Chip Roundup: AMD's Lisbon Punctuates Q2

Lightning-Fast Xeons For February?

Here’s a heads up for builders of whitebox servers -- Intel will bring out a 3.6-GHz, Westmere-class Xeon server chip in February 2011 that could flirt with 4.0GHz without overclocking, courtesy of the chip giant’s built-in Turbo Boost technology, according to an Intel product road map chart obtained by Custom Systems Magazine.

The Xeon X5687 is a 32-nanometer, quad-core processor that is part of a big update to Intel’s Xeon 5600 Series of dual-socket server chips planned for early next year. The X5687 has 12 MB of L3 cache, eight compute threads and draws 130 watts of power.

The coming Xeon 5600 Series release also includes a 3.46-GHz, six-core processor, the Xeon 5690, also with 12 MB of L3 cache and sitting in the 130w thermal envelope.

Capabilities on both chips include Hyper-Threading, AESNI, Intel Virtualization Technology (VT), Trusted Execution Technology (TXT) and, of course, Turbo Boost. Based on maximum Turbo Boost frequencies listed for current Xeon 5600 Series chips, a quad-core, 3.6-GHz chip like the Xeon X5687 ought to throttle up a single core to about 3.9GHz when the other three cores aren’t in use, and Intel’s Turbo Boost technology automatically and dynamically gauges that certain thermal conditions are met.

Intel released the first 12 processors in the Xeon 5600 Series in March. Also known by the code-names Gulftown and Westmere EP, the 5600 Series is made up of 32nm processors with as many as six cores and 12 compute threads.

Clock speeds for the current Westmere EP product line top out at 3.46GHz for the quad-core Xeon X5677, currently priced at $1,633, and 3.33GHz for the six-core Xeon X5680, also priced at $1,633 in 1,000-unit quantities. Prices for the Xeon 5600 Series chips planned for February 2011 were not listed on the Intel road map chart obtained by CSM.

NEXT: GPU Computing War Heats Up

GPU Computing War Heats Up

Nvidia, headquartered in Santa Clara, Calif., has led the way in recent years in pushing the boundaries on just how much traditionally CPU-handled computing can be taken on by graphics processors. The presence of so many GPUs in the world’s top supercomputers is just one indicator of what the graphics chip maker has done right.

There’s also the entire CUDA ecosystem of software developers and custom system builders that Nvidia has built up from scratch. And how about the annual Nvidia-hosted GPU Technology Conference, which will kick off its third edition in San Jose, Calif., in late September?

Throw in Nvidia’s Tesla Preferred Partner program, which has received glowing accolades from larger system builders like AMAX Information Technologies, who have jumped feet-first into the business of providing customers in academia and a variety of industrial verticals with GPU-based supercomputers for compute-intensive workloads.

And lastly, Nvidia has been steadily introducing its powerful new Fermi architecture to 2010 refreshes of its three main product lines -- first to its GeForce consumer products, then to its high-end Tesla GPUs and finally, in late July, to its Quadro professional graphics cards.

But now Nvidia may just have woken up a sleeping giant for all its efforts. After seemingly ceding the evangelism to its competitor for years, AMD has recently started to show some signs that it’s ready to play harder in the general-purpose GPU computing theater.

AMD, which acquired graphics chip maker ATI Technologies in 2006, launched a pair of teraflop-chomping, fanless FireStream GPU accelerator boards for servers and workstations in late June.

About a month earlier, the company poached CUDA integrator Manju Hegde from Nvidia to head AMD’s Fusion Experience Program.

As AMD readies itself for its Fusion era in 2011, when the chip maker plans to start shipping chips that integrate CPU and GPU functionality on a single silicon die, Hegde is charged with building up a similar partner and developer network for those products to the one Nvidia currently enjoys.

One can only wonder what Intel’s plans are in this increasingly interesting space, what with its own Larrabee discrete graphics project seemingly shelved.

AMD Launches Lisbon

AMD served up perhaps the most important new processor product launch of the second quarter when the Sunnyvale, Calif.-based chip maker unveiled its Opteron 4000 Series of server processors in late June. Formerly code-named Lisbon, the Opteron 4000 processor series for single- and dual-socket servers and workstations features one chip priced at less than $100 and others that draw as little as 35 watts in power.

The initial Opteron 4000 Series includes nine server processors priced from as low as $99 to $698. All but two of the new Lisbon chips have six processor cores.

“This is the second part of our server platform strategy, adding to the Opteron 6100 Series,” said Brent Kerby, senior product manager of AMD’s Server and Workstation Division.

AMD released the multicore Opteron 6100 server platform for servers with two processor sockets or more in March. That lineup of products, formerly code-named Magny-Cours, features server chips with as many as 12 processor cores and targets the traditional industry-standard server market in tower, rack and blade systems.

The Opteron 4000 Series features several processors that sit in exceptionally low power bands—a key demand from cloud computing and scale-out customers, Kerby said. AMD expects to see its Lisbon chips in custom, purpose-built systems and appliances, “skinless” servers for extreme scale-out deployments, containerized data centers and some entry-level SMB-targeted servers.

“We’ve actually had customers say, ‘We like your processor but can you ratchet back performance to save more power?’ That’s unheard of in my experience. It’s like, ‘Wow, what a change,’” he said.

The early returns on Lisbon from AMD’s system builder partners back up that claim.

“Performance is still important but what we’re hearing from our customers is that high core counts and just pure horsepower are not their sole criteria,” said Travis Scott, director of marketing for Secaucus, N.J.-based ZT Systems.

Scott said ZT Systems is seeing “a lot of momentum in the marketplace around the cloud” -- whether that means customers outsourcing IT to public cloud providers or building their own private clouds -- and that has put concerns about power efficiency, density and thermals in the context of scale-out computing in the data center at the top of many customers’ minds.

“Decisions now are being made in terms of a compromise of certain vectors. Power efficiency is pretty much involved in every discussion,” he said.

AMD’s new processor lineup includes two six-core chips that sit in the 32-watt thermal envelope and draw 35 watts of power, the 1.8-GHz Opteron 4164 EE, priced at $698 in 1,000-unit quantities, and the 1.7-GHz Opteron 4162 EE, priced at $316. The chip maker is claiming they’re the most power-friendly x86 server chips in existence.

NEXT: Components Shortages Bedevil Semiconductor Firms

Components Shortages Bedevil Semiconductor Firms

A number of sources reported that key components used to manufacture semiconductors were in short supply towards the end of the second quarter of 2010. In addition to a dearth of DRAM -- and really, when is memory ever not suffering a shortage? -- research firm iSuppli reported that in June, lead times for ordering key components like power MOSFETs and rectifiers were hitting 20 weeks.

Intel reportedly had some trouble in that time frame meeting demand for some of its Arrandale-class processors designed for thin-and-light notebooks, notably Core i3 and Core i5 chips featuring integrated 45nm GPUs in the same package as 32nm CPUs.

But by mid-July, contract notebook manufacturing giant Quanta said Santa Clara, Calif.-based Intel’s Arrandale shortages were in the rearview mirror. Quanta is the largest contract laptop maker in the world and completed its best-ever month in June with 4.8 million units shipped -- so we’re guessing the Taiwanese manufacturer is in a position to know.

Intel’s Mobile Moves

Finally, as CSM went to press, Intel was rumored to be on the verge of completing a billion-dollar deal to acquire Infineon’s mobile chip business -- a move that would mean a great deal for Intel’s chances of gaining ground in the handset and smartphone market it so desperately wants to crack.

The German semiconductor company competes with such companies as Broadcom and Marvell in supplying chips to the likes of Apple, Nokia and Research in Motion for their handheld products based on non-x86 chips.

Such a development might not immediately affect Intel’s traditional partner channel of whitebox builders and system integrators, who continue to exclusively produce x86-based PC clients and servers. But at the very least, it’s a hint that the chip giant is deliberating a break from its long-standing policy of insisting that the Intel architecture is the only one worth selling.

When all you have is x86 chips, everything looks like a PC -- and just maybe Intel is starting to reconsider the wisdom of that position.