IDF: McKinley to be twice as fast as Merced, says Intel

At its Developer Forum in San Jose, Intel releases the first performance figures for McKinley and it looks promising, but 32-bit applications will see little benefit
Written by Matthew Broersma, Contributor on

Intel on Tuesday unveiled for the first time some of the hardware and performance specifics around its next-generation server processors, McKinley and Madison.

Executives are talking up the Itanium line of 64-bit processors here this week during the Intel Developer Forum, attended by engineers, executives and technical marketers from across the PC industry. The Itanium line is Intel's first foray into high-end server computing, with the first processor in the line -- code-named "Merced" -- having been launched about seven months ago. Intel claims to control about 34 percent of the lucrative high-end market, far below its dominant 84 percent share in front-end servers.

While Merced has been seeping onto the market for about a year, the servers are only considered useful as development systems. That is expected to change with Merced's successor, McKinley, due in the first half of next year.

Intel says that despite improvements, such as the machine checking architecture discussed earlier today, McKinley will attempt to build as much as possible on the Merced legacy. It will, for example, run Merced binaries with no recompiling necessary, speeding up performance by about 70 percent, based on specint2000 benchmark testing.

The added speed is due to the McKinley core running at 1GHz, up from Merced's 800MHz, as well as a higher front-side bus frequency, more execution units and faster memory cache.

While Merced's level-3 cache -- a cache of memory frequently used by the processor -- is on the processor cartridge, with McKinley this cache is moved onto the die. It will be 3MB, smaller than Merced's 4MB but with reduced latency. A mid-tier version of McKinley will also be available with a cache as small as 1.5MB.

McKinley will not be physically compatible with Merced, using a different front-side bus. The system bus will be 128 bits wide, run at 400MHz and transfer 6.4 gigabits per second.

The microarchitecture will be enhanced, with six integer execution units to Merced's four, two loads and two stores per clock (double that of Merced), and 11 issue ports. Such features add to McKinley's ability to address large amounts of data.

Intel representatives said that recompiled Merced code could run up to twice as fast on McKinley. Those figures are based on server processes, which are unpredictable and therefore difficult to speed up using the speculative execution process built into the Itanium architecture.

Those who want to run 32-bit applications on McKinley will be disappointed to find that such code will not see any significant performance gain. The 64-bit Itanium line can run 32-bit software, but only at a fraction of normal processor speed. "We don't see users putting lots of 32-bit applications on [Itanium]," said Lisa Hambrick, co-director of Intel's enterprise processor marketing.

Madison, the third iteration of the Itanium line, will appear in 2003 with a 6MB L3 cache. It will share the form factor, chipset and front-side bus architecture with McKinley, but will be manufactured to the 0.13 micron manufacturing geometry, reducing its power consumption and heat output.

Intel also discussed some of the details around Deerfield, a version of Itanium optimised for high-density computing which will arrive in 2003. Merced is not practical for high-density computing uses, like data centres, because of the immense amount of heat it generates -- current Itanium boxes are mostly filled with cooling units. By the time of Deerfield Intel will have brought the 0.13 micron manufacturing process to the Itanium line, which will begin to address heat issues. Intel said that Deerfield will be used in two-processor systems with the 2U form factor, which is used in rack-mount systems. The company has brought McKinley systems down to the 1U "pizza box" form factor in the laboratory without using special cooling techniques, a spokesman said.

For denser form factors using more processors, or for blade servers, Intel will continue to rely on the Xeon processor, the company said.

Despite the drastic downturn in the fortunes of the telecommunications industry, Intel is stepping up its push into telecoms servers, a move which the company detailed at IDF Europe earlier this year. The space is currently dominated by Sun Microsystems.

"We're investing heavily within the telco space," said Tom Garrison, Intel's co-director of enterprise processor marketing. "We're going to go in there with the Intel economies of scale and the overall benefits that the Intel architecture has and be able to offer platforms at a much lower cost. [We will] emerge in that space with very high growth."

ZDNet UK's Matt Broersma reported from San Jose.

See Chips Central for the latest headlines on processors and semiconductors.

See the Developer News Section for jobs and the latest headlines.

Have your say instantly, and see what others have said. Click on the TalkBack button and go to the Chips Central forum

Let the editors know what you think in the Mailroom. And read other letters.

Editorial standards