AMD Salvages Low-power Argument with New Opteron CPUs

In the fourth calendar quarter of last year, Intel made the most headway against arch-rival AMD in the server CPU market in several years, surging ahead in shipments and reportedly taking back a few points of market share. Now, AMD is working to take those points back and then some, and to do so, it's working hurriedly to seize the low-power, high-performance advantage that the company is perceived to have lost in the wake of Intel's Woodcrest introduction.

Two of AMD's introductions today include "high-efficiency" versions of its high-performance line. Whereas the Opteron 8220 and 2220 models for dual-processor systems, introduced today, are both rated for a thermal design point (TDP) of 95 watts, their 8218 HE and 2218 HE counterparts are rated for 68 W TDP, and five other single-processor HE models introduced today are rated for 65 W TDP.

Intel might be the first to point out that these numbers don't seem good enough, with its recently introduced 2P model Xeon 5160 rated for 80 W TDP and its 5150 rated for 65 W TDP. In previous years, AMD deflected Intel's claims of power advantages by saying they were using a skewed metric, describing average power dissipation instead of maximum, implying that Intel server chips could easily run hotter.

And as engineers there have also pointed out in the past, TDP is a measure of how hot a chip might run - specifically, a relative measure of the maximum wattage a CPU could run and still be cooled to nominal levels by its on-board fan.

So now that Intel appears to have adopted AMD's metric for TDP, AMD today is working a little harder to attempt to prove that TDP is not proportionate to power draw. In internal test results the company released today, AMD says its new Opteron 2218 HE draws 141 watts of power on average while idle, and has a maximum power draw of 245 W ("at load"). By comparison, Intel's Xeon 5150 draws, under AMD tests, 231 W of power while idle and 311 W at load.

AMD then calculated the potential cost savings for servers, in locations where utility companies charge 10¢ per kilowatt-hour. (A December 2006 IBM study showed New York City utility companies charging data centers as much as 15¢/KWh.) The savings costs for using an Opteron 2218 HE instead of a Xeon 5150 were calculated to be $93 per server per year, or as much as $46,253 per year for a server farm with 500 processors.

The message from AMD: Although it may appear on the surface (and surfaces do tend to get hot) that Intel has the efficiency advantage, carefully studied measurements should indicate otherwise.

The company also resumed its ongoing message that Intel chooses skewed metrics when convenient. Today, AMD cited an Intel benchmark showing a Xeon 5160 scoring a 123 in the SPECint_rate2000 benchmark - a rating of relative performance in integer operations - versus a rating for the existing Opteron 2220 SE of 78.3.

As AMD has been telling BetaNews repeatedly in recent days, where Intel appears to hold the lead in a particular test, a modernization of the test procedures will prove otherwise. In tests using the updated SPECint_rate2006 benchmark, compiled by the authors of the benchmark itself, an Opteron 2220 SE server scored a 51.7, whereas a Xeon 5160-based server scored a 50.3.

BetaNews confirmed those figures today with the Standard Performance Evaluation Corporation database, and learned these were peak performance figures that pitted a Tyan Thunder server tested by AMD against a Fujitsu Siemens Celsius server with a 5160 on board, tested by the manufacturer.

In terms of base performance (the type of figure AMD would prefer to cite if we were talking about heat dissipation), the AMD-based Tyan server scored a 46.1, while the Intel-based Fujitsu Siemens server scored a 48.8. AMD submitted to SPEC its own test figures for an Intel 5160-based Supermicro server with a lower base score of 45.2, though it did not provide a peak score for that server.

So AMD continues its incremental improvements to its Opteron product line, and in so doing, tactfully steers customers' attention toward the bright spots while avoiding the not-so-bright. Last year, that was an easier job for AMD. But this year, until the company can implement its next wave of efficiency-improving advances, it will be a far tougher sale to make.