AMD: Intel's Conroe power advantage claim based on skewed metric


Austin (TX) - With Intel's release of desktop processors based on its new Core Microarchitecture now expected at the end of this month, early tests from Tom's Hardware Guide are lending credence to Intel's claim that it has regained the advantage in the race for best CPU performance per watt of energy. In advance of Intel's release announcement, AMD has mounted what press managers for a political campaign would call a "pre-spin cycle:" an early response to the claims Intel is likely to make. One of AMD's key bullet points is that the numbers Intel boasts for power consumption, and those AMD boasts, aren't determined on a level playing field.

"One thing you need to be aware of," Mike Field, AMD's Athlon FX product manager, told TG Daily, "is the difference in how we measure power versus how Intel measures power." As Tom's Hardware Guide's detailed Core 2 Extreme test showed, AMD reports a Thermal Design Power (TDP) figure of 125 watts for its high-performance Athlon 64 FX-62 processor, while Intel claims 75 watts. "What Intel does states is a typical wattage figure. So there's a difference in those two numbers right off the bat."

In the definition of the term set forth by Tom's Hardware Guide years ago, TDP is "the maximum amount of power the thermal solution is required to dissipate." In other words, TDP is a benchmark figure relative to how a cooling device, such as a heat sink or a fan, dissipates heat. Its original intent was to show the maximal power consumption that a physical system could withstand and still dissipate heat at necessary levels. But as Uwe W. D. Weyden wrote for us almost six years ago, "CPU vendors provide the values TDPmax and TDPtyp to the designers" - meaning, the maximal level of heat dissipation and the typical level under normal loads.

So there are two TDPs... or, at least, there used to be, six years ago. Anyway, AMD's claim is that its 125 watt number represents the FX-62's TDPmax, and Intel's 75 watt number represents Core 2 Extreme's TDPtyp. If Intel represented the maximal TDP instead, the numbers might look much different.

There's also this bit about "thermal... " If you look at it closely enough, TDP isn't really a measure of power consumption anyway, AMD argues. It was never designed to be, states David Schwarzbach, AMD's Athlon 64 X2 product manager: "TDP's origin was as a thermal design point. This is for engineers to design their heat sinks, their ventilations, their fans, all that stuff. It's not a good way to communicate to end users what's actually happening in energy consumption within the box."

"At the end of the day," said Field, "what it comes down to is, what is the power utilization of the platform as a whole?" He reminded us about such perennial factors as the memory controller, which in Intel architecture is on a separate chip drawing power, while in AMD architecture remains on the CPU. "We do have some advantages in our architecture, with our integrated memory controller, that allows other components to use less power on the system as well. So all of that needs to be taken into consideration when you're doing a head-to-head power comparison."

It just so happens, however, that Tom's Hardware Guide did a head-to-head comparison, with some results that AMD won't much like: Under maximum workloads, an overclocked Core 2 Extreme X6800 processor drew 218 watts. That's down remarkably from the Pentium D 840's power draw of a colossal 344 watts, and it's now lower than the Athlon 64 FX-62's draw of 283 watts. There's nothing thermal about this figure.

Schwarzbach was only willing to concede just a little ground to Intel, but not much. "With the Core 2 Duo announcement, he said, "Intel is catching up with the 65 watt TDP specifications. So they're closing the gap a little bit, but I still believe AMD has the leadership position there."

The problem going forward, AMD believes, is not to re-widen the gap, but instead to change the way it's measured. "We would like to see the development of uniform ways of communicating actual power," said Schwarzbach. "TDP has been the default way of doing it to this point. I think it does the industry a disservice overall, because we're still in the dark about what the actual energy consumption is, and I think that's where the industry sensitivity is right now. In commercial [markets] with fleets of PC systems, and mobile with battery life, you really want to know at the plug - which implies this is a total platform energy measurement - what am I actually consuming in terms of wattage? What's the cost to me?"

If the Tom's test is indicative of the results one could get from a more realistic power measurement, then AMD's argument may be an easy target. AMD, however, could continue to tout the differences between Core Microarchitecture and AMD64, in hopes of substantiating one or more "asterisks," if you will, that some could put in front of Intel's power figures.

"While Intel's architecture is an improvement," said Schwarzbach, dusting off one such asterisk, "our architecture still maintains advantages because an integrated [memory] controller means that the overall platform level of energy consumption should continue to be an advantage... We don't have a northbridge chip consuming power. So average wall power, or power at the plug, is going to be a relevant communication point for us, as well as average CPU power consumption."

Would AMD support an independent effort, perhaps from a standards body, to implement a fairer approach to publishing power consumption numbers? "As consortiums and standards-making bodies emerge in this area," Schwarzback replied, "we'll certainly be one of the founding, if not influential, members in those communities."