silicon and accompanying services. Joshi said, as a result, the RTL from OpenSPARC can be the basis for custom processors and a multitude of variants. As the world moves to more open-source software, Joshi said he would like to see some important changes in third-party EDA tools. Some of the lower-end tools should move to an free open-source model, and all of the tools need to be more interoperable through open standards. Joshi ended by saying that the tools need to migrate to operating systems that can handle multiple threads and CPUs, particularly today as multi-core processors are becoming mainstream. ******************************** EDA industry observer and commentator Tets Maniwa can be reached at maniwa_at_sbcglobal.net
|