AMD Clinches Quad-Core Price/Performance Lead After First SPEC Results
The first official benchmark test results for Intel's Tigerton series and AMD's Barcelona series were released by the Standard Performance Evaluation Corporation. From a pure prowess perspective, neither the new quad-core Xeon MPs nor the new quad-core Opterons walk away with an all-around victory, with Intel's X7350 stealing the show in integer throughput, and AMD's 8350 winning the race in floating-point throughput.
Intel tested a Lenovo R630 G7 system with four of each of its new quad-cores dropped in at some point, including its new top-of-the-line 2.93 GHz Xeon MP X7350. Though it has yet to publish results for all Tigertons across the board in the SPECfp_rate 2006 floating-point throughput test, a quartet of X7350s scored about 29% better in base rate and nearly 34% better in peak rate than four of AMD's top-of-the-line Opteron 8350s in the SPECint_rate 2006 integer throughput test.
But in the SPECfp_rate test - for which we only have Intel results from the X7350 - the Opterons performed somewhat better: by about 15% in base rate and 17% in peak performance. AMD built its own system based around a Tyan Thunder n4250QE, a four-way Socket F motherboard that uses nVidia's nForce Professional 2200 chipset.
AMD had earlier advised BetaNews that we'd see its architecture start to shine when the workload and complexity of the task scaled up, and this is the first clear indication of what we've heard. With AMD's new processors selling for substantially less than Intel's in the same category - for the top-of-the-line, 56% less than Intel's price for 1000-unit quantities (thus likely an even greater percentage for single units) - the mixed results actually end up benefitting AMD.
But with a key element of Intel's results yet to be published, one of the Tigertons may yet close that gap: the Xeon MP E7330, which is in the middle of its product line. As with the first Core 2 Duo Conroe series processors in July 2006, Intel's middle of the line has been priced to deliver the best performance value for the dollar.
Back then, it delivered the E6600 dual-core model at a price that quickly dropped on the street to just over $300. Compared to Intel's sweet spot at the middle of its Xeon MP product line, its top of the line again carries a heavy premium, while the bottom of the line could be overpriced for what it delivers.
Though our gut feeling has been telling us the E7330 could be Intel's value leader, and a true contender against AMD, we wanted a relatively simple though calculated way of verifying this suspicion. So we created a quick worksheet that attempts to quantify the performance value of the quad-core CPUs in question, in terms of how much performance and efficiency does a dollar can buy. This way, wattage (expressed as Thermal Design Point), clock speed (frequency), price and perceived performance all play equal roles.
Our goal is to create an index that lets you envision the relative efficiency value of quad-core processors, so you can see how much power your company may be purchasing with the same money.
|
| A table calculating the relative performance index for new AMD and Intel quad-core processors. |
Our formula is crude, but effective. First we divide frequency by TDP to arrive at a relative "power index" - how many cycles per second can be driven by one watt of power. We then divide each CPU's power index by its price, then multiply the result by a 50/50 combination of the SPECint_rate and SPECfp_rate scores. In both categories, we counted base rate as contributing to 75% of the final score, and peak rate as contributing to 25%. Then we moved the decimal point several places so that we're not dealing with small change. All we need is a set of indexes we can compare with one another.
At $1,019, AMD's Opteron 8350 sells per unit for less than half of Intel's price for the Xeon MP X7350 when sold in quantity: $2,301. So even though the X7350 is almost a 30% better performer at integer tasks, your company's dollar buys you almost twice as much processor efficiency with an 8350 than an X7350, as denoted by our index scores: 28.18 for the Opteron versus 14.52 for the Xeon MP.
At $389, meanwhile, AMD's two-way Opteron 2350 seizes the overall efficiency value lead, with an index score of 41.58. That means two 2350s could conceivably be a better investment for your company than two 8350s - if their scores were even, the 8350 would be just as efficient for the dollar as the 2350, even though the former is the four-way model. For its SPEC submission, AMD constructed its own 2350-based system on an ASUS KFSN4-DRE motherboard, which is what AnandTech chose for its recent Barcelona review. It uses nVidia's nForce Pro 2200 chipset.
Our results are not final by any means, however. Since we don't yet have SPECfp_rate results for the other processors in the Tigerton array, though we do have integer rate scores, we wondered what floating point scores would Intel have to post to enable their midrange quad-core CPUs to match the final index score for AMD's 8350.
Working the spreadsheet backwards, the news is not good for AMD. With extremely impressive integer rate scores for the Xeon MP E7330, if it were only to post floating point rate scores that merely matched those for the Opteron 8350, it would still beat the Opteron in the final value index, for about a 30.9. Take away 1 MB of L3 cache, and the Xeon MP E7320 would still beat the 8350 if its floating-point scores were to match, for a final score of 29.3.
There's a big problem, though: The 8350 bests the X7350 in floating point rate scores, so it's extremely unlikely that the lesser units in the Tigerton line will match Opteron's performance there. Still, if the E7330 were to match the X7350's score there, it could still eke out a final efficiency index score of 28.8, enough for Intel to claim it has the most efficient quad-core in the server space. We may know the final results in a matter of days.
"According to the SPEC CPU2006 license, if numbers are actual results," said SPEC media relations director Bob Cramblitt, "then whoever provides the numbers must be able to provide a full-disclosure for the results; otherwise, the results must be labeled as estimates."