Commentary: Heat continues to plague CPU design

Heat is not a new problem in computer and semiconductor design.

It goes back all the way to ENIAC in 1946. In the early 1950s, companies air-conditioned their computer rooms to combat heat. Starting in the 1960s, IBM used water cooling to control the heat generated in its mainframes. In 1980, chip design moved from bipolar to CMOS in part to combat heat problems.

	See news story: Intel CTO: Chip heat becoming critical issue

Heat--and battery life--will continue to enforce a bifurcation in the processor power of different devices. At the low end, PDAs and other pervasive devices will remain limited in their display and processor technologies. However, the main limiting factor with these devices will be the intersection of battery life and form factor/weight, rather than heat.

Portable computers will continue to use faster chips, but these will be designed specifically for lower power and lower heat and will be sized to fit into thin portables. Because these devices will always lag desktops and servers in processing power, they will continue to use thin-client implementations, such as Citrix's WinFrame or proxies, to enable access to power- and bandwidth-hungry server applications.

For instance, some vendors of ERP (enterprise resource planning) software now provide thin clients that enable users to use some ERP functions from PDAs. This underscores the importance of the communications infrastructure to provide the connectivity these solutions need.

Desktop systems are large enough to contain air conditioning if necessary to dissipate the heat of high-end CPUs. Even on the desktop, however, not everyone needs the fastest possible CPU. The average business user or home user does not come close to using the full power of the systems they currently have, except when opening or closing large, complex applications. Adding a faster CPU will make little difference other than shortening start-up times.

The point of diminishing returns
Based on discussions with a large number of customers who have studied performance, and testing with a wide range of machines from P166 through P4-1.5GHz, there appears to be a point of diminishing returns with current business applications at about 800MHz. Above that, there is no appreciable benefit. The vast majority of business systems are currently between 450MHz and 600 MHz, so there is still one more turn of the processor crank, even with current apps.

However, this is a temporary plateau. The next generation of interface technology should arrive in 2003 or 2004 and will place much larger demands on the processor and system capabilities, driving adoption of faster processors. Servers will have even faster CPUs, but their heat problems will be handled by controlled environments and more complex internal design.

One other measure to control heat and power consumption is the manipulation of power supply parameters. Supply voltages have dropped to reduce the heat generated and increase the switching speed of the integrated components. Dynamic manipulation of the supply is gaining momentum as a means to further stretch the speed and heat issues. Intel's Mobile Voltage Positioning (IMVP) is such a technology. Speed and heat are directly related; heat and battery life are also directly related. Shrinking chip features reduces the problem, but architectural improvements, like IMVP, are the true key to future improvements.

For IT organizations, the system with the fastest processor does not always make sense. They need to weigh the real needs of their users and balance cost against overall system speed. In portable and pervasive devices, they also have to consider battery life, display technology, and overall size and weight. A slower portable that has three times the battery life may meet some users' needs better than the fastest machine that only works for a half hour on battery.

Many factors besides CPU speed--including bus speed, memory availability, database architecture, network performance and peripherals such as graphics cards--contribute to the speed of the overall computer system. Therefore, the system with the fastest, "hottest" CPU may not in fact provide the fastest performance at a given time.

Meta Group analysts Jack Gold, William Zachmann, Steve Kleynhans, David Folger, Glenn O'Donnell, Mark Shainman, Timothy Hickernell and Val Sribar contributed to this article.

Visit Metagroup.com for more analysis of key IT and e-business issues.

Entire contents, Copyright ? 2001 Meta Group, Inc. All rights reserved.