eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
Intel is upping its commitment to lowering server power.
The chip maker on June 6 disclosed to analysts plans to launch its dual-core “Woodcrest” Xeon DP 5000 series processor on June 26 and to pull the introduction of its “Tulsa” Xeon MP ahead by a quarter, while also delivering lower-power versions of that chip.
The acceleration of the new Xeons—Woodcrest was originally slated to come in the third quarter—shows Intel is taking energy efficiency seriously enough to snatch performance per watt leadership from rival Advanced Micro Devices, whose momentum it aims to stem with the new chips.
Intels quick move to new Xeons is one attempt by the chip maker, which has been criticized in the recent past for offering power hungry chips, to address senior IT managers growing concerns about datacenter electrical and cooling bills.
Woodcrest is “showing tremendous performance, even exceeding our own engineering goals,” Kirk Skaugen, vice president of the Digital Enterprise Group and Co-GM Server Platforms Group, said during an analyst briefing on June 6.
While it promises high performance, Woodcrests “power looks outstanding…beating our 80-watt goal with the majority of our volume coming at 65 watts.”
Indeed, the majority of the two Xeon processor lines will each use less power than previously estimated, with Woodcrest arriving at 65 watts and Tulsa coming at 95 watts. The top performing versions of the chips will still hit 80 watts and 150 watts, however.
But the power consumption numbers alone are not the only measurements IT managers should base their buying decisions on, Skaugen said.
Skaugen encouraged IT managers to instead look at the total amount of power a given machine draws from an electrical socket.
This watts-at-the-wall measurement, he said, is more accurate than adding up chip specs, which tend to reflect maximum power consumption because they show what a server behaves like under various loads.
The Intel executives took issue with past statements by Advanced Micro Devices executives that have contrasted AMDs power consumption specifications for Opteron platforms.
AMD has totaled up its own Opteron processor and memory power consumption specifications and compared them with those of Woodcrest Xeon and come out ahead.
Opterons totals differ as they include memory controllers and work with DDR2 (double data rate 2 dynamic RAM), while Woodcrest chips, like other Intel processors, use discrete controllers and FB-DIMMs (Fully-Buffered Dual Inline Memory Modules).
FB-DIMMS embed a buffer chip to boost their performance, and also use more power—roughly nine to 10 watts of power—versus non-buffered DDR modules.
“Measuring thermal design point [specifications] is kind of like going around and determining your monthly electric bill by going around and counting every appliance and every light bulb,” Skaugen said.
Average system power, instead, should be the focus, he said.
AMD executives dont necessarily disagree with what Skaugen had to say, an AMD spokesman said.
AMD officials have also said that power at the wall should be the main focus of IT managers attentions. Several groups, which have varying levels of involvement from both Intel and AMD, are now actively working on server power benchmarks.
AMD also kicked off the Green Grid Alliance, which is working to teach companies more about data center power efficiency.
Next Page: Chips put to the test.
Chips Put to the
Test”>
But, to underscore his point, Skaugen referenced a series of Intel tests that showed a Woodcrest server drawing roughly 40 to 50 fewer watts at the wall then a similar AMD Opteron machine when both were attached to a power meter and put through the SPECint Rate Base 2000 test.
The Woodcrest machine, an Intel “Star Lake” white box system fitted with its 3GHz Xeon 5160 and eight 1GB memory modules, used 267 watts in the test. The AMD machine, an HP ProLiant server based on 2.6GHz Opteron 285 chips and eight 1GB DDR2 memory modules, used 307 watts, the Intel test showed.
The test basically showed that the Intel Woodcrest server averaged less power over a given period of time on a series of five benchmarks, including marks such as SPECs SPACjjb2005 and SPECint_rate_base_2005, versus the AMD-based machine, Skaugen said.
“What we really want to know is whats the watts at the wall, if you will,” he said. “Watts at the wall is lower for Intel than for AMD.”
The AMD spokesperson criticized Intels choice of benchmarks and said that the test was one of many potential matchups. Others, he said, have shown AMD systems using less power.
“Theres certainly going to be some configurations, Id guess, where Opteron will be at parity,” Skaugen said.
The Intel watts-at-the-wall-test underscores Intels assertion that Woodcrest servers average power consumption will be less than AMD Opteron machines, while offering better performance, giving Intel back the edge in performance per watt, Skaugen said.
AMD doesnt necessarily agree there, either, the spokesperson said. The chip maker plans to deliver increased performance per watt during the quarter, when it will roll out a revised version of its Opteron chip that sports redesigned circuitry and a new DDR2 memory controller, the AMD spokesperson said.
Turning away form Xeon MP, Skaugen said that Intel feels confident about Tulsas performance and that the company would being producing Montecito, its dual-core Itanium 2 chip, this month.
The Itanium 2 chips will also be identified with model numbers in the 9000 range, similar to Intels Xeon MP 7000s and Xeon DP 5000s, the company indicated.
Tulsa, Skaugen said, will allow Intel to get back into the game in multiprocessor servers, an area where AMDs Opteron has been strong of late. Even Dell, once an Intel-only shop, has chosen to offer multiprocessor servers based on the Opteron.
“We have lost share on Xeon MP particularly in the performance space. We got behind on performance, shame on us,” he said.
But, “With the performance were seeing out of Tulsa…were confident that we have a significant lead on transaction performance…the primary criteria on which I think we lost share is addressed by the Tulsa product.”
Although he didnt offer additional details on the 95-watt Tulsa chips, Skaugen said more than half of the Tulsas it offers will be 95 watts. The company will offer a Tulsa chip with 16MB of onboard level 3 cache memory. That chip, it has said, will run at 150 watts.
But despite its performance of late in the multiprocessor space, Intel didnt lose Dells business completely, Skaugen said.
“We expect them to be one of our best Tulsa customers,” he said. Instead, “What Dell announced was augmentation of their product line.”
Given that PC makers are likely to continue offering the choice, he said, its up to Intel to convince their customers to pick machines based on its processors.
“Ultimately, the customer will decide,” he said.