A bumpy ride ahead

By Mark Hachman  |  Posted 2004-10-20 Print this article Print

?"> For system vendors and analysts, the question is how smoothly that transition will actually proceed. Microprocessor companies including AMD, IBM and Intel have all suffered rocky transitions when shifting to new microprocessor designs and processes. Several companies have revised product roadmaps for faster processors and new integrated capabilities, analysts pointed out. Intels Houlihan indicated that Intel will provide some sort of a safety net for vendors for the single-to-dual transition.
"Theyll coexist for some period of time, he said, noting that the company has had decade-long contracts with some suppliers for its Pentium desktop microprocessors. Houlihan said, however, his comments should not be interpreted as the length of time that Intel would support its single-core chips.
The shift from a single-core to a dual-core architecture is in part a recognition by processor manufacturers that a single-threaded, single-core architecture isnt as efficient as one that uses multiple cores to process data. In all processors, a primary goal is to keep the processor core as active as possible, constantly acting on and outputting data. Traditionally, thats been done as quickly as possible, with extremely fast processors that can only work on a single thread. By moving to a dual-core architecture, the workload can be spread over the second core, allowing the chip to run at a reduced clock speed. The dual-core approach can also impact the need for large data caches on some systems. For example, Intel has leveraged its manufacturing efficiencies to produce successive generations of larger chips with larger caches that serve as a repository for frequently-used information. However, in microprocessor production, large die-size and the real estate for caches come with a cost, both for the chip maker and for the user in increased power consumption. The caches can be replaced by a more efficient architecture, such as one that employs multithreading, or the ability to work on more than one instruction thread at a time. "I improve the performance of my machine, and thats really at this stage of the game thats what dual-core processors are about," said Paul DeGroot, an analyst with Directions on Microsoft, Kirkland, Wash. Although the dual-core designs may be more efficient, reducing power and heat, the exact performance gain from such a processor when compared with a faster single-core processor is still an open question. At the recent Fall Processor Forum in San Jose, AMD officials took the wraps off a dual-core version of the Opteron architecture. AMD Fellow Kevin McGrath said dual-core system running 5 clock speeds down from the companys fastest part will generate approximately 125 to 140 percent the performance of the comparable unicore system, McGrath said. For many system vendors, the single to dual-core shift will be a "non-event," said Gordon Haff, a server analyst with Illuminata in Nashua, Mass. Server OEMs with proprietary multicore microprocessor architectures, such as IBM, have gone through all of this before, he observed. "Even in desktops the dual-core transition is not going to be like a light switch, overnight," Haff said. "Theyve got multiple product lines, and it will take a bit of time to transition over." Analyst DeGroot added that the lack of a hardware infrastructure was mirrored in the software community. "My general sense is it gives Microsoft an advantage that reflects the fact that dual-core processors are kind of a workaround for the heat of other processors," DeGroot said. Theyre not intended, at this stage of the game, to give the ability to run multiple instances on separate cores." "An app like SQL server isnt easily capable of distinguishing between the cores," DeGroot said. "We need to see what future development in virtualization can provide as well. But until its possible to virtualize a single core, it would seem to me the customer is kind of helpless." In addition, another reason for delay might involve the fact that for most applications involving repeated disk and memory accesses, a dual-core chip might not offer that much of a performance boost, noted Dean McCarron, principal analyst at Mercury Research, of Cave Creek, Ariz. Dual-core chips will first find a home in 3D visualization workstations and scientific applications, where a lot of repetitive calculations are made, he said. While dealing with incomplete information, McCarron understood that Intels dual-core may not be as synchronized as competing products. "I think in the more distant future there will be a lot of evolution on the socket bandwidth," McCarron said. "Whats happening here is that Intel is skipping ahead on the processor side without necessarily advancing the supporting infrastructure. My expectation is that the significance of dual cores will be the ability to turn the clock rates down and resolve some of the thermal issues," he said. According to McCarron, that transition will be a "conservative" one. From Intels statements that observation holds. On Tuesday, Barrett said the decision to shift to a dual-core architecture was driven by power concerns. Intels decision to move to move to a "dual-core, multicore, multithreaded capability is precisely what we have to do to avoid that power challenge," he said. Meanwhile, Intels dual-core Xeon will use an undisclosed chipset, replacing the Lindenhurst chipset. Intel officials said previously that the first dual-core version of the Xeon MP line will arrive in early 2006 and be dubbed "Tulsa, followed by "Whitefield", a multicore Xeon MP, in early 2007. Lisa Vaas contributed to this story. Check out eWEEK.coms Desktop & Notebook Center at http://desktop.eweek.com for the latest news in desktop and notebook computing.


Submit a Comment

Loading Comments...
Manage your Newsletters: Login   Register My Newsletters

Rocket Fuel