Nanotech - The Circuits Blog

Updated on March 30 at 1:50 p.m. PDT with additional information throughout.

Intel's Nehalem-architecture chips will now try to make their mark in servers, after debuting in desktops in November.

On Monday, Intel is rolling out new Nehalem-based Xeon models targeted at servers using up to two processors. Nehalem offers some important firsts for Intel, including an integrated memory controller for better performance, hyper-threading for up to 16 virtual cores (which improves multitasking), and Turbo Boost Technology, which dynamically increases the processor's frequency (speed), as needed.

Pat Gelsinger, senior vice president at Intel, speaks at Nehalem server chip launch Monday.

(Credit: Screen capture by Brooke Crothers)

"Our most important server launch in well over a decade," said Intel Senior Vice President Pat Gelsinger, speaking on Monday at the Nehalem server chip launch event in San Francisco, which was streamed live over the Web. Gelsinger said Intel has already shipped hundreds of thousands of quad-core Xeon 5500 processors--which pack in just over 730 million transistors--to mostly high-performance computing customers. Gelsinger also said future 32-nanometer "Westmere" versions will have six cores and be essentially drop-in replacements for the current 45-nanometer Xeon 5500.

The announcement is anticlimactic to some extent. Apple has already announced a new Mac Pro using the Nehalem Xeon 5500 processor and last month Intel discussed how Nehalem will be used in new "Willowbrook" servers in mega data centers.

But fresh announcements were made Monday from the largest server suppliers in the world, including IBM. "You can get anywhere from a 20 percent to a 200 percent performance improvement going from (Intel's) existing 'Harpertown' processor," said Alex Yost, vice president IBM BladeCenter. IBM is the largest server supplier in the world based on revenue, holding about a 33 percent share of the market, according to Gartner.

"We did install a bunch of early systems at key clients on Wall Street and I am very encouraged," he said. Yost added, however, that at the processor level Nehalem offers virtually no difference in power savings over previous generations of Intel chips. Improved power efficiency can be achieved other ways, though: for example, using virtualization to condense many virtual servers into a single physical server or using more power-efficient motherboards. Moving from rack-based servers to blade servers can also boost power efficiency.

Dreamworks has also been a high-profile early adopter of Nehalem. Prior to Nehalem, Dreamworks had to wait overnight to get an animation rendering project completed but this can be done almost in real time with the new processor, according to Intel and Dreamworks.

Josh Crowe, vice president of engineering at Savvis, an outsourcing business, said Nehalem offers better virtualization technology for customers whose budgets "have been cut to next to nothing" because of the tough economic times. "The consolidation ratios are going to be massively improved," he said, referring to the ability to bring many virtual servers into one physical server box.

Analysts don't expect Nehalem to cause seismic shifts in chip market share because Intel processors already claim a disproportionate chunk of the server market. "We expect only an incremental upside on a unit basis," said Ashok Kumar, an analyst at investment bank Collins Stewart. More importantly, Kumar is closely watching Cisco's entry into the server market. "If the Cisco relationship opens up new market opportunities, that's where the upside will be."

Ponder this: Is an Intel product launch still a launch, if the product debuts very publicly in an Apple computer?

I won't presume to answer that question. But the fact is that Intel will launch Nehalem-EP server processors later this month, despite their manifestation Tuesday in the new Mac Pro under their official model names: the Xeon 3500 and 5500.

The chips--in their desktop variant known as the Core i7--are being offered in eight-core or four-core configurations and, like all Nehalam-architecture processors, come with an integrated memory controller for (theoretically) better performance. (Intel's Core architecture does not integrate the memory controller.)

Other Nehalem-architecture features include: Hyper-Threading for, according to Apple, "up to 16 virtual cores" (which improves multitasking), and Turbo Boost Technology, which dynamically increases the processor's frequency, as needed.

The Mac Pro also offers high-end Nvidia and ATI graphics. Systems can be configured with either Nvidia GeForce GT 120 or ATI Radeon HD 4870 graphics chips.

Intel has added to its stable of Xeon processors and shaved the price on an Atom chip.

On Sunday, Intel introduced two low-power Xeon processor models rated as low as 45 watts and a higher-end processor.

The L3110 (3.00GHz) integrates 6MB of level-2 (L2) cache memory and is rated at 45 watts, one of Intel's lowest TDP (Thermal Design Power) ratings for a Xeon processor. This is priced at $224.

The Xeon L3360 (2.83GHz) comes with 12MB of L2 cache and is rated at 65 watts. This is listed at $369.

A higher-end X3380 Xeon (3.16GHz) has 12MB L2 cache, is rated at 95 watts, and lists for $530.

The chipmaker also cut the price of the Z530 Atom processor (512k cache, 1.60 GHz) 7 percent, from to $65 from $70.

Unisys may have written Itanium's epitaph on Wednesday--at least for some of the largest server vendors.

Colin Lacey, vice president of Systems and Storage at Unisys, discussed in a phone interview why Unisys--one of the top 10 U.S. server vendors--doesn't see a future for Itanium, including the long-delayed quad-core Itanium "Tukwila" processor.

Lacey said Itanium's appeal has almost vanished for many vendors in server industry. "It's appeal has certainly narrowed down. It's almost exclusively down to a single vendor," he said, referring to Hewlett-Packard. "The current shipping platform is overdue for a technology refresh (and) it's been delayed a couple of times already," he said. In short, Itanium's chronic delays and underwhelming performance mean it's not a viable option anymore for Unisys--which offered Itanium-based servers in the past.

Lacey went on to say that Xeon can be "harnessed" in a high-end server environment to deliver performance that surpasses Itanium with the same reliability. Unisys announced on Wednesday that its newest enterprise server--the ES7000 Model 7600R--has set a record for price/performance in the Transaction Processing Performance Council's (TPC) TPC-H benchmark test. "The performance of the Unisys server, which uses the latest Intel Xeon 6-core processors, shows the increasing superiority of Xeon-based systems for mission-critical applications such as business intelligence over those based on Intel Itanium processors." the Blue Bell, Penn.-based company said in a statement.

Unisys' comments uncannily echo a statement made more than 10 years ago (in 1998) by an analyst who was discussing the long-delayed "Merced" chip--the processor jointly developed by HP and Intel that eventually became Itanium. At that time, an analyst said, "if the performance is similar to Merced," server vendors will opt to "squeeze more profit out of Xeon" instead of adopting Itanium.

Lacey spelled out how Unisys did its testing. "It's not like we're loading the dice. I know that Itanium has only two processors, but the configurations we're comparing have exactly the same number of processor cores. We're comparing 64-core to 64-core. There is no compute engine deficit between one and the other. So, it's looking at the architecture and what works."

With latest generation of Xeon technology "we can deliver pretty compelling raw performance as well as a very significant cost reduction by migrating over to a Xeon architecture. We're talking about a Windows SQL database environment and there's no real pain, if you will, from doing that on Itanium to doing that on Xeon," he said.

At Unisys, there is no significant difference in reliability either, one of Itanium's purported marquee features. "We track the unplanned downtime of our customer base and we track pretty much identical results between Xeon and Itanium architectures with respect to downtime. We don't see any material difference whatsoever and we keep very detailed tracking on that," he said.

Intel, on the other hand, believes both processors offer distinct advantages. "Both platforms offer unique advantages for different needs and applications," Intel spokesman Patrick Ward said Wednesday. "It's great that Unisys is being so aggressive with Xeon's price/performance strengths. Itanium offers terrific scalability and reliability strengths that are a better fit for some of the most mission critical workloads," he added.

Jointly, HP and Intel, not surprisingly, have a different take on Itanium's value for the large enterprise customer. In this video, Intel CEO Paul Otellini and HP CEO Mark Hurd discuss the technology. "Itanium is our architecture for enterprise-class machines. And that's a big segment of the market. About $28 billion. This is a very, very critical architecture for us," Otellini said in the video. "There are things that we're putting into the architecture, the RAS (Reliability, Availability and Serviceability) features, the reliability characteristics, the power-performance characteristics. These are things that we're tuning for the enterprise at the highest end," he said.

"The alliance around Itanium is also unique," Otellini continued. "Multiple operating system environments; 13,000 applications. It's outgrown every other mainframe architecture on the planet over the last five years," he said. "Itanium is about 10 years old now. It's really hitting its stride. It's at the point where we outsell other architectures in Asia. (Sun Microsystems) SPARC and (IBM) Power-based machines," Otellini said. "There are over a thousand silicon engineers at Intel working on this product line," he added.

More details on the Unisys test results can be found in the Unisys February 18 news release.

Intel will have a lot to say at the International Solid-State Circuits Conference, spanning the spectrum of silicon from mobile to server processors. Here are a few of the highlights from abstracts of Intel sessions at the ISSCC, which kicks off Sunday in San Francisco.

Nehalem, currently marketed as the Core i7, will scale down to sub-10-watt chips--that's ultraportable notebook (think MacBook Air) territory:

• "A family of next-generation IA processors...The family has a coherent point-to-point link and integrates memory controller, power-management microcontroller and power-gate transistors and scales from sub-10 to 130W in mobile, desktop and server applications."
  

Part of the message will be more brute-force silicon: more processor cores, bigger caches--especially for Intel's high-end Xeon processor line:

• 8-core Xeon processor (aka Nehalem-EX): "An 8-core 16-thread enterprise Xeon processor has 2.3B transistors in 9M 45nm CMOS...operation up to 6.4GT/s...Core and cache shut-off techniques are used to minimize leakage." (Note: '9M" means nine metal layers; "GT/s" is giga-transfers per second.)
  
• 6-core Xeon (aka Dunnington): "A monolithic 6-core Xeon processor has 1.9B transistors in 9M 45nm CMOS with a 9MB L2 and 16MB L3 cache and exceeds 1M transactions/minute TPCC in 8-socket configuration. The FSB (Front-Side Bus) I/O circuits are implemented in the center of the die to reduce I/O latency. A low-leakage process variant with cache-sleep and shut-off modes enables low-power 6-core 65W and 4-core 50W variants."

And let's try not to forget Itanium--Intel's, some would say, ill-fated silicon for very-high-end severs:

• "The clock system for a 700mm2 65nm quad-core Itanium processor has a cascaded PLL (phase locked loop) architecture and enables dynamic frequency switching."

Intel will also present on graphics-related mobile silicon:

• "A 4-way SIMD (Single Instruction Multiple Data) accelerator for power-constrained microprocessors fabricated in 1.1V, 45nm CMOS occupies 0.081mm2...Enables mode-dependent power savings while achieving wide operating range (1.3V to 230mV) with 2.3GHz, 161mW operation at 1.1V and peak SIMD energy efficiency of 494GOPS/W at 300mV, 50 (degrees) C."

Intel and DreamWorks plan to show off the fruits of their 3D collaboration in a Super Bowl 3D extravaganza this Sunday as DreamWorks prepares to tap into future Intel Larrabee graphics silicon.

The Super Bowl ad sponsored by DreamWorks Animation, Intel, and NBC will feature a 3D trailer of the animated movie Monsters vs. Aliens, coming out in March. A second spot will be a 3D commercial for PepsiCo's SoBe LifeWater energy drinks. Viewers--as they will in the movie theater--will need special 3D glasses to see the effects. (Intel has made 125 million of the InTru3D glasses, which are available for free at stores such as Target and Best Buy.)

Stereoscopy--which in a primitive form has been around since the 1840s--creates the illusion of depth by presenting a slightly different image to each eye. Starting this year, DreamWorks will produce all of its feature films in stereoscopic 3D for use with the special glasses.

DreamWorks CEO Jeffry Katzenberg dons Intru 3D glasses that are used for viewing the Super Bowl 3D trailer of Monsters vs. Aliens

(Credit: Intel)

The InTru3D technology will provide more vibrant colors than traditional technologies that use 3D glasses, according to Jeffrey Katzenberg, chief executive officer of DreamWorks, in an interview posted on an Intel Web page.

"Instead of (traditional) red and blue lensing, there's a different set of filters that are used" that are better at reproducing color, said Katzenberg. "The second thing is a greater level of precision in terms of the broadcast signal--right eye, left eye. The blurry kind of stuff is cleaned up a lot," he said.

But there's a lot more going on with Intel and DreamWorks than meets the eye. Think Intel's future Larrabee graphics chip is just a smoke screen or paper tiger? Listening to Katzenberg, it sounds very real. "We are well on our way to upgrading our software to really take advantage of Larrabee," said Katzenberg, in the Intel video interview. "Larrabee raises the bar of what we can do not just by 2X or 3X but by 20X," he said.

DreamWorks is also using Intel software tools. "This is allowing us to create a completely new paradigm in movies," Katzenberg said, referring to Intel's InTru3D technology, which Intel describes as "uniting the best in computer-generated moviemaking with the latest high-performance processing technologies."

Last year, DreamWorks said it was dropping technology from Advanced Micro Devices in favor of Intel--and at that time the two companies announced a strategic partnership aimed at redefining 3D filmmaking technology. DreamWorks had been in a three-year partnership with AMD.

DreamWorks uses rendering farms with thousands of Intel processing cores to create animation.

Before it adopts Larrabee (later this year), DreamWorks will move part of its rendering farm to Intel's Nehalem processor for servers, due later this quarter.

• Click here for more Super Bowl stories.

Intel is planning price cuts to its lower-end mainstream quad-core processors on January 18.

Barron's Tech Trader Daily first reported the news, citing Pacific Crest analyst Michael McConnell.

These cuts are happening because of the recent introduction of Advanced Micro Devices' 45-nanometer Phenom II and "Shanghai" Opteron processors.

AMD's quad-core Phenom II "Dragon" processor platform has been garnering solid reviews and its Shanghai server chip has been adopted by top-tier server suppliers including Hewlett-Packard, Sun, Dell, IBM, and Fujitsu.

Cuts are expected mainly on quad-core processors, though other processors may also receive cuts.

Small start-up Tilera still beats chip giants like Intel handily on core counts. Tilera updated its line of many-core processors Monday, adding a 36-core version to the mix.

Tilera, which made a splash last year when it introduced its first 64-core processor, announced a scaled-down 36-core Tile processor on Monday, in order to broaden its market reach.

The TilePro36 "is giving us a midrange product. This type of device would be used in a high-end video conferencing (system)," said Bob Doud, who is the director of marketing at Tilera. The TilePro36 chip is also targeted at applications such as networking and security applications. "Anything in the one- to five-gigabit-per-second range," Doud said.

Along these lines, Doud also clarified the markets that Tilera is not targeting. "We target the embedded space. We don't target products like servers and home PCs," he said.

Tilera has also updated its original 64-core Tile64 to TilePro64 by doubling the performance while keeping the power consumption increase to less than 5 percent.

Key features of the TilePro36 and TilePro64 processors.

(Credit: Tilera)

The TilePro family improves the performance of highly threaded and shared-memory applications through the introduction of Dynamic Distributed Cache technology, according to Doud. "We've been able to double the performance of our caching system that supports threaded programming models," he said.

The TilePro processors have twice the level-1 cache memory of the previous generation and incorporate an additional on-chip communication network dedicated to cache management. Cache memory is crucial component for increasing processor performance.

In the Pro family, the "big breakthrough" is "making caching and cache coherence scalable as you start to go into more and more cores," according to Doud. In other words, designing the cache memory so it can be shared efficiently as more cores are added.

Tilera has also added new instruction set extensions for audio and video that deliver up to 2X improvement in multimedia signal processing.

So, what is the Tilera chip? The processors consist of small, individual building blocks, or tiles. Each tile has a RISC processing core that runs up to 866MHz as well as a switch that can send data in four directions: up, down, right, and left. These switches form a mesh network, called iMesh, that lets the chips communicate.

Each tile contains its own cache and the tiles can access all of the cache, depending on how it's programmed.

Power consumption is low for a many-core processor, using only between 10 and 16 watts for the 36-core version. Intel dual-core processors typically use over 25 watts.

The key to the chip's performance is tied to the nature of a distributed network. A network of slower processors can get jobs done quicker (in some cases) and with less overall energy than faster quad-core processors from Intel, for example, that have more complex cores.

The company say it now has more than 45 customers.

As Intel officially unveiled its six-core "Dunnington" Xeon 7400 processor Monday, Unisys rolled out servers boasting up to 96 cores--with a catch.

As expected, Intel launched the Dunnington chip for high-end servers, the company's first six-core processor and last of its Penryn-class chips. Penryn will be followed by the Nehalem microarchitecture, due to appear initially as the Core i7 processor in the fourth quarter.

The Xeon 7400 boasts significantly better performance due to its 16MB cache memory and half-dozen cores.

Intel Xeon 7400 is first 6-core processor

(Credit: Brooke Crothers)

The Xeon 7400 is also one of the first Intel chips to have a monolithic design. In other words, all six cores will be on one piece of silicon. To date, for any processor having more than two cores, Intel has put two separate pieces of silicon--referred to as die--inside one chip package.

Unisys is in the vanguard of server vendors offering systems using the 7400 series processor. On Monday, the Blue Bell, Pa.-based computer vendor announced the ES7000 Model 7600R Enterprise Server, a 16- socket server providing up to 96 processor cores.

In addition to being better at handling complex database applications, one of the most compelling features of six-core servers is consolidation: collapsing many servers into a few servers. Unisys said it has demonstrated consolidation of 64 SQL Server databases into a single four-socket, Xeon 7400 processor configuration, with 24 total processor cores. In essence, this collapses a conventional "commodity server farm" of 64 single-socket, dual-core Xeon processor servers into a single server configuration, Unisys said.

There's an odd catch, however, that will affect the highest of high-end configurations. "Because Microsoft Windows operating system support is limited to a 64-core environment, within a single OS instance, we'll support up to 64 cores," said Colin Lacey, a Unisys marketing vice president.

"You'd actually have 96 cores physically within the system. But then you would disable two cores in each socket. So you'd actually be running these sockets at four active cores each (out of six)," Lacey said.

Lacey said this condition is necessary to deliver the highest performance in a Xeon 7400-based server running Windows, though he expects to rectify the 64-processor limitation in the future. He added that he wouldn't consider 64 cores to be a limitation in the "real world" for most customers, who would in most cases opt for servers with a smaller number of cores.

Linux does not have this limitation, Lacey said.

Advanced Micro Devices pre-emptively chimed in on the Xeon 7400 series last week. "Intel has taken the old front-side bus architecture and added 6 cores to it," according to an AMD statement. The company's Opteron processors have jettisoned the front-side bus--a data path between the processor and the memory controller--in favor of putting the memory controller on the same piece of silicon as the processor to speed performance. Intel has done this with its upcoming Nehalem architecture too.

Prices for the ES7000 Model 7600R range from $26,430 to $135,000.

Software vendors are also supporting the Xeon 7400 based platforms, including Citrix, IBM, Microsoft, Oracle, Red Hat, SAP, and VMware, Intel said.

Pricing for the Xeon 7000 Sequence processor in quantities of 1,000 ranges from $856 to $2,729.

Other server vendors announcing servers include Sun Microsystems, Hewlett-Packard, and Dell.