ServerWatch content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
First-Gen Problems
Late to the 64-bit party, Intel clearly made up the lost ground rapidly. But will the same thing happen in dual core? Maybe not. One of problems with Intel’s first-generation dual-core processors was that when installed they were the equivalent of 2-way servers. Brookwood explains that they were little more than a combination of two single-core processors on a piece of silicon. The problem, however, was that they acted as though they weren’t sitting next to each other; they used an external link to communicate over the front side bus. This process is less efficient, slower, and requires more power than true dual-core.
That’s all changing, however, with the newest batch of Xeons. The impending release of the dual-core processor “5000 sequence,” codenamed Dempsey, is a good example. These second-generation dual-core chips support EM64T and Hyper-Threading; they run at 2.80 GHz, have 2 MB of Level 2 cache per core, a new Intel E7520 chipset, and built-in virtualization technology. They are aimed primarily at the Web, infrastructure, and e-mail server audience. Intel also has a 32-bit Xeon dual-core processor on the way that will deliver improved performance per watt. Codenamed “Sossaman,” it is based on its mobile core and targeted at low-power and thermal environments, such as blades.
By mid-2006, watch for a new Itanium-2 dual-core processor called the “9000 sequence” (codenamed “Montecito”). This is just one of more than a dozen dual-core and multi-core projects Intel currently has in development. As a result, the vast majority of Intel chips will be fully dual-core by year end.
The improvement between the first generation and the second generation is evident in OEM uptake. HP remained cool to the first offerings, but it is introducing servers based on the latest Intel dual-core processors, such as the ProLiant DL320 G4, DL360 G4p, DL380 G4 SAS, ML370 G4, and BL20p G3 Blade.
Two-Horse Race
A year ago, some hoped that other chipmakers, such as Transmeta or VIA, might become serious contenders. It now appears that the server market will remain firmly a two-horse race for some time to come.
And what does the future hold?
Intel has big plans. By the end of the year, it intends to have 85 percent of its server processors dual-core capable. Instead of shrinking the gap, however, Brookwood believes AMD’s performance/price lead might widen with the introduction of Dempsey. He explains that an already power-hungry Intel chip design employs a memory architecture that actually consumes more power (36 more watts from six memory modules). Later in the year, Intel will cut processor power consumption to 90 watts. Although this is still not on par with AMD, the memory design pushes the total wattage up to about 126 watts.
“Intel is taking steps to narrow the performance gap per watt, but whether they’ll catch up to AMD is hard to say, if they continue using this memory architecture,” says Brookwood. “Intel probably won’t have onboard memory controllers until 2008 or 2009.”
He predicts neither party will deliver quad-core chips until 2007, with AMD once again likely to be the first guest to arrive.
“We believe the server market will eventually be made up solely of dual-core and multi-core processors,” says Patla. “We plan to continue with dual-core through 2006, and you can expect multi-core (such as quad-core) from us in 2007.”
