Processor vendors increasingly reflect the need for a more balanced approach in the way they convey performance. AMD now names its AMD Athlon XP processors with model numbers that indicate the performance of each processor relative to other AMD Athlon XP processors and the prior AMD Athlon family
of processors. The company adopted model numbers to discourage megahertz-to-megahertz comparisons. Motorola, supported by its system partner Apple, has historically tried to balance clock speed and efficiency in its PowerPC processors through a number of architectural innovations, including its Altivec technology for multimedia processing and the use of multiple levels of cache. Even Intel, the leading proponent of megahertz, has long used architectural innovation (such as advances in front-side bus speed) to improve the efficiency of its processors. Intel also is raising public awareness by pouring substantial resources into its mobile processor, code-named Banias, which operates at a lower clock speed than the current Pentium 4 processors. Intel will market the processors on factors like long battery life, small size, low weight, and wireless connectivity, features for which clock speed cannot be the sole measure. The need for a more balanced approach to performance will grow as future processor and system architectures continue to diverge. For example, Transmeta’s upcoming 256-bit VLIW processor, code-named Astro, will break further away from the industry’s standard processor architecture and the AMD Athlon 64 processor, with an integrated memory controller, will innovate further away from the industry’s standard partitioning.
Challenges
While the PC industry’s products have evolved, its performance measures have not. As a result, the industry has lost the ability to communicate product performance effectively. Changing this situation will be a significant challenge for the industry because the standards are very high. IDC believes that the PC industry owes buyers a new measure that is:
•System-based. PC performance is about the whole system of interdependent components, not just the processor.
• Simple. Any new measure must be as easy to understand as clock speed.
• Flexible. Novices should be able to look at just one or a few numbers. Advanced users, however, should be able to go deeper into their research on PC performance, should they choose.
• Tailored to usage models. In order to be useful and relevant to buyers, a measure must reflect how a PC will be used. The need for a more balanced approach in the way they convey performance.
Changing this situation will be a significant challenge for the industry because the standards are very high.
• Built around clock speed and efficiency for delivered application performance. Processor and system architectures reflect a series of different design decisions. Designers will continue to rely on both factors to improve application performance.
• Consistent. While underlying tests will evolve to enhance the measure’s ability to convey performance, the actual measure presented to PC buyers must remain consistent.
• Repeatable. Allowing for margin of error, the underlying tests run on a PC must give the same or similar results when run again independent of who is doing the testing.
• Transparent. In order to be credible, the underlying tests and the testing methodology must be open to scrutiny.
• Given broad industry support. PC buyers should only accept a single, unified method. It’s not acceptable to have competing methods from multiple sources and different backers in each camp.
• Administered by a credible, independent party. All parties involved must trust that all products are being measured fairly and impartially.
• Systematically updated. In order to evolve with changing usage models and configurations, the measure would need to be updated accordingly.
PC buyers of all kinds must insist on industry action. When buying a PC, consumers should demand an indicator of how a certain PC is suitable to run their applications and meets their individual needs. To ensure a higher return on investment, corporate and commercial buyers should insist that their requests-for-proposals account for more than frequency when evaluating performance. When lacking a true measure of PC performance, all PC buyers should ask prodding questions: What is clock speed not telling me about the processor’s and the system’s performance? Isn’t good PC performance more than just a matter of the processor? With only clock speed to go on, how am I supposed to gauge how fast my applications will run?
Conclusion
IDC believes that tackling these challenges will benefit the industry by allowing vendors and users to segment the market and help it grow in the future. Traditional scaling is not enough and will not help the industry prosper. Measurements that judge PC performance based on more than traditional scaling will encourage more diverse ways of achieving higher performance. They will also acknowledge that good performance can be many things, such as fast applications, but also long battery life and high frame rate. In the meantime, don’t base your buying decision on megahertz alone and, whenever possible, rely on industry benchmarks that will give you a more accurate picture of the performance of the entire system.
When lacking a true measure of PC performance, all PC buyers should ask prodding questions: What is clock speed not telling me about the processor’s and the system’s performance? Isn’t good PC performance more than just a matter of the processor? With only clock speed to go on, how am I supposed to gauge how fast my applications will run?
