Transmeta's Crusoe

Authors

Authors: Ravi Madapati
Faculty Member
ICMR (IBS Center for Management Research).

Links

Free Resources,
Articles,
Book Reviews,
Interviews

Tools

Market Acceptance

A 'megahertz treadmill' drove the entire computer industry. New processors with new megahertz ratings appeared on the market, with overwhelming rapidity. Sometimes, even before the current model arrived at retailers' shelves, it had to be marked.

Consumers often complained that the PC they had bought in one week became obsolete by the next week. The treadmill effect periodically played havoc with computer manufacturers' supply chains. Crusoe aimed at getting around this problem. It offered just enough megahertz to deliver a satisfying user experience. Manufacturers could plan for long production runs that allowed superior products to be built, while driving down costs. In addition, Crusoe's software could be used to deliver extended and enhanced benefits of longer battery life and cooler operation without having to re-configure the underlying hardware. Crusoe promised the performance for a range of applications from office suites to multimedia. Desktop processors and systems were optimized for peak performance. In mobile systems, the situation was quite different.

The issue of battery lifetime entered the picture. Crusoe had been designed around the belief that running the processor faster than necessary to get the job done was a waste of energy and reduced battery lifetime. Even before Crusoe had hit store shelves in 2000, critics were already skeptical about its performance. Tests suggested that the chip sacrificed speed for longer battery life-at a price that might not be worth it. But Transmeta claimed that these tests could not be applied to Crusoe because of Crusoe's unusual architecture, which combined software and hardware. Hence the company claimed that standard benchmark tests that were used to gauge the performance of hardware-based processors like Pentium were unlikely to capture the benefits of Crusoe.

Indeed this was the reason Transmeta gave for not releasing benchmarks in the first place. During operation, the chip recompiled x86 code into native Crusoe code. This slowed down performance drastically. However, Transmeta emphasized that the recompiling process only had to be carried out once so that the operation speeded up during the course of use. Transmeta maintained that while doing a repetitive task, performance went up after a couple of iterations, so Crusoe was fast enough for typical applications, and far more power-efficient than similar processors from Intel and AMD.

Next >>