AMD vs. Intel: Barcelona’s Slight Energy Advantage

Here we go again. In the inexhaustible game of leapfrog between Silicon Valley chipmakers Advanced Micro Devices (AMD) and Intel Corp (INTC), it’s AMD’s turn to jump. Today, after more than six months of delays, the Sunnyvale, Calif. underdog of the multiprocessor arch-rivalry finally announced the launch of its quad-core processor “Barcelona” chip.

The release of the Quad Core Opteron chip series certainly has market share implications for AMD, but we’ll let Om fill you in on what it means for the business. We’re more interested in just how power-efficient AMD managed to make its new chip.

An AMD spokesperson claims that the 75-Watt 2 GHz Barcelona processor, when compared to Intel’s 2.3 GHz Xeon processor, is on average 26 percent faster across a variety of server workloads, while managing to draw less power and require less cooling than Intel. In one given server workload test, AMD’s new processor proved to be “65 to 70 percent faster” while using “five less watts, on average,” of energy compared to Intel’s Xeon processor.

That’s using AMD’s new energy consumption metrics, called ACP, or average CPU power, that includes more than just the processor cores when determining energy usage of the microprocessor. Since Intel doesn’t have an integrated memory controller, it “allows the chip maker to post lower energy consumption data for its chip,” but it also “draws additional energy to the chipset and impacts performance,” says Tom Sanders of VNU Net.

Meanwhile, AMD’s “numbers are a little fuzzy,” says Roger Kay, president of research firm Endpoint Technologies. Right now the only stats on Barcelona’s power-per-watt efficiency are from AMD, and we can’t trust those stats entirely without any third-party research. Mercury Research President Dean McCarron agrees. “The way these things get measured there’s a lot of room to alter the results,” he said.

Even so, AMD is touting the power efficiency of its new chip. How does AMD reduce the energy usage of its chip? Having the quad-core processor on a single die, the company explains, allows the cores to “efficiently communicate and perform work without having to leave the processor package.”

The chip’s technology also reduces energy consumption by allowing processors and cores to “operate at various voltages and and frequencies, depending on usage.” When a chip has four cores, if all of them are draining energy all the time, that’s a major energy waste. But if a chip can turn off unused parts of the processor, it will be more energy efficient.

Kay tells us that by using AMD’s “direct connect” memory architecture, the company’s latest chip is able to do its processing work without “much energy overhead.” But Kay said the difference is a slight one.

“On a raw performance measure they might have a slight edge at the introduction of Barcelona, but the performance-per-watt edge might be more durable. It might last a little longer because architectural changes are difficult to make. Intel is not going to incorporate [their version of] direct connect memory until 2008. But most advantages in this industry are really temporary.” – Roger Kay, Endpoint Technologies.

When it comes to IT purchasing decisions, the power drainage of a chip is almost, if not equally, as important as processing speed. The energy consumption of servers and data centers has doubled in the past five years and is expected to nearly double again over the next five years, to more than 100 billion kWh, according to the EPA, at a cost of about $7.4 billion a year.