Business Tech

It's a question we all face: with chips getting more processing cores instead of more gigahertz, is your next computer going to actually run your software faster?

Microsoft is one of the companies that feels the pressure to most acutely when it comes to putting those cores to work. Though it doesn't pretend to have the problem licked, Microsoft does believe Windows 7 provides a better foundation for using multicore systems than earlier versions of the operating system.

Jon DeVaan, head of Windows Core Operating System Division

(Credit: Microsoft)

One key part of solving the PC's multicore problems draws from the world of big iron, and Windows 7 can support much bigger iron--servers with as many as 256 processor cores compared with 64 for its predecessor. Now a few years into the multicore era, even today's laptops are able to juggle as many tasks as reasonably powerful servers from just a few years ago. Intel's new Core i7 "Clarksfield" processor for mobile computers has four cores that manage a total of eight separate "threads" of work.

"One dimension is support for a much larger number of processors and getting good linear scaling on that change from 64 to 256 processors," said Jon DeVaan, senior vice president of Microsoft's Windows Core Operating System Division. "There's all kinds of depth in that change."

Linear scaling means that doubling the number of processors means a doubling in performance--something rarely achieved in real-world computing. But what does 256 or even 64 processors have to do with a PC with four or eight cores? In short, updating the Windows plumbing to support bigger servers also helps work run more smoothly on smaller multicore machines, for example by ensuring data cached in memory is close on hand to the processor core that needs it, DeVaan said.

It's crucial that Microsoft help solve multicore issues. The company is responsible not just for the most widely used personal-computer operating system but also for the programming tools many use to create the software that runs on it. That's why another broad attempt to ease multicore pains takes place within Visual Studio 2010, the upcoming version of Microsoft's programming tools.

"People have been working on this for a long time. So far there haven't been any magic bullets," Devaan said. "The commercial reality is creating a lot more urgency now, so I think we'll see a lot more approaches taken."

Unlocking multicore power is a point of competition, too: Apple's newest version of Mac OS X, Snow Leopard, adds a facility called Grand Central Dispatch to centralize management of all the various threads of programs as they run on a system.

Intel and Advanced Micro Devices bear responsibility, too, since they embraced multicore designs once heat problems put an end to the clock-frequency race, but Microsoft has much more clout in developer relations.

Windows 7 is due to ship October 22.

(Credit: Microsoft)

Multicore designs can help easily when people are running many separate programs or when running programs that are "embarrassingly parallel"--in other words, when a task has many naturally independent subtasks, such as rendering each of a video's many frames. But many programs won't easily make the jump to a parallel design when they're set up as a single sequence of steps today.

"An operating system is never going to be able to take an application that isn't already parallel and make it so. Developers still need to multi-thread their apps," said Evans Data analyst Janel Garvin.

Visual Studio 2010
So it's good Microsoft is working on parallel programming aids within Visual Studio.

"Microsoft has done surprisingly little until recently to help developers write parallel applications, except for their alliance with Intel to promote Parallel Studio," an Intel collection of programming tools for parallel programming, Garvin said. "However, in the last year they've made some announcements and promises for Visual Studio 2010 about enhanced tools for parallel programming. It's likely that the success of Parallel Studio has impressed upon them the importance of providing Windows developers with the tools they need to remain competitive going into the future when manycore will be the standard."

Eventually, programmers will have to embrace parallel programming to be competitive, Garvin said. Parallel Studio helped bring the concepts to a much more mainstream audience, she said, and Evans Data's spring 2009 global developer survey found 40 percent of programmers are working on multithreaded applications today and another 15 percent plan to in the next year.

"Parallel programming is complex, difficult and labor-intensive, for even the most skilled developers, which has led developers to avoid writing parallel programs, leaving many CPU cycles unused," according to Steve Teixeira, Microsoft's principal product unit manager of parallel computing. The company's attempt to improve the situation comes not just in Visual Studio 2010 but also in another future product, version 4 of the company's .Net Development Framework.

Parallel programming tools
Among those features:

• The Task Parallel Library, which lets .Net programmers write more parallel code in familiar terms. For example, programmers are used to "for loops" that repeat a particular task a specific number of times; library lets each step of the loop happen simultaneously instead of sequentially.

The new Intel Core i7 processor for mobile computers has four cores and can run eight threads.

(Credit: Intel)

• The Microsoft Concurrency Runtime can provides a shared resource for scheduling tasks and allocating resources--and which works better on Windows 7.

• The Asynchronous Agents Library can permit separate threads of execution to pass messages among each other. That's useful in cases where separate threads need to head off no-no conditions such as when

• Parallel Language Integrated Query (PLINQ) technology lets programmers perform some operations with data in parallel rather than sequentially.

• The Parallel Pattern Library is designed to make parallel programming easier for those using the C++ language.

Microsoft knows none of this is truly easy, though. DeVaan wonders about cases when existing software is being parallelized--is each step in a parallel for loop really independent of the others? He sees "a lot of hand-waving" around the computing industry that glosses over the true difficulties.

"As an industry, we're going to be working hard to make it work better and working with broad set of developers to target (multicore programming) without undue work," DeVaan said. "Will these approaches really accomplish it? That's an open question."

Cambridge, England-based chip company ARM on Wednesday announced the development of dual-core, quad-core, and eight-core Cortex A9 processor designs, explicitly aimed at markets currently served by Intel's x86 chips and IBM's PowerPC.

"This is a huge departure from what we've done in the past", Eric Schorn, vice president of marketing for ARM's processor division, told ZDNet UK. "We really wanted to take off the handcuffs and see what could be done with performance, performance, performance."

The new designs, available in two variants optimized for low power consumption or high performance, are intended for use by companies building their own chips. ARM claims that the new processors, which can run at up to 2GHz, are up to eight times more efficient than Intel's low-power chips in terms of performance per watt, with the high-performance part running at five times the throughput of Intel's Atom chip for similar power levels.

The low-power part delivers twice the performance at a quarter the power, according to the company's published benchmarks.

"The sweet spot for most customers is dual-core," said Schorn, "but the base design can go up to quad-core and some partners are already building those. Eight way is coming. Everyone's high-end road map is putting down more cores, and we do that. We're headed in the direction of Intel's mainstream processors. We have other plans that surpass the current performance, and we'll intercept Intel in a high-margin area, not just with Atom."

The dual-core parts, which are designed to be made using fabrication company TSMC's 40-nanometer chip manufacturing process, can be licensed now with delivery of the finished designs to partners in the fourth quarter of 2009. ARM itself will be making evaluation chips available to partners and software designers in the first quarter of 2010.

To date, ARM has mostly partnered with companies making components for wireless, consumer, and automotive equipment. However, this new design will see new enterprise partners coming on board. In particular, Schorn said that the high-performance multicore ARM approach would open up parts of the market currently dominated by companies with large proprietary design teams -- "blowing the doors off that by offering freely available IP," as he put it.

"Enterprise is a key opportunity," continued Schorn. "Our existing partners are executing extremely well in their existing markets. We have a new license signed, with a number in the pipeline, and enterprise is well represented. The design is applicable to all sorts of servers, is cache-coherent so can do SMP, and will be wonderful for Linux, Apache and other parts of the enterprise stack".

Earlier this week, ARM announced that it had joined the Linux Foundation.

ARM also intends the multicore Cortex A9 design to be used in consumer equipment. "If you look at the high end of embedded systems, Netbooks and the like, there's not much innovation relative to the mobile phone area. We want to take the rate of change of mobile phone design and innovation into other areas. Consumers will see a lot more diversity at a lot faster pace," said Schorn.

The company says it gets its claimed level of performance at low power by having very finely tuned control over the different areas of the chip, with seven power zones able to turn off parts of the cache, maths, media, and general-processing areas automatically when idle.

Schorn said that he was not concerned by ARM's lack of Windows 7 compatibility. "We don't have a Big Windows announcement to make. We do have staff at Redmond, and we'll see what the future will hold. Talking about Windows is the wrong way of looking at it.

"If you look at what's happening with Web-centric, internet-oriented demographics and things like Java virtual machines, just-in-time compilers, widgets and so on, it's not architecturally dependent. Look at Samsung with Yahoo widgets in its televisions. I don't see the need for Big Windows on your television."

Rupert Goodwins of ZDNet UK reported from London.

Apple began shipping Snow Leopard on Friday, but the true importance of the Mac OS X update likely will emerge well afterward.

That's because Mac OS X 10.6 begins a longer-term Apple attempt to get ahead by cracking a problem facing the entire computer industry: squeezing useful work out of modern processors. Instead of stuffing Snow Leopard with immediately obvious new features, Apple is trying to adjust to the new reality in which processors can do many jobs simultaneously rather than one job fast.

"We're trying to set a foundation for the future," said Wiley Hodges, director of Mac OS X marketing.

Apple shed some light on its project, called Grand Central Dispatch, at its Worldwide Developer Conference in June, but most real detail was shared only in with programmers sworn to secrecy. Now the company has begun talking more publicly about it and other deeper projects to take advantage of graphics chips and Intel's 64-bit processors.

The moves align Apple better with changes in computing. For years, chipmakers such as Intel and Advanced Micro Devices had steadily increased the clock rate of their processors, and programmers got accustomed to a performance boost with each new generation. But earlier this decade, problems derailed the gigahertz train.

Intel and Microsoft will hold an event next week to discuss collaboration on improvements to Windows 7.

The event, on September 1 in San Francisco, will "share how the two companies collaborated on key enhancements during the development of Windows 7," according to Intel. Steve Smith, vice president and director, Intel's Digital Enterprise Group Operations, and Michael Angiulo, general manager of Windows Planning and PC Ecosystem at Microsoft, will talk at the event. Microsoft plans to launch Windows 7 on October 22.

Windows 7 collaboration will be demonstrated by engineers from both companies, according to Intel. Not surprisingly, Microsoft is working closely with Intel, whose chips will power the vast majority of PCs running Windows 7.

In a blog posted in July, Intel described how Microsoft and Intel "saw unique opportunities to optimize Windows 7 for Intel processor technology" in the areas of performance, power management, and graphics.

The blog discusses improvements to multitasking based on "SMT Parking," which provides additional support to the Windows 7 scheduler for Intel Hyper-threading Technology. With Hyper-threading, the operating system sees a single processor core as two cores (i.e., a dual-core chip becomes a virtual quad-core processor), thus potentially improving multitasking--or doing tasks (threads) simultaneously.

In addition, improvements over Vista for boot and shutdown times have been implemented during the Windows 7 development cycle, according to the blog.

And on Intel's Web site, the chipmaker lists desktop motherboards and associated drivers that have passed logo certification for Windows 7.

Another beneficiary of improved Windows 7 technology: Intel solid-state drives, which are typically faster than hard-disk drives and gaining ground in niche markets such as high-end laptops, gaming PCs, and servers. SSDs will be able to take advantage of Windows 7 technology called the Trim Command. Trim will allow blocks of data to be freed up for reuse to better maintain the performance of the SSD.

Windows 7 will also do more than previous operating systems with graphics via DirectX 11. Advanced Micro Devices has described DirectX 11-related technology that enables games developers to create smoother, less blocky and more organic looking objects in games. And, beyond games, Windows 7 has the potential to turn a graphics processing unit (GPU) from AMD or Nvidia into a general-purpose compute engine, used to accelerate everyday computing tasks like a central processing unit, or CPU. Specifically, "the compute shader" can be used to speed up more common computing tasks. The buzz word used to describe this technology is a mouthful: GPGPU or general-purpose graphics processing unit.

Intel has quietly snapped up two software companies in the last 30 days with aim of boosting development of applications that take better advantage of chips with more than one processing core.

In a company blog, the chipmaker indicated the acquisition of Cilk at the end of last month and then Rapidmind earlier this week. Both are small companies that employ under 50 people. The acquisitions follow the purchase of software company Wind River Systems in June.

"Over the last few years, there has been a gradual emergence of multicore microprocessors. It's put parallelism in more and more machines," James Reinders, chief evangelist and director of marketing and sales at Intel, said in a phone interview Friday, explaining why Intel bought the two firms.

"If you look at traditional applications, ones that we use everyday, it's fair to say that most are not exploiting parallelism--at least not to the full extent," Reinders said.

A multicore processor is defined as any chip with more than one processing core. Today, almost all Intel chips that go into laptops, desktops, and servers have at least two cores. The challenge for Intel is to make sure that applications take advantage of all the cores--so-called parallelism. This has historically presented a challenge for software programmers.

"The operating system does stuff for applications in parallel," Reinders said, referring to operating systems such as Windows. "But considering that we can produce more and more cores every year, to truly get the benefit of what the future holds, applications need to change. And most applications haven't changed," he said.

The goal is to facilitate the development of parallel programming. "How do we help software developers tackle parallel programming? Both companies had teams of experts that had been focused on this problem. So, they're kindred spirits," he said.

Writing about Cilk in a blog, Reinders said Intel sees "great opportunities for Cilk to integrate with our parallel tools...including Intel Parallel Studio." The firm's technology enables "mainstream programmers to develop multithreaded (or parallel) applications...Providing a smooth path to multicore for legacy (older) applications that otherwise cannot easily leverage the performance capabilities of multicore processors," according to Cilk's Web site. Original Cilk research was done at MIT.

Rapidmind was founded five years ago as Serious Hack and grew out of work at the University of Waterloo. It boasts advanced technology for helping software developers with data parallel programming for multicore processors and accelerators.

The cost of the two acquisitions was not disclosed.

Earlier this month, I encountered an Adobe Photoshop Lightroom analysis by consultant Lloyd Chambers that expressed surprise with a facet of the image editing and cataloging software: it didn't export photos as fast as possible.

Chambers found that if a photographer wants to produce JPEG or TIF images from the originals in the program, the fastest way is to divide the batch into thirds and export each third separately. Using a modern Mac Pro system, exporting a test set of photos took 351 seconds as one batch and 189 seconds divided into three batches running at the same time.

"The big disappointment is the sluggish performance importing and exporting files, which are tasks that are key to efficient workflow--tasks one has to do over and over. Most of the 'juice' of a Mac Pro goes untapped," Chambers concluded. "You have to load it up with more than one job to force more of the available CPU cores to be used. Lightroom should do this automatically!"

The study caught the attention of others, including Scott Kelby, head of the National Association of Photoshop Professionals. I was intrigued, too, because although many programming chores are difficult to spread across multiple processor cores, exporting photos is trivially easy since it breaks conveniently into independent bite-sized pieces. So I thought I'd see what Adobe had to say for itself.

This story was corrected. See below for details.

SAN FRANCISCO--Apple wants Mac OS X to do a better job dealing with the new directions that Moore's Law has taken computer chips.

At its Apple Worldwide Developer Conference here, Bertrand Serlet, senior vice president of software engineering, shed light on technology called Grand Central Dispatch that's designed to make Mac OS X 10.6, called Snow Leopard, take better advantage of multicore processors and graphics processors.

Power play at Apple's WWDC 2009.

(Credit: James Martin/CNET)

Computer chips for years improved in performance through faster clock speeds, but processor engineers ran into problems with chips consuming inordinate amounts of power and producing inordinate amounts of heat. In addition, the faster clock speeds sometimes meant chips just idled faster because memory access speeds couldn't keep up.

The new direction: multicore processors that put multiple processing engines on the same chip. The problem with the approach, though, is that PC software typically had been written to run with one thread of instructions at a time. Multicore processors work best when software does many things at the same time, which is much harder to program.

Grand Central Dispatch is designed to address that problem for software developers, making it easier to program multithreaded software, use operating system services, and tune program executions.

It also improves how Mac manages those threads, Serlet said. For example, when running Apple's Mail app, today's Leopard OS uses about the number of threads when busy as when idle.

"When it's busy, it uses more threads to take advantage of multicores. When idle, all those threads go away, giving back resources to the system," Serlet said. "When you apply that to every application, you get a big win in performance and responsiveness."

Graphics chip power
The new Mac OS X also is designed to support a programming technology called GPGPU--general-purpose graphics processing unit--which lets a graphics chip run some computing jobs in addition to its ordinary job displaying graphics.

To make its GPGPU technology work, Apple uses OpenCL, a C-like programming technology that has the support of graphics chipmakers Nvidia, AMD's ATI, Intel, and others.

Graphics chips aren't good for every sort of computing task, but they are good for mathematical calculations--including they physics calculations often needed in video games that simulate flowing fabrics, bouncing balls, and other real-world actions.

Mac OS X will be available in September with an upgrade price of $29, a big notch less expensive than the $129 price of earlier upgrades.

Apple also is working to support 64-bit x86 processors, now the prevailing standard. One big advantage of 64-bit processors is support for more than 4GB of memory; Serlet also touted faster mathematical processing such as the doubled speed of fast Fourier transforms.

Apple has been gradually making its operating system fully 64-bit. "Snow Leopard is final stage where all the major system applications are written in 64-bit mode," Serlet said.

Correction 6:13 a.m. PDT Tuesday: This story misidentified the speaker. It was Bertrand Serlet, senior vice president of software engineering.

Intel is back, pitching its processors for gaming graphics.

The chipmaker will attempt to promote its silicon for sophisticated game effects at the upcoming Game Developers Conference in March, as it strives to make a case for quad-core processors in lieu of graphics chips from Nvidia and Advanced Micro Devices.

The pitch goes like this: "Learn how to easily add real-time 3D smoke, fog and other fluid simulations to your game without using up the GPU." That's according to an Intel Web page entitled Intel at Game Developers Conference. (The CPU is the central processing unit, or main brains of a computer; GPU stands for graphics processing unit.)

The session abstract goes on to say that the "source code to a fluid simulator optimized for multi-core CPUs...can easily be integrated by game developers into their engines to produce unique 3D effects."

Intel's argument raises the question, how should the CPU and GPU divvy up their tasks? In games, the CPU can handle things like physics and AI (artificial intelligence), and certain older games actually run some graphics on the CPU. Generally, however, the GPU is much more efficient (that is, faster) at handling most of the high-end effects that the gamer sees on the screen.

But there are exceptions. "Not all algorithms and processes map well to a GPU," said Jon Peddie, president of Jon Peddie Research. "You have to have a problem that is naturally parallel, and except for the rendering of, say, a water surface and subsurface and reflections, the wave motion equations will run just fine on a CPU," Peddie said.

Intel may also be seeking ways to make better use of its quad-core processors, according to Tom R. Halfhill, an analyst at the Microprocessor Report. But, he added: "I need to be convinced that a CPU can do those 3D effects better than a GPU can."

Then, there's also the Larrabee factor. Larrabee is an upcoming high-end graphics processor due late this year. "I'm sure some of it may also relate to Larrabee, which will include x86 cores, if or when it comes to market," said Jim McGregor, an analyst at In-Stat.

(This Mythbusters demonstration at an Nvidia conference is oversimplified and self-serving but it crystallizes the difference between CPUs and GPUs.)

In another GDC session, Intel is also pushing the CPU for physics and AI: "How can your game have more accurate physics, smarter AI, more particles, and/or a faster frame-rate? By threading your game's engine to take advantage of multi-core processors. Intel has built a threaded game engine and demo called 'Smoke' that shows one way of achieving this goal," the abstract states.

It continues: "This presentation examines the Smoke architecture and how it is designed to take advantage of all CPU cores available within a system. It does this by executing different functional and data blocks in parallel to utilize all available cores."

Intel won't stop there. It will also focus on the bane of many PC game developers: gaming on Intel integrated graphics silicon--a relatively low-performance platform that prohibits game titles from being displayed in all their glory at higher resolutions. The session will focus on "programming for scalable graphics applications" and cover "performance considerations when programming for integrated graphics in general with specific tips for Intel Integrated graphics."

Gartner sounded a warning on Wednesday about the impact on software of the rapid growth of multicore chips and the number of threads each processor can handle.

In a research note, the analysts argued that software is struggling to keep pace with the fast growth of multicore processors, first from two and four cores per processor, and now to eight and even 32 cores in high-end servers. With 32 processors per socket already shipping, four years from now machines could host 1,024 processors, Gartner said.

Analyst Carl Claunch said: "Many of the software configurations in use today will be challenged to support the hardware configurations possible, and those will be accelerating in the future."

Running advanced multicore machines with today's software is like "putting a Ferrari engine in a go-cart," he said.

Part of the problem, Claunch said, is the speed of innovation in today's chip design. Chips develop with more cores and each core gets more threads as well, adding to the issue, he said. Each generation turns the same number of sockets into twice as many processors.

Claunch said the software that runs today's servers has both hard and soft limits on the number of processors. Part of the problem was that it was difficult to find out what the limits were, he added.

"An operating system might use an eight-bit field to hold the processor number, meaning a hard limit exists of 256 processors," Claunch said. "Soft limits, however, are uncovered only from word-of-mouth, real-world cases. They are caused by the characteristics of the software design, which may deliver poor incremental performance or, in many cases, yield a decrease in useful work as more processors are added."

The issue meant there were limits on the architecture of systems, caused by the software. "The net result will be hurried migrations to new operating systems in a race to help the software keep up with the processing power available on tomorrow's servers," Claunch said.

Colin Barker of ZDNet UK reported from London.