"Dawning 6000" to use Chinese-made Loongson processor

The high-performance computer "Dawning 6000," which will have a computing speed over 1,000 trillion operations per second, will adopt the Chinese-made general processor Loongson for the first time as its core component. This is according to the Dawning Information Industry Company.

"Dawning 6000" is currently jointly developed by the Institute of Computing Technology under the Chinese Academy of Sciences and the Dawning Information Industry Company.

Li Guojie, chairman of Dawning Information Industry Company and director and academician of the Institute of Computing Technology, said research and development of the Dawning 6000 is expected to be completed in two years. By then, Chinese-made high-performance computers will achieve two major breakthroughs: first, adopting domestic-made central processing units (CPUs) will be technically obstacle-free; second, the existing cluster-based system structure of high-performance computers will be changed once the computing speed reaches 1,000 trillion operations per second.
Resource - People’s Daily Online

Yukon Ho AMD shows off new ultraportable roadmap

Yesterday, AMD announced that its integrated CPU+GPU Fusion processors wouldn't debut until 2011. Instead, we'll see a plethora of new mobile parts and chipsets rolling out of Sunnyvale through 2009, including new mobile platforms codenamed Congo and Yukon. Both will debut in the first half of 2009, with Yukon positioned as the most netbook-like solution. AMD isn't really using that term, however, and is attempting to position these products as "full-featured" ultraportables with form factors that more closely resemble the Macbook Air.

Let's pause a moment, and define a few terms. Yukon and Congo are high-level terms that refer to an overall platform. Congo, the higher-end version, will use AMD's upcoming Conesus CPU. Conesus, shown below, is a 65nm, dual-core CPU with 1MB of cache (presumably 512K per core), and will support DDR2. Combined with the Congo platform, it will carry an RS780M GPU, the SB710 southbridge, and will support at least some of the Radeon's HD decode features. Yukon is the low-end variant of Congo and will be paired with a low- end Conesus-derived CPU, codenamed "Huron." This new chip will pair up with the RS690E video core and an SB600(?) southbridge.

I dropped in that question mark because I'm having a hard time seeing the logic for an SB600 in an ultraportable or mini-notebook. SB600 used more power than SB700, offered less connectivity, and was a bit buggier than the SB700 and SB750 that have since replaced it. I've no doubt it can do the job, but it's an odd choice for an netbook ultraportable. AMD is targeting a sub-25W TDP for both of these platforms. Granted, that's well above Intel's Atom, but disclosing total platform TDP as opposed to CPU TDP alone is a more accurate representation of how much power the shipping product is going to require.

As for Huron and Conesus, a quick check of AMD's existing product line yields several interesting possibilities. AMD currently has two embedded solutions that might fit Yukon's power consumption profile. The first is virtually a guaranteed fit—it's an integrated Athlon 64 2000+, with a 1GHz core clock, 512K L2, and an 8W TDP. I say "guaranteed," in this case, because if AMD can't deploy a motherboard chipset+IGP that draws less than 16W, the company has no business in this space to begin with. There's also an embedded Athlon 64 2600+—single-core, 15W, 512K of L2 cache. That's a bit tighter of an envelope, but it's still potentially possible.

Conesus is a bit trickier. AMD currently only has just one mobile core in the 25W space, the QL-62 (2GHz, 1MB L2 cache). Then again, since this is a 25W part at 2GHz, it shouldn't be too hard to knock 500MHz off the core, trim the voltage, drop the system bus from 3600MHz to something in the 2GHz range. The reason I'm spending so much time discussing current options, for the curious, is because I seriously doubt Conesus or Huron are actually new CPUs. They might be respun for lower power consumption (we know AMD's 65nm process technology has improved in that regard, based on current Phenom TDPs compared to what the company originally shipped), but the CPUs themselves, I think, are likely to be K8 derivatives.

I have seen some questions on whether or not AMD could ship this sort of part in time to compete with Intel's Atom refresh next year, but again, I don't think that's an issue, for reasons of power consumption and the fundamental performance differences between an out-of-order execution engine and an in-order design. Attempting to compare the two, at this point, smacks more of an inherent desire to grab onto something that's currently popular (Atom) and associate it with something new (Huron/Conesus) even if the two have very little in common.

Resource - Ars Technica

AMD Fusion now pushed back to 2011

One of the updates that came out of AMD's Financial Analyst day today is that the company's much-discussed Fusion CPU+GPU hybrid will not appear until the company transitions to 32nm technology. That means no Fusion CPUs until 2011 at the earliest. The Shrike platform, once planned to be some sort of next-generation Shanghai + integrated GPU on 45nm has now been canceled altogether. Long-term, AMD still claims to believe in the power of Fusion—just not on a 45nm process.

This may be disappointing news to some, but it's probably a good move for the company. It was never very clear how AMD was going to economically deliver an integrated GPU on 45nm, even if it went for two separate dies on the same physical packaging. A Fusion part needs to be cheap enough that OEMs prefer it (as opposed to an integrated GPU + discrete processor), while simultaneously performing well enough to make customers want it. It can't be all that expensive, either, considering the fact that a relatively modern HD Radeon 3450 sells for just $29 at NewEgg. Waiting for 32nm to come online buys the company both breathing room and die space, and AMD could use a bit of both.

A block diagram of Fusion

Over the next two years, AMD plans to split itself in half, successfully launch The Foundry Company, manage its own 32nm transition, and break ground on Fab 4x in New York, which will come online as a 32nm plant. Bobcat and Bulldozer, meanwhile, may have been delayed, but they aren't gone; both are currently scheduled for launch in 2011. On the server side, AMD has to scale Shanghai through its 45nm lifespan, release the six-core Constantinople Istanbul variant (coming in second half 2009), and successfully launch its first in-house server chipset, Fiorano. On the GPU side, AMD must continue to gain ground (or hold what it has gained) in the GPU market, and transition the Radeon series down to 40nm, with 32nm possibly following after.

That's an aggressive launch schedule, particularly from a company in as precarious a position as AMD is, and successfully executing it will require a great deal of skill. Shanghai is far stronger out of the gate than Barcelona ever was, but it'll be quite some time before Sunnyvale is back on any kind of good financial footing. As for what's coming post-Shanghai on the desktop, AMD did release a few details. It plans to introduce two 32nm Bulldozer cores in 2011, Llano and Ontario. Llano will be a quad-core part with support for DDR3 and up to 8MB of cache, while Ontario will be a dual-core laptop part with an integrated GPU. There's no word yet on the company's "Quebec" processor, but we expect that core will feature AMD's Separatist Technology. Rumor has it that the project is bogged down in negotiations, most of which are aimed at persuading the upcoming chip that it is not, in fact, a Xeon.

Resource - Ars Technica

November 13, 2008 -By Mark Hachman AMD Answers Atom with 'Conesus,' Roadmap Update

AMD dramatically updated its desktop roadmap on Thursday at its analyst meeting, revealing a top-to-bottom approach to address the market from ultraportables and netbooks on up to high-end entertainment PCs. Included in the update was the company's answer to Intel's Atom processor.

AMD plans six platforms: "Yukon", for the netbook market, due in the first half of 2009; "Tigris," for the mainstream notebook, due in the second half of 2009; "Kodiak," for business desktops, scheduled to be released at the same time; and Pisces, a similar stratregy for consumer desktops, also planned for the second half of 2009. In the near term, however, is "Maui", due this quarter for home theater PCs, and "Dragon," for the entertainment PC category, which will be released in the first quarter of 2009.

AMD also added six new cores to its processor roadmap, extending it into 2009.

For mini-notebook and netbook enthusiasts, the key additions are "Caspian" and "Conesus," both 45-nm cores apparently built on the same architecture as the "Shanghai" processor AMD introduced on Thursday and its desktop counterpart, Deneb, which will be launched early in 2009, at the Consumer Electronics Show in January, according to a leaked roadmap. That roadmap did not identify either Caspian or Conesus by name; speculation had been that the two would be based on the Geode processor. Randy Allen, the senior vice president of AMD's Computation Solutions Group who identified the new cores here, did not specify whether the two new chips are indeed Geodes.

click on image for full view

In any event, both cores mean that AMD now has an answer to the Atom processor that AMD's chief rival has used to power netbooks. Both Caspian and Conesus will be dual-core parts, each with an integrated DDR-2 controller. Caspian, the processor designed for ultraportable notebooks, will contain 2 Mbytes of cache; Consesus will contain just one. Conesus will also be manufactured using a BGA package, allowing it to fit within the cramped confines of netbooks.

click on image for full view

However, the new chips aren't specifically designed for netbooks. AMD chief executive Dirk Meyer said that since AMD had been unable to assess the total available netbook market so AMD wasn't directly addressing it for now. "First order, we're ignoring the netbook phenomenon, concentrating on PC notebooks above that form factor," Meyer said.

Allen offered further clarification, stating that AMD will cede part of the netbook market to Intel, including so-called Mobile Internet Devices. Customers of the Yukon netbook market don't want a "compromised PC experience," Allen said. "We will deliver a full-fledged PC experience," Allen said. "We won't be going to the bottom where Atom is going, it won't be to the lowest [thermal design power] where Atom is going."

AMD's 2009 desktop roadmap remains largely unchanged: on the high end of the desktop lies "Deneb," a four core chip with 8 Mbytes of cache, accessing either DDR-2 or DDR-3. The mainstream desktop will be built on "Propos," a quad-core chip with 2 Mbytes of cache, also connecting to either DDR-2 or DDR-3.

In 2010, Caspian will be replaced with Champlain, which will move the ultraportable notebook category into quad-core territory. Conesus, meanwhile, will be replaced with "Geneva," still a dual-core part. Champlain will contain four cores, connect to DDR-3 memory, and includes 2 Mbytes of cache. Geneva, also a BGA-based chip, will contain just two cores, but will be otherwise identical to Champlain, the roadmap indicates.

AMD also extended its roadmap into 2011, farther than the chipmaker has previously gone before. At that time, netbooks will be anchored around the "Ontario" processor, a dual-core, BGA-packaged chip that will be AMD's first to integrate graphics functionality. Ontario will contain 1 Mbyte of cache, and connect to DDR-3 memory.

At present, the 2011 roadmap also calls for both notebooks and desktops to be served by a single processor: Liano, a four-core part that also contains a GPU. Sporting four cores and a 4-Mbyte L3 cache, it too will connect to DDR-3 memory.

AMD also removed "Shrike" from the roadmap, previously the first of the so-called "Fusion" processors combining CPU and graphics capability. Allen explained that Shrike offered just modest improvements, while Liano promised "huge" improvements.

On top of AMD's roadmap in 2011 will be "Orochi," adding more than four cores and more than 8 Mbytes of cache. Further details were not disclosed

Resource - ExtremeTech

AMD introduces 45nm quad-core “Shanghai” Opteron processors

Today AMD announces 45nm quad-core Opteron processors, codenamed “Shanghai” designed to deliver superior virtualization experience, performance and power efficiency.

The touted highlights of these new pieces of silicon are many and varied:

• Extending virtualization capabilities.
• Up to 35% more performance compared to “Barcelona” Opteron CPUs while having the same power envelope.
• The only x86 processor spanning 2, 4 and 8 processor servers while maintaining socket and thermal compatibility with previous generation processors.
• Approximately 35% less power usage at idle “Barcelona” Opteron CPUs.

A lot is being made of the virtualization capabilities of these new Opteron processors:

Here’s an interesting video for those thinking about moving from Intel to AMD. It shows a live migration from Intel to AMD via Red Hat virtualization:

These new “Shanghai” Opteron processors deliver superior performance at a lower price point than Intel can deliver. For example;

• 6% better performance for two-processor servers at a 29% lower price point
• 24% better performance for four-processor servers at a 21% lower price point

Some more AMD vs. Intel comparisons:

When it comes to getting these processors out, AMD has partnered with big name OEMs such as IBM, Dell, HP, Sun and Cray to make sure that there are plenty of “Shanghai” Opteron-ready platforms.

What about price? AMD have priced the “Shanghai” Opteron processors to offer excellent performance-per-dollar.

Resource - ZDNet

Google demanding Intel's hottest chips?

When purchasing server processors directly from Intel, Google has insisted on a guarantee that the chips can operate at temperatures five degrees centigrade higher than their standard qualification, according to a former Google employee. This allowed the search giant to maintain higher temperatures within its data centers, the ex-employee says, and save millions of dollars each year in cooling costs.

If the chips failed prematurely at these higher temperatures, the former Googler says, Intel was obliged to replace them at no extra charge.

Intel denies this was ever the case. "This is NOT true," a company spokesman said in an email. Google declined to comment on its relationship with Intel. "Google invests heavily in technical facilities and has dozens of facilities around the world with many computers," reads a statement from the company. "However, we don't disclose details about our infrastructure or supplier relationships."

The ex-Google employee learned of this Intel pact a little more than a year ago, during a Google "Tech Talk" open to anyone at the company. The talk was given by a Google thermal dynamics engineer, part of a small team - perhaps no larger than two people - that oversees heat issues inside the company's data centers.

According to the same ex-employee, it is now the norm for Google to construct its data centers by piecing together intermodal shipping containers pre-packed with servers and cooling equipment. In 2003, Google filed for a patent on this sort of modular data center, and the patent was granted last October.

This modular setup - known internally as Project Will Power - allows Google to construct the building blocks for each data center at a central location and then ship them around the world as needed. Robert X Cringely first leaked word of Google's modular data center work in 2005, claiming the project began after representatives of the Internet Archive pitched the idea to Google co-founder Larry Page.

Heat Saves

It's no secret that Google builds its own data centers - and many of the servers within. It's long been said that after the top server vendors, the ad broker consumes more processors and hard disks than anyone else in the world - and Intel freely acknowledges that it provides at least some of the chips.

Intel also customizes power saving motherboards for Google - a service not afforded other customers. But our source tells us that any motherboard pact is separate from the companies' high temperature chip agreement.

If Google can leverage an extra five degrees centigrade from Intel, it can save a few penguins - and some serious cash. According to Data Center Knowledge, Google runs at least 36 data centers across the globe, and several more are under construction. One of the newer sites, in The Dalles, Oregon, includes three data center buildings, each measuring 68,680 square feet.

According to Mark Monroe, Sun Microsystems' director of sustainable computing, data center managers can save 4 per cent in energy costs for every degree (Fahrenheit) they raise the thermostat. If you run your data center at a higher temp, you use less power and spend less money on the equipment needed to cool it down.

But if you raise the thermostat, you may void your hardware warranties. "The hardware manufacturer will usually have a temperature range," says Rich Miller, editor of Data Center Knowledge. "If the equipment is running outside of the range, the manufacturer might be inclined to challenge any sort of returns or credits. That's the thinking across the industry."

Resource - The Register