The new chip line offers a performance bump over its predecessor, the Xeon 7200 series, Intel said. Much of that increase comes from adding a 16M-byte level 3 cache. The 7400 series processors are the first Xeon chips to use a level 3 cache, which stores data closer to the processor cores, helping to boost overall performance.
"With the level 3 cache, that does contain additional performance for some of the high-compute-intensive and data-intensive enterprise applications," said Adesh Gupta, regional server platform manager at Intel Asia-Pacific.
The extra cores also help. Unlike desktops and laptops that rarely run applications capable of tapping the full processing power of quad-core chips, many server applications, like virtualization, run better on multi-core processors.
The first processors to come out of Intel's India Design Center in Bangalore, the Xeon 7400 chips run at clock speeds up to 2.66GHz and have either four or six cores. They are priced ranging from US$856 to $2,729, in 1,000-unit quantities. Servers based on the chips will be available starting Tuesday from vendors like Hewlett-Packard, IBM and Dell, among others.
The Xeon 7400 contains all six cores on one piece of silicon, while Intel's existing line of quad-core Xeon chips pack two pieces of silicon inside a single package. This was possible because the 45-nanometer process used to make the new chip reduces the size of the features on a chip, increases performance and reduces power consumption.
"We knew that this process would help us pack in more transistors," Gupta said.
The Xeon 7400 series is the last member of Intel's Penryn chip family to be released. Later this year, the company will shift to a new processor architecture called Nehalem.
Like earlier chips, the Xeon 7400 relies on a memory controller located in an external chip, which can cause memory bottlenecks in certain applications. The level 3 cache helps to alleviate this problem, but cannot eliminate it entirely. Nehalem will move the memory controller onto the processor itself, which is likely to speed up memory access considerably.