Intel: Never mind Dempsey, here's Woodcrest

Intel overshadows launch of new server chip by disclosing performance numbers of unreleased Woodcrest chip.

Tom Krazit Former Staff writer, CNET News
Tom Krazit writes about the ever-expanding world of Google, as the most prominent company on the Internet defends its search juggernaut while expanding into nearly anything it thinks possible. He has previously written about Apple, the traditional PC industry, and chip companies. E-mail Tom.
Tom Krazit
3 min read
The early numbers look good for Intel's server group, but preproduction results do not a comeback make.

Intel released the results of 20 benchmarking tests Tuesday showing how much the performance of its server processors has improved over the last few years. The numbers overshadowed the expected release of Intel's Dempsey processor, which won't have much time to make an impact before systems based on the Woodcrest chip touted in the benchmarks arrive in the third quarter.

Woodcrest, the first server processor that will be released using Intel's new Core architecture, posted better results than Advanced Micro Devices' Opteron 285 processor on several general-purpose server benchmarks developed by industry consortiums such as the SPEC (Standards Performance Evaluation Corporation) and the TPC (Transaction Processing Performance Council), as well as tests measuring performance on applications like Microsoft Exchange or high-performance computing.

The tests were all conducted by Intel on preproduction servers from several different partners, including Dell, IBM, Hewlett-Packard, Fujitsu and others, that used Intel's Woodcrest processor. The results were compared to the most recent results posted by AMD or its partners, an Intel representative said, meaning some of the tests used Opteron processors a little slower than the 285 model. Details about the specific results, configurations and Intel's testing methods can be found on Intel's Web site.

Intel's server group hasn't really wanted to talk about performance over the last year, with AMD's Opteron chip holding an advantage on most benchmarks, as confirmed by many of the benchmark graphs posted Tuesday. As a result, AMD's server market share has grown from virtually zero three years ago to more than 20 percent, according to Mercury Research. But Intel has high hopes for Woodcrest and the rest of the Core architecture processors, which will replace the Netburst architecture used by Intel's current server processors, including Dempsey.

The release of the Woodcrest results puts a bit of a damper on the Dempsey unveiling. Intel plans to market Dempsey as a low-cost alternative to Woodcrest, Intel server chief Pat Gelsinger told CNET News.com in March.

Hewlett-Packard and IBM announced plans Tuesday to ship servers based on Dempsey, now known as the Dual-Core Intel Xeon 5000 series processors. But Dell, up until last week an exclusive customer of Intel's, announced plans only for workstations based on the Xeon 5000 series processors Tuesday. Last week, Dell announced plans to use Opteron in four-processor servers by the end of the year. A Dell representative said the company would ship servers based on Dempsey in the coming weeks.

The benchmark results are similar to ones released under a nondisclosure agreement to analysts at the last Intel Developer Forum, in March, with Intel actually having improved its performance in certain areas, said Nathan Brookwood, an analyst with Insight 64. Earlier this year, an Intel executive claimed the company's Core architecture would deliver a 20 percent improvement overall as compared with AMD's chips. The Woodcrest numbers deliver that much of an advantage in some tests, but less in others.

Woodcrest, which will be known as the Dual-Core Xeon 5100 series processors, will be formally released next month, with servers expected to become available in the third quarter. At that point, independent reviewers and potential customers will be able to conduct their own benchmarks. SPEC and TPC are well-known benchmarking tests, but chip and server companies spend millions making sure their hardware runs well on those tests, which don't always reflect the performance of homegrown applications on the hardware.