A Multi-Core Future

As the Threading for Multi-Core community manager I've been reading a lot of technical papers and blog posts about the hype and reality of the Multi-Core revolution. I'm living the Multi-Core life and working to promote it globally so that developers are building applications that run best on multi-core systems.

As I commented on in my last post, it does not appear that everyone is developing applications that take advantage of multi-core systems. Mark Nelson wrote a great analysis of the current state of the multi-core revolutions. He has some interesting views of Intel and AMDs approached to scale from dual-cores to 128-core machines. I agree that a lot of work needs to be done on the software side to deliver the gains that this hardware could deliver. It is up to the developer community to sort out the hype from reality. I hope that I can provide trustworthy support in this area and do what I can to answer your questions (or find the experts who can).

To add another point of view I suggest that you check out Michael Suess's Thinking Parallel blog that pretty much goes point by point through Mark's original thesis. I agree with both of them that the main stream operating system companies are provide a stable framework for the application to take advantage of threading on multiple core systems.

Please take a moment to review both articles and provide your thoughts.

AMD Promotes Vista Design

Advanced Micro Devices debuted a program Tuesday aimed at convincing other hardware makers to develop PCs and components that will give consumers and business users a better ride on Windows Vista than its rival Intel.

AMD’s Better by Design program will provide a kind of Good Housekeeping seal of approval on systems that meet AMD’s criteria for the kind of performance needed to run the processor-intensive Vista operating system.
Vista

As Microsoft prepares to release the consumer and small business versions of Vista at the end of the month, many users have been wary of upgrading, worrying that their computers may not be powerful enough to run the enhanced graphics and other features.
Vista

Others are concerned about bugs, incompatibilities, and security vulnerabilities that have already been uncovered.

Better by Design is the second branding program that AMD has unveiled in two days at the International Consumer Electronics Show in Las Vegas.
Las Vegas

On Monday, the Sunnyvale, California-based chip maker launched its AMD Live branding program in an attempt to compete with Intel’s Viiv consumer electronics technology and its Intel Inside logo (see AMD Intros ‘Live’ Brand Devices).

AMD launches new chips in Beijing

US microprocessor maker AMD launched its latest quad-core processors in Beijing yesterday, the first stop of its global launches.

The quad-core Opteron processor claims to be the world's most advanced computer chip with a performance 50 percent higher than the current dual-core products.

Its arch-rival Intel launched a similar quad-core product last week, but AMD claims its product is the only one with four cores in the same semiconductor die.

"The selection of China as the first location to release the chip, Barcelona, proves the importance of the China market to AMD and also shows our development in the country has entered a new stage". The supercomputers, whose speed can put them among the four fastest computers in the world, are likely to make China the second country with such high-performance computers, which are widely used in complicated projects like weather forecasts and oil exploration.

AMD vs. Intel: The Challenger's New Plan

AMD Module Shift Manager Peter John presents a 200mm wafer in Dresden, Germany. Norbert Millauer/AFP/Getty Images
Chipmaker Advanced Micro Devices says it’s about to go through a major change concerning how it makes chips, but it hasn’t said exactly what that change will be. Speculating about it has become a great guessing game among Wall Street investors and Silicon Valley’s chattering classes.
AMD (AMD) is suffering through one of the toughest stretches in its history, racking up losses of $1.6 billion on sales of $2.8 billion so far this year. It’s now struggling with a nagging question: how to continue making chips for personal computers and servers that can compete with those of rival Intel (INTC) without having to bear the heavy expenses required to operate its chip factories known as “fabs.”
AMD has said it’s pursuing a new strategy it calls “asset smart,” aimed at saving money while at the same time preserving its manufacturing muscle. Hector Ruiz, CEO up until he stepped down on July 17 (BusinessWeek.com, 7/17/08), says that in his new position as executive chairman he’ll focus his attention on completing the transformation. Many chip industry analysts believe the outcome will either save the company or doom it.
At Least Some Outsourcing
The company has managed to keep a tight lid on its plans, unusual for the gossipy chip industry. The company declined to comment on its plans to BusinessWeek, beyond issuing a brief statement: “AMD continues to look at multiple options that leverage our world-class manufacturing capabilities and relationships to achieve an optimum blend of internal and external operations.”
The phrase “optimal blend” is important, because it suggests that AMD is going to outsource at least some of its existing manufacturing operations. The backbone of its manufacturing operations are two fabs in Dresden, Germany, and all the chipmaking equipment in them. It also has two large test and assembly plants in Malaysia and Singapore, and a smaller one in China. One popular theory has AMD turning to a third-party chip foundry company like Chartered Semiconductor Manufacturing (CHRT) of Singapore to step in and operate the fabs under contract. Chartered already makes some of AMD’s chips under contract to help AMD keep up with demand surges. Taiwan Semiconductor Manufacturing (TSM) handles manufacturing for AMD’s graphics chip unit ATI.
Selling off fabs would have clear financial benefits for AMD, but would also expose some problems. For one thing it would go a long way toward cutting AMD’s operational expenses, which were $1.6 billion in the first six months of the fiscal year, eclipsing its $1.3 billion gross margin. Unloading the fabs would put AMD in a class of chip companies who don’t own their own factories, and thus hire companies like Chartered, TSM, United Microelectronics (UMC), and even IBM (IBM) to make chips under contract. Going “fabless” has meant success for chip companies as varied as Nvidia (NVDA), Broadcom (BRCM) and Qualcomm (QCOM). Still, such a move would fly in the face of AMD tradition: Its founder, Jerry Sanders, once famously quipped “Real men have fabs.”
Maybe, but there are pitfalls to owning fabs. They’re expensive to build—$3 billion to $5 billion each—and expensive to maintain. Competing with Intel means upgrading a fab with billions’ worth of new manufacturing equipment every two years or so. There are also costs to holding chip inventory, which, given the volatile nature of the PC market (BusinessWeek.com, 5/30/07), has caused AMD to occasionally be left holding a lot of costly chips that are hard to unload.
Intel Deal Could Stand in the Way
Unloading the fabs to a foundry like Chartered might solve some problems, but it would create others. One important one: AMD would lose control of the up-close, day-to-day management of its manufacturing processes, which have over the years been a strategic advantage. “They know how to run a fab, and they do it well, and for the most part that’s been a big help,” says analyst Nathan Brookwood, head of Insight64, a Silicon Valley consultancy.
Going fabless might also run afoul of an important cross-licensing agreement between AMD and Intel. The deal, which is a leftover from their legal battles in the 1980s and was renewed most recently in 2001, expires on Jan. 1, 2010. Analyst Tim Luke of Lehman Brothers (LEH), in a research note issued July 23, says the deal likely limits the extent to which AMD could pursue a fabless arrangement, and that if AMD were to bring in an outside partner to run its fabs, it would likely have to retain ownership of at least 50%.
So is there a third alternative for AMD, something other than holding on to its fabs or selling them outright? Perhaps. Luke suggests splitting AMD into two companies, one devoted to design, the other to manufacturing. “A separate foundry function might enable it to compete with Intel without the capital required to keep up from a manufacturing standpoint,” and manufacturing could be funded by an investment from a partner, probably Chartered.
There’s another possible in-between path. Brookwood thinks AMD will sell the fabs and the equipment in them to a third party, and then lease them back. Such a move would allow AMD to retain full control of the fabs and operate them, yet greatly reduce the operational expenses. AMD would essentially be paying rent on the fabs and the equipment in them to a landlord.
Regulatory Approval
So who would that landlord be? Brookwood thinks a leading candidate might be a sovereign wealth fund like Mubadala Development, the Abu Dhabi-based fund that paid $622 million for a stake in AMD last November. At the time, the fund paid $12.70 a share for 49 million shares, an 8% stake. AMD’s stock price has since dropped to less than half that, closing Aug. 8 at $5.13. “By taking ownership of the fabs, this fund might go a long way toward seeing its investment pay off,” Brookwood says.
Such a deal would likely require the approval of government regulators in Germany, the European Union, and probably the U.S. That might explain why it’s taking so long for Ruiz and new CEO Dirk Meyer to announce the plan. But would it pass muster with Intel under their cross-licensing agreement? After all, selling the fabs might arguably be considered the kind of change in control that gives Intel the right to cancel the deal.
But such a move is unlikely. Intel in recent months has been dogged by antitrust investigations in the EU, Japan, and the U.S. An unusual anti-AMD move by Intel may hurt the company’s ability to argue that it’s not a monopolistic bully.
So it looks like a major strategic shift is ahead for AMD. With lower operational expenses, the company would be able to concentrate on designing and selling chips. The company has proven that at times in the past it has been able to go head to head with the fearsome Intel. Even now it has a number of promising chips coming out in the near future, including one code-named Shanghai for servers and another code-named Puma for notebooks.
Unloading the financial burden of its fabs may let AMD do even more in the future. “There are some real bright spots in their product line,” says analyst Dean McCarron of Mercury Research. “They do look like they are starting to turn a corner.”

Is Intel scared of Nehalem?

Intel has officially acknowledged the existence of Nehalem, its next-generation microarchitecture, during several keynotes during this week’s Intel Developer Forum Fall 2008. It’s even been productised to Core i7, so it’s real, very real.
However, as much time as Intel has put into disclosing what makes this particular core such an engineering leap forward over the present generation, hard-and-fast performance numbers have been difficult to come by - kind of surprising for a chip that launches in the next few months, right?
One would expect a class-leading architecture which will become the backbone of Intel’s server, desktop and mobile parts for at least the next two years, to be heralded with the usual glut of ‘oh-my-lord-it’s-fast’ benchmarks, suitably skewed (ahem, normalised) to AMD’s fastest, thereby showing the delta that, Intel believes, exists between the semiconductor rivals’ best.
So why no angelic trumpets and red-carpet treatment for a design that, on paper, takes present Core 2 (Penryn) to the cleaners in a number of memory-bandwidth and heavily-threaded instances?
Is it because of AMD?
Could it be that AMD’s announcement that it will pull its next processor update, Shanghai, forward to Q4 2008 has Intel execs quaking in their expensive suits? We don’t think so, because Shanghai’s performance improvements over current-generation Barcelona are generally known, and it’ll struggle to add more than 20 per cent extra oomph when evaluated on a clock-for-clock basis against Phenom X4. Knowing this, Shanghai will probably perform somewhat akin to Intel’s Penryn.
Could the lack of numbers have something to do with Nehalem’s performance not being quite up to scratch? That seems highly unlikely, especially if our Nehalem performance preview is accurate, and we have no reason to doubt that our 2D numbers will stack up against retail samples.
It’s a question of economics, we reckon
Ultimately, we reckon that Nehalem performance has been deliberately kept under wraps by the powers that be. Why? Because letting a full suite of numbers out for public consumption, which has been Intel’s method of disseminating its engineering excellence since first-generation Core microarchitecture hit AMD Athlon in the nuts, inextricably dampens - nay, crushes - sales of present-generation parts.
As a consumer or business, why would you buy a Core 2-based system when something potentially better, lots better, is just around the corner? - a product that will require a new motherboard and, potentially, new memory - kerching! Knowing just how much of a whipping Nehalem can potentially hand to Penryn in heavily-threaded scenarios, Intel would be driving potential customers away from mid-to-high-end sales of a chip that’s been yielding well for some time. Cutting off your nose to spite your face comes to mind.
Intel is scared of Nehalem, insofar as its prodigious performance makes Intel’s current line-up look, well, a little tardy by comparison, and why tell people that when there are millions of Core 2-equippe d machines waiting to be sold at the likes of PC World and Best Buy? Why spend $400 on a chip, or $2,000 on a system, now when the same money will buy you so much more performance in just three months?
Of course, after so little sleep during IDF, this particular hack could well be off his rocker. We’ll let you decide.

AMD Launches ATI Radeon HD 4600 Series Graphics Cards

AMD has introduced ATI Radeon HD 4600 series graphics offerings which, according to the company, deliver exceptional gaming and HD multimedia performance while consuming less power than a standard light bulb. The series comprises -- ATI Radeon HD 4670 and ATI Radeon HD 4650 graphics cards -- and leverages the technology found in the ATI Radeon HD 4800 series, including support for the latest DirectX 10.1 games, HD multimedia capabilities and leading performance-per-watt.
Based on second generation 55nm process, the ATI Radeon HD 4670 and ATI Radeon HD 4650 graphics cards feature TeraScale graphics engine, with 320 stream processing cores. The ATI Radeon HD 4600 series incorporates AMD's Unified Video Decoder (UVD 2.0) for advanced hardware HD video processing which ensures a smooth HD viewing experience on HD-capable monitors. In addition, it features support for HDMI 7.1 surround sound audio; ATI Avivo HD technology for sharp and crisp images; and ATI PowerPlay technology which allows for automatic power adjustments based on the activity of the graphics processing unit (GPU).

The ATI Radeon HD 4670 operating at under 60 watts and the ATI Radeon HD 4650 operating under 50 watts under full load -- eliminates the need for additional power connectors.

The ATI Radeon HD 4670 is available with a frame buffer of 512MB GDDR3 memory with an MSRP of Rs 5,000 with a 1GB DDR3-based variant is scheduled to ship later this month. The ATI Radeon HD 4650 features a frame buffer of 512MB GDDR2 memory and is expected to be available later in the month at an MSRP of Rs 4,250.

Post-production houses test Intel Core2 Quad technology

Both houses felt almost satisfied with their current processing power and performance, but felt that if time and energy could be saved it could be greatly beneficial to them.

Running a successful business today, whatever field it may be, requires a high degree of efficient technology and equipment, especially when dealing with a highly competitive environment. In the field of production and post-production especially, their efficiency relies heavily on powerful computers to help them in creating and editing work as quickly as possible, until it is flawless. Many post-production houses have come to realize that investing in computers that can take on this difficult task is indeed the path to competitive performance.

The challenge
Digital Media Cairo (DMC) and FX Productions both tested the Intel® Core™2 Quad processor for ten days and kept a log of performance results comparing to their usual results. Testing focused on 2 of the most common challenges that they face today: time management and speed, and equipment over-work. With the long hours spent editing heavy images, the computers were likely to over-heat and even need to be shut down before being able to continue.

Both DMC and FX Productions relied on Mac Pro computers to conduct their business and claimed that so far the computers never gave them too many set-backs, other than the usual heat-up or occasional shut-down. The computer processors they were using included Intel Core Duo processors and Intel Xeon processors. Both post-production houses commented on the fact that they felt their processors were up to speed and allowed them to work efficiently and at the pace they desired. The challenge was to show them that they could be working at a much faster pace, in return being able to take on more work.

Tests results
Assessment of the Intel® Core™2 Quad processor showed a definite improvement in computer functioning, equaling in an improvement in the pace that work is done and therefore the amount completed. Both production houses used applications such as Mac Final Cut Pro, Soundtrack 2, and Motion 3, in addition to other applications in Final Cut Studio.

Rendering time for certain animated frames took an average of an hour and 45 minutes using the post-production houses' usual processors. But with the Core™2 Quad processor, rendering time was decreased to about 45 minutes to an hour, almost half the time. In editing other visual media such as videos or documentaries, rendering time was about 20-30 minutes comparing to a much lower 12-18 minutes using the Core™2 Quad processor. Encoding of images was also reduced to almost half the time when using the Core™2 Quad. Employees who tested the Core™2 Quad processor also mentioned that the computers didn't get hot, a good indication of the thermal performance of Intel's white box machines, used on the shelf components and chassis.

Overall test results show that the Intel® Core™2 Quad processor was faster and more powerful than previously used processors, that it saved energy (decreasing heat), and that employees in the post-production field would recommend that houses invest in Core™2 Quad processors. Both houses mentioned that they never realized that they could have better functioning computers until they tested the Core™2 Quad.

Customer quote
'The Intel® Core™2 Quad processor is very beneficial because it saves you time and makes you more relaxed while you are working. It also makes a huge difference when the computer doesn't heat up, especially because our rooms are usually quite hot because of all the computers and equipment inside.' - Waleed Abdel Moneim, Head of Editing, DMC.

'Once you see the difference and the benefits, you are much more likely, as a post-production house, to go out and buy some Core™2 Quad computers. Our Mac computers all had Intel Core Duo processors which were already very good. The Core™2 Quad of course was only better. Very fast, very efficient, and kept up with all the different things we were doing while editing. We never realized how much computing power we were missing until we used Intel's Core™2 Quad.' - Ahmed Niazy, Editor, FX Productions.