Intel Tracks Power Costs, Pushes Performance

The company is working to build benchmarks for power consumption and releasing a new instruction set to increase server performance.

SAN FRANCISCO—As part of Intels effort to be more open about its technology, the company plans to share how much it costs to power a PC or a server for a year.

The chip maker at its Intel Developer Forum here announced a series of moves aimed at helping IT managers take stock of the annual electrical costs of their PCs and servers.

For example, Intel is working with BAPCo (Business Application Performance Corporation) and SYSMark, two benchmarking bodies, to establish new standards of measurement for power consumption by client PCs and servers, respectively, while also committing to publishing its own stats.

The work arrives at a time when rising electrical rates, coupled with increases in the deployment of power-hungry servers, is causing major concerns about electricity consumption for many large businesses.

Meanwhile, Intel and its chip-making rival Advanced Micro Devices are heavily emphasizing performance per watt, a measurement of how much work a chip can accomplish for a given amount of energy it consumes, over delivering raw performance.

The benchmarks for PCs will grow out of work done by Intel and others to create measurements for notebook batteries, Patrick Gelsinger, general manager of Intels Server Products Group, explained in a keynote address at IDF.

"Much like [the] auto industry, with miles per gallon, we needed a way to measure" notebook power and battery life, he said. "A BAPCo [battery mark] standard was created. But we have no equivalent today with clients."


That will change now: BAPCo and Ecma International, two performance measurement bodies, will work along with Intel and others to create EECoMark, a measurement for PC energy efficiency based on performance, Gelsinger said.

EECoMark tries to measure how a typical PC would be used in a typical workday in order to determine how much energy the machine consumes and thus rate its efficiency, he said.

"What we are doing is taking a workload ... and looking at how that workload would occur and actually measure the energy cost of that platform throughout the workday," Gelsinger said.

Efficiency measurements can also be extrapolated to show annual power consumption and thus determine annual electrical costs, he said.

An energy-efficient PC—a machine based on Intels Core 2 Duo—would consume about $23 worth of electricity per year, based on the North America average of 10 cents per kilowatt hour, while others consume up to $44 worth of energy, Gelsinger said, illustrating this with in a chart. Spread over a number of machines—many companies have thousands of PCs—the figures add up to a significant cost.

"Its great to see the industry coming together" to focus on an energy consumption measurement standard, Gelsinger said. "Were also working to establish the same metrics for the server platform."

To that end, Intel is working with SPEC (Standard Performance Evaluation Corp.) to create a measurement for server energy efficiency.

/zimages/4/28571.gifIntel touts the advantages of quad-core chips. Click here to read more.

The measurements are expected to come out in the first half of 2007, but Intel knows that its going to take a while for the Ecomark and the SPEC server power measurements to be finished, Gelsinger said.

"What were committed to doing today is … measuring ourselves and publishing the results," he said. That way, the industry can measure Intels data against Ecomark, for example, giving additional points of reference for PC energy efficiency.

The Intel efforts come after similar moves by companies such as Sun Microsystems, Advanced Micro Devices and others toward choosing a standard measurement for taking stock of server power consumption.

Meanwhile, a group including AMD, Dell and IBM has launched Green Grid, which focuses on creating best practices for designing power-efficient data centers.

While power consumption is a major concern for IT managers, so is performance. Intel is also making several efforts to increase server performance, including opening its front-side bus to accelerator chips, working on new chip-to-chip interfaces and developing more advanced, performance-enhancing processor instructions for 45-nanometer manufacturing.

To that end, Gelsinger announced SSE4, the fourth iteration of the SSE instruction set. SSE4 is made up of a set of 50 new processor instructions Intel will deliver over the next two years to speed up the processing of functions like rendering multimedia and serving Web pages or applications. Gelsinger also disclosed plans to open Intels server platforms to application accelerator chips and to create a new interconnect, based on PCI Express, with IBM.

Intel will extend the front-side bus, its pathway for shuttling data into and out of its processors, by allowing chip makers Altera and Xilinx to develop application accelerators for its servers. This gives the two chip makers the chance to work with Intel customers and build a broad range of accelerators.

Finally, Intel and IBM will deliver a new interconnect, dubbed "Geneseo," a PCI Express-based interconnect that is designed to speed up the pathway between the processor and application accelerators. The specification could roughly double bandwidth and cut latency, and therefore increase performance.

Intel and IBM will take the approach to the PCI Special Interest Group, which manages the PCI specification.

Gelsinger also highlighted Intels new single-socket Xeon 3000 chip, a Core 2 Duo-based chip designed for single-processor servers.

Intel will continue to use FB-DIMM (Fully buffered dual in-line memory modules) for its dual processor and for the above server platforms between now and 2009. Its Xeon 3000 line, however, will use DDR3 (double data rate 3) DRAM (dynamic RAM.)

/zimages/4/28571.gifCheck out eWEEK.coms for the latest news, views and analysis on servers, switches and networking protocols for the enterprise and small businesses.