Intel ready for 2008 with Penryn
The planned launch of Intel's Penryn processors on Monday is the first blow in a one-two punch that might stagger AMD heading into 2008.
Just a few months after the launch of AMD's quad-core Barcelona chips, Intel is hitting back with Penryn, now known as the Xeon 5400 family of processors. A total of 15 server chips are set to launch Monday as well as a new Core 2 Extreme desktop processor, with Penryn chips for mainstream desktops and notebooks scheduled to launch in the first quarter of next year.
Penryn is essentially a shrink of the Core architecture that brought Intel out of the woods in 2006. But these are also the company's first chips to use Intel's 45-nanometer manufacturing technology, and they will usher in the first change to the basic properties of the transistor since the 1960s.
For the first time, Intel plans to use a metal gate and a new material for the oxide layer around the gate in its transistor designs. This fundamental part of the transistor provides the foundation for computing as the part that determines whether a transistor is off or on, a "0" or a "1."
"We needed the scaling and power/performance, and it would be very hard to do it on the previous technology," said Dadi Perlmutter, senior vice president and general manager of Intel's Mobility Group.
Intel and other chipmakers were running into problems making the gates smaller and smaller. The gates were getting down to the point where the gate dielectric--an insulating layer that sits between the gate and the rest of the transistor--was only a few atoms thick. The chipmaking industry has figured out lots of amazing things, but it hasn't figured out how to split an atom without causing a bit of an energy problem.
The new materials allowed Intel to build thicker gates with the same switching properties as the older ones, which helps control current leakage and also buys the industry a few more generations of scaling. IBM and AMD plan to release chips based on similar technology in 2008.
And the combination of the new transistors and some design tweaks appears to have been enough to give Intel a performance lead with the Penryn generation of chips. The company said some of its partners set world records for scores on well-known benchmarks such as TPC-C and SPECint_rate2006 with the basic Xeon chips. When paired with the 1600MHz front-side bus available on some chips, Intel said it also set records on SPECfp_2006rate, long a stronghold of AMD's. SPECint_rate is a general measure of transactional performance that's important to business customers, while SPECfp_rate measures floating-point performance and is important to high-performance computing customers.
Intel avoided making direct comparisons to AMD's chips in briefing materials distributed ahead of the announcement. It plans to have a Web site up and running on Monday with more detailed performance information.
Intel did say that the new Xeons will be about 28 percent faster than their older brothers on SPECint_rate2006, and 30 percent faster on SPECfp_rate2006. Barcelona barely edged out the older generation of Xeon chips on SPECint_rate2006, so it looks like Intel will have an edge in that area.
If you make a server-buying decision based solely on these numbers, however, you're making a mistake. There are some truths to benchmarks, but companies like Intel and AMD spend millions of dollars trying to get an edge on benchmark results, which don't always mirror real-world performance. Still, they're the best comparison vehicles we've got, though those who are contemplating a serious server purchase test their applications on a system before making the leap.
Penryn marks the second generation of Intel's quad-core designs. Around this time last year, Intel packaged its dual-core Core generation processors into quad-core chips that enjoyed several months free from competition from AMD.
That free ride ended with the Barcelona launch. AMD gained back some market share in the third quarter, as Barcelona systems trickled into the market. Still, going into the launch the company didn't expect Barcelona to contribute meaningful revenue until the fourth quarter.
And it seems that AMD is having a little trouble getting Barcelona into the market. Reports surfaced last week in the run-up to the Penryn launch that some server vendors are quoting 2008 as the time frame for Barcelona's availability, even though AMD executives said they plan to ship "hundreds of thousands" of Barcelona chips this quarter.
That, of course, is exactly when Intel will fire back with the Penryn chips. The new Xeons will arrive in the same pricing bands that Intel's current lineup of Xeon chips occupy, and Intel plans to have systems available right away from the usual suspects in the server market.
And next year, it will get even tougher for AMD. The company has two new designs for desktops and notebooks (known as Spider and Puma) that are set to arrive over the next few months. But Intel isn't sitting still, either: the first quarter of next year will see Penryn chips arriving for desktops and notebooks, as well as an extremely low-power chip called Silverthorne that could open new markets for Intel that AMD can't touch until 2009.
To top it all off, Intel's main plan for 2008 is to release chips called Nehalem that borrow many of the same design techniques, such as an integrated memory controller and point-to-point connections, that made AMD's Opteron chips a winner for several years. AMD would say it's a sign that it was right all along, but it doesn't really matter: Intel has managed to stay very competitive without those techniques, and when it adopts them, AMD could be in more trouble next year.
The smaller chip company is clearly pinning its hopes on 2009, when it aims to release "Fusion" chips that integrate a high-powered GPU with a PC and server processor. The "Bulldozer" chips will also be assembled from smaller building blocks, which could let the chipmaker target specific customers with designs tailored for their needs.
For now, though, Intel is in excellent shape--assuming it doesn't run into any problems during the first few months of the Penryn launch. In just two years, Intel has managed to get beyond the embarrassment of its abrupt change in course at the hands of AMD to get its server group back on track.
Few of us will ever buy a server based on these chips, but this market is extremely important to both Intel and AMD because it's so much more profitable than cranking out chips for your desktop or notebook. That helps fund the development of other technologies that do have an impact on the rest of us, meaning that the competitive balance between the two companies in this segment has far-reaching implications.
For now, advantage Intel.
Tom Krazit writes about the ever-expanding world of Internet search, including Google, Yahoo, online advertising, and portals, as well as the evolution of mobile computing. He has written about traditional PC companies, chip manufacturers, and mobile computers, spending the last three years covering Apple. E-mail Tom. 
Break the Wedge!
www.BreakTheWedge.com
I'm tired of not getting what I pay for.
Penryn is basically the 45 nanometer die shrink for the Core 2 architecture. It will come with 50% more cache, double the transistor count, and is about 25% smaller than Conroe (65nm Core 2).
Wolfdale and Yorkfield are dual and quad core Penryn's.
Anandtech's benchmarks (at same clock speed) had Penryn 4-5% faster than Conroe/Kentsfiel in non SSE4 optimized benchmark and 10% better in power consumption. Nothing ground breaking, but there were a few very specific cases where Penryn bested Conroe by ~45% in SSE4 optimized media encoding.
AMD has states Phenom will be competitive with Conroe, but Penryn will be out before Phenom. There are no samples/benchmarks of Phenom released yet that I know of.
The $10,000 question is obviously if Phenom is competitive enough to be faster than Penryn when it is released.
My best guess is no, but I did say guess.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092
Take it for what it is - a preview, not the final product.
Great grasp of the English language mate...
Meanwhile, looks like exciting times ahead in the development of faster and more efficient processors. No doubt these new server technologies will trickle down soon enough to the desktop and laptop market. On the other hand, for most users doing internet surfing and word processing, it seems to be more along the lines of change for the sake of change?
Mike
- AMD Cannot Compete Unless...
- by eightwings November 12, 2007 9:43 AM PST
- AMD is fighting a losing battle. Intel defined the current market and AMD cannot beat them at their own game. They are condemned to always play second fiddle unless they can find a way to redefine the market. They can only do so by reassessing the current state of the art in multicore CPU architecture and computer programming and correct what is wrong with it. And there is a lot that is wrong with it. I call it "The Age of Crappy Concurrency". Check it out: http://rebelscience.blogspot.com/2007/09/age-of-crappy-concurrency-erlang-tilera.html
- Like this Reply to this comment
-
(7 Comments)Now that the industry is transitioning to massive parallelism, AMD has the chance of a lifetime to change the computing landscape in its favor and leave Intel and everybody else in the dust.