Nelson said his initial test was limited to two older single-core chips, a 2.4-GHz Opteron and a 2.4-GHz Xeon with Netburst architecture. The Opteron delivered 115 percent more throughput at peak, he said.
"I was shocked," said Nelson, adding that he has never seen such a transaction benchmarking gap between two chips in the two decades he has been running tests.
The machines used for the testing were identical, configured with equal memory and disk drive characteristics and application code compiled by the same compiler. Apache 2 Web server software was used to test the throughput capability, said Nelson, who also intends to compare the newer dual-core processors from the two chip makers.
Nelson, who is based in East Dundee, Ill., said his testing is independent and self-funded. "I think I'm like Ralph Nader -- I am running the test that the manufacturers don't necessarily want to see run," he said.
Nelson said he doesn't know exactly why Opteron is delivering better throughput, but he said the chip's architectural differences, including its Direct Connect Architecture -- which moves chip processing closer to memory -- probably account for the difference.
Gordon Haff, an analyst at Illuminata, in Nashua, N.H., said that "no single benchmark result is truly significant." Pointing to Nelson's plans to test Intel's more recent chips, its Woodcrest and Conrow family, Haff said those chips are Intel's current generation, and they have closed the dual-core performance gap with AMD -- at least for now, he said.
Haff said that Nelson's testing of the 2.4-GHz chip is "essentially a backward-looking benchmark test. It's not news that Netburst Xeons were significantly slower than AMD," he said. "It's far more interesting and relevant how Woodcrest Xeons stack up against AMD and how the various quad-core processors line up as they roll out into the market."
Intel officials declined to comment on the test. AMD officials could not immediately be reached for comment.