Nanotech - The Circuits Blog

Advanced Micro Devices' server roadmap is solid but its mainstream mobile lineup is languishing.

AMD's six-core "Istanbul" processor release was moved up to June

(Credit: AMD)

First, the good news. These days AMD is walking the talk. This is a radical change from the AMD of 2007-2008, which always seemed to have a hopper full of Intel-vanquishing paper processors that, if they did materialize, disappointed.

Fast forward to AMD's Tuesday earnings announcement, when the company said it was actually moving up the introduction its most sophisticated processor, the six-core Istanbul, to June.

And AMD has proved its silicon mettle at large server customers such as IBM and Sun Microsystems--the latter's executive vice president John Fowler had nothing but praise for AMD processors in high-end Sun server systems.

In a "Global Webcast" on server technology Wednesday, Patrick Patla, a vice president in AMD's server and workstation business, revealed a strong roadmap, saying that 8- and 12-core "Magny-Cours" processors will appear in 2010. "We're currently working on new processors which we expect will deliver more than 35 times the performance of the original single-core AMD Opteron processor released in 2003," Patla said in a statement.

Intel, of course, will also bring out many-core processors, but AMD is keeping pace, and, according to people who should know, like Sun's Fowler, maybe more than keeping pace.

Now, the bad news. This post today on CNET's Crave blog says it all: "One of our biggest issues with HP's Pavilion dv3z was its AMD processor, keeping it from beating out the performance of comparable 13- and 14-inch laptops with Intel Core 2 Duo CPUs." This is, by no means, the first review that expresses this sentiment. In short, AMD mobile platforms consistently come up short in the high-profile, burgeoning laptop market. Will AMD close the gap in 2009?

Maybe one answer to that question is AMD's Neo chip that powers the low-cost, ultra-thin HP Pavilion dv2 laptop. More than a Netbook but less than a mainstream laptop, this kind of sleek mobile device could eventually eclipse the high-end Netbook segment.

AMD CEO Dirk Meyer said Tuesday that the single-core Neo processor will get a dual-core sibling dubbed "Congo" by summer. A dual-core processor in this low-cost, MacBook-Air-for-the-masses category is a compelling proposition. AMD needs to stay ahead of the game, especially when Intel brings out chips for this category in the May-June timeframe.

In the wake of reported merger talks with IBM, Sun Microsystems executive vice president John Fowler talked about Big Blue as a rival.

Sun Microsystems executive vice president John Fowler

(Credit: Sun Microsystems)

Fowler, in a phone interview Friday, discussed IBM as a competitor in the server computer market and the competitive differences between Intel and Advanced Micro Devices.

"IBM is obviously a classic competition space where we bring to bear all of our technology innovation," Fowler said. "In this particular case, how we've incorporated and done networking technology, how we've incorporated and done flash (memory). The fact that we have an open operating system that runs across our RISC platforms and our x86 platforms and that we have an open storage offering on the 7000," he said.

Fowler continued, "Those are all great discussions to go have with IBM customers. Those are all things that IBM just doesn't have and represent a significant amount of value."

In a question about comparing IBM--which also offers a RISC architecture server platform (PowerPC) as well Intel and AMD--to other competitors like Hewlett-Packard and Dell, Fowler said: "They (IBM) are more similar (to Sun) than the others and interested in more of the same things."

And how does Sun compete with larger rivals? "We have for decades now innovated in a marketplace where companies like HP and IBM have been consistently bigger than us. What we do is incorporate new technologies or invent new technologies more quickly," he said.

Fowler also spoke of some of the differences between Sun offerings on AMD and Intel platforms. He began by saying that a recent Intel Nehalem server chip announcement covers only two-socket (a socket accommodates one processor) servers and that Intel's product is an "entry level" product. "It's a solid product but it really applies to the volume entry-level systems in the world today," he said.

Fowler continued: "Obviously, AMD and (Sun) SPARC cover a much broader range of application scale. In AMD's case, extending up to eight-socket servers and in SPARC's case up to 64-socket servers," he said.

An IBM server vice president discussed IBM's strategy to tap into Sun Microsystems' customer base, in the wake of reports that IBM is in talks to buy the Santa Clara, Calif.-based server supplier.

IBM is rumored to be working toward a merger with Sun mostly due to the strength of Sun's server business. SPARC is Sun's chip architecture, while Solaris is Sun's operating system that runs on both SPARC chips and x86 processors from Intel and Advanced Micro Devices.

"Sun has a terrific installed base," Alex Yost, vice president IBM BladeCenter, said in a phone interview earlier this week in response to a question about Sun as a competitor.

Yost went on to say that IBM has an active business of migrating customers from Sun's SPARC architecture to x86-based servers. "I have a number of clients that are looking to go to Solaris on x86 or Linux on x86," he said. "That's very much something that we're actively doing."

Yost added that there are some IBM customers that require Sun's SPARC architecture. "We also have some clients in very specialized environments that require native Solaris on SPARC," he said. For these clients, IBM has partnered with Themis to offer SPARC blade server on IBM BladeCenter, Yost said.

Part of the challenge of absorbing Sun would be to integrate Sun's products with IBM's. On its Web site, Themis describes its T2BC Blade Server as enabling Solaris applications "to run natively, on an UltraSPARC T2 chip...within an IBM BladeCenter." The Themis product description continues: "The T2 Blade Server can share the same chassis with server blades that utilize other processor architectures and operating systems."

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.

New "Nehalem" servers will anchor Intel's renewed push into cloud computing, as the chipmaker focuses on mega data centers with hundreds of thousands of servers.

Intel's cloud-computing efforts this year will be centered on a new server that uses upcoming Nehalem technology, Intel said Tuesday in a teleconference on its cloud-computing strategy. Nehalem is Intel's new chip architecture currently used only in its Core i7 desktop processors.

Mega data centers potentially mean mega-growth. The world's largest chipmaker sees between 20 percent and 25 percent of server shipments going to mega data centers by 2012. Today mega data centers represent about 10 percent of the server market, according to Intel.

And what is cloud computing to Intel? A cloud architecture aimed at mega data centers with hundreds of thousands of servers that "can be balanced automatically. Automatically resized and scaled," according to Jason Waxman, general manager of high-density computing at Intel's Server Platforms Group. "Your service is stateless: it's not the same server every time. At any point in time I'm not necessarily accessing the same server."

Intel's goal is to optimize this massive mesh of server hardware. "Optimization is key. When you're talking about hundreds of thousands of servers, every server, every watt, every network connection represents cost," he said.

Waxman said Intel will use its upcoming Nehalem silicon to spearhead its renewed push into mega data centers. "We've designed a server for a Nehalem-based board that's optimized for our cloud-computing infrastructure," said Waxman. The "Willowbrook" motherboard will be launched later this quarter, according to Waxman.

Willowbrook is designed with "very efficient voltage regulation," he said, and "we've optimized the layout of the boards" so air can flow more efficiently across the board. Waxman added that "idle power" has been reduced--a crucial metric for mega data centers. "We've been able to take out power. At idle, a standard Nehalem platform consumes 110 to 115 watts, we've been able to get it down to the sub-85 watt range," he said.

Overall, optimization and power savings boils down to cost. For a large cloud service provider, 50 percent of the total cost is the compute infrastructure--servers and storage--and 25 percent is delivering the power and cooling, he said. "75 percent of the (total cost of ownership) is computer, power, and cooling. And this is what Intel is focused on. Optimize the servers and get every watt we can out the servers."

Waxman said repeatedly that Intel is not going to be a service provider but wants to enable customers to take advantage of Intel cloud-computing technology. "We're not trying to become a service provider but we bring all this core technology and expertise together. The capability to look at a cloud and optimize it," he said.

He cited Salesforce.com, IBM, and Microsoft as service providers and added that "it's sort of a wild west frontier" as many of the more comprehensive cloud-computing service products from major companies are not in production yet.

Other technologies that Intel will roll out with Nehalem server chips include Virtual Machine Device Queues (VMDQ) that allow traffic to be queued up and aim to resolve an outstanding problem in which one virtual machine can hog all the bandwidth. Waxman also discussed the "I/O hub" technology that Intel is implementing with Nehalem. "It has a tremendous number of PCI Express Gen 2 lanes. Gen 2 for speed and more lanes--that's kind of our strategy," he said. The Peripheral Component Interconnect or PCI bus is a data path to a computer's peripheral devices such as a network card or graphics card.

Waxman also discussed a Node manager. "Within a data center, I'm trying to figure out how to use as many servers as I possibly can and one of the challenges of optimizing a cloud is how do you make sure you don't overload a server and create a server hot spot," he said. The Node manager will reside in the motherboard BIOS, he said.

Intel has delayed a high-end server chip, billed as the world's first 2 billion transistor microprocessor, originally expected as long ago as 2007.

Tukwila is a quad-core update to the Itanium processor, which has had a less-than-promising run since the original version was announced back in 2001. The chip's architecture--based on explicit instruction-level parallelism--is a radical departure from the x86 design used in PCs today. It was believed at one time that Itanium would replace x86 chips in many Intel-based computers.

So, what's delaying it this time? Intel has "made some engineering enhancements to the Tukwila platform," according to a statement Thursday from Intel. As one enhancement, Tukwila and its follow-on processors on the Itanium roadmap, Poulson and Kittson, will be socket compatible. Intel is also introducing a technology called "scalable buffered memory" to Tukwila platforms. Scalable buffered memory enables higher memory capacities and uses newer DDR3 (Double Data Rate, third generation) memory.

The number of CPU pins that can be dedicated to a memory interface is limited. Scalable buffered memory acts as a memory expander by connecting to multiple DDR3 RDIMMs (Registered Dual In-line Memory Modules) for each CPU memory interface, thus increasing the total memory capacity compared to natively attached memory configurations, according to an explanation provided by Intel.

"Validation testing of the Tukwila processor with the new memory will move the Tukwila platform launch from early '09 to mid '09," and Intel spokesperson said.

Current and past users of Itanium processors include Hewlett-Packard, SGI, NEC, Fujitsu, Unisys, and Hitachi.

The processors are targeted at high-end enterprise servers and high-performance computing systems.

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.

Intel is shipping new server processors that consume as little as 12.5 watts per core.

Cumulatively, the racks and racks of servers in large data centers can require power rivaling that consumed by entire city blocks. So, getting power consumption as low as possible while delivering adequate performance has become a delicate balancing act for Intel and Advanced Micro Devices.

New additions to the Xeon 5000 Series include the L5430 and X5270 processors, the fastest of which boasts a clock speed of 3.5GHz, Intel said.

The low-voltage L5430 uses only 50 watts of power or just 12.5 watts per core. The dual-core Xeon X5270 draws 80 watts, according to Intel.

"Much of the energy efficiency these new processors provide comes from Intel's...45 (nanometer) manufacturing capability and its reinvented transistors that use a Hafnium-based high-k metal gate formula," Intel said in a statement. Transistors with high-k metal gates can control current leakage better than those with silicon dioxide gates, which Intel had used in the past.

Not all processors, however, boast low power consumption. Intel will also ship high-performance versions with relatively high power consumption, including the X5492, which consumes 150 watts.

All of the new Xeon chips use packaging materials free of halogens, which can release toxins if incinerated. This is a goal Intel has set for all of its processors.

Vendors expected to bring out systems include Asus, Dell, Fujitsu, Fujitsu-Siemens, Gigabyte, HP, IBM, Microstar, NEC, Quanta, Rackable Systems, Sun Microsystems, Supermicro, Tyan, and Verari Systems.

The processors are targeted at organizations using workstation and blade and mainstream servers, Intel said.

Pricing ranges from $562 for a quad-core Xeon L5430 (2.66GHz) to $1,493 for a quad-core Xeon X5492 (3.4GHz).

Samsung and Sun Microsystems have developed a flash chip for use in solid-state drives that offers higher endurance levels than current devices, the companies say.

The chip is targeted at server applications.

Samsung solid-state drive

(Credit: Samsung)

The Samsung flash memory chip is based on single-level-cell (SLC) NAND flash technology and offers a fivefold increase in data write-and-erase cycles over standard SLC flash memory, according to Samsung.

SLC-based flash chips are faster and offer more write-and-erase cycles than multilevel cell (MLC)-based devices. MLC, however, offers greater capacities and lower cost, making it suitable for notebook computers. MLC-based solid-state drives from Samsung, Intel, Micron Technology, and Toshiba are expected to flood the market in the second half of the year.

Samsung's new SLC NAND memory is designed for server-grade SSDs, particularly high-transaction data processing servers. The "ultra-endurance" memory has been developed in close cooperation with Sun over the past several months, Samsung said.

"We have been working with Sun to develop this new...server-grade SLC flash memory, which will give IT managers the best in high-density, high-endurance memory design with markedly less energy consumption than we see today," Jim Elliott, vice president of memory marketing at Samsung Semiconductor, said in a statement.

Probable applications include video streaming and search engine operations. The chip will provide a 100X increase over conventional hard disk drives in the number of data transfers (input/output per second or I/OPS) per watt, offering significant power savings, according to Samsung.

Sun said it plans to incorporate the technology into its line of servers and storage, according to Michael Cornwell, lead technologist for flash memory at Sun.