The microprocessor battle between Intel Corp. and AMD can be likened to many things, but in the eyes of industry analyst Nathan Brookwood, sports analogies seem the most apt.
“This is better than March Madness” — the annual U.S. college basketball championship playoffs — Brookwood said of last week’s Intel announcements of its upcoming Penryn and Nehalem chip families.
“There you only see the teams go up against each other once. Here we’re seeing this thing go on and on.”
Comment: info@itbusiness.ca
What’s going on and on are new generations of chip architecture the two companies are building, giving rise to multi-core chips of increasing speed and decreasing power consumption.
Last week Intel released details of the Penryn family, to go into production at the end of the year with products likely appearing in early 2008, Brookwood said, which will be the first to use a 45-nanometer manufacturing process. Intel now makes its CPUs with a 65nm chips, a size AMD is only just getting into.
Thanks to the tinier size, Intel says it can squeeze 820 million transistors into the Core 2 quad-core versions of the Penryn CPUs.
It’s promising a range of dual- and quad-core desktop processors, dual-core mobile processors and quad-core Xeon server processors will be released, as well as a processor for high-end server multiprocessing.
Laptop lovers will appreciate the new mobile Penryn processor’s power management, said to significantly reduce the power of the processor during idle periods.
What was new in the public briefing, Brookwood said, was that dual-core Penryn chips will have caches of up to 6MB (an increase from the maximum 4MB in its current CPUs), and quad-core chips up to 12MB, and that and front side bus speeds will increase to 1.6Ghz (up from 1.33Ghz).
But what struck him was Intel’s willingness to talk about the Nehalem chips, which aren’t scheduled to go into production until 2008.
“I’m puzzled as to why they’re trying to undercut the Penryn attention,” said Brookwood, whose California-based firm is called Insight 64.
“When you have set of new products that are pretty exciting the last thing you want to do is say, ‘Look at the stuff that’s coming out two years from now.’”
Nahelem CPUs will have up to eight cores and including technology similar to Intel’s hyperthreading technology, allowing the operating system to process several threads simultaneously. Some older Pentium 4 CPUs use hyperthreading, but the current Merom, Conroe, and Woodcrest processors don’t.
Although Intel didn’t outline how it will build its eight-core chips, Brookwood believes it will follow its current process of using pairs of processors and put two four-core CPUs in a package rather than eight cores on one piece of silcon. Penryn quad core CPUs are made of two pairs of dual-core processors.
On the other hand AMD’s upcoming Barcelona family of quad-core CPUs, to be released in the summer, have four cores on one piece of silcon. That means they share caches and external interfaces and, in theory, will be faster than Intel quads. AMD won’t be able to do eight cores until it’s Singapore family of processors comes out.
“Some purists will argue that’s not really quad core,” Brookwood says of Intel’s approach. “I don’t have a problem, but clearly the performance benefits are somewhat tempered by the implementation techniques. AMD, because everything is shared, has the potential for a much higher level of performance.”
More interestingly for Brookwood is that Nehalem chips will include integrated memory controllers, which AMD began doing three years ago and Intel dismissed as insignificant.
“The good news for Intel is they needed to make this transition,” he said. The bad news for AMD is that it only has a year and a half to exploit that advantage.
At least until its Barcelona family debuts. And how will it do against Nehalem chips? Brookwood can’t say yet because details of the architecture haven’t been released.
Reflecting on his basketball analogy, Brookwood had a second thought.
It’s like two tennis players volleying back and forth, he said, with onlookers watching for a mistake.
“Or,” he added, “who’s going to come up with that powerful shot the other can’t return.”