An OEM Technology White Paper by

Auri@envisioneering.net

www.envisioneering.net

The ongoing trend in the PC industry is to provide great value, with ever-increasing access to services, with steadily increasing power, all at lower cost. Operating systems from Microsoft and others continually include more features, and bundled applications which require — actually, demand — more processing power from the main CPU. Each of these bundled functions (web browsers, network access routines, etc…) relies heavily on new technologies provided by coprocessors.

These coprocessors are now optimized to specialize in one of the hottest areas of PC consumer functions — DVD playback — while also assisting 2D and 3D acceleration, sound processing, and other multimedia technologies. The prime benefit to PC designers and manufacturers in offloading much of this media task-juggling from the main CPU is freeing-up the central CPU to process "constant computing" processor-intensive tasks. These include file encryption, decryption, virus checking, compression and decompression, Internet channel paging and more. In general, they satisfy the ever-increasing need for power demanded by today's accomplished and demanding operating systems and the infotainment applications consumers want to have at hand.

Lower cost hardware solutions which deliver sustained quality provide new and upgraded computers the performance capabilities consumers want, including high-end DVD-Video playback and ultra-realistic gaming, allowing the main processor to provide even more robustness and usefulness to the user. Implementers benefit from the ability to upgrade existing equipment at minimal cost to new capabilities, allowing minimal Bill of Materials increases while expanding the overall system margin with new capabilities at minimal cost.

Coprocessors also provide the cost-conscious user an easy upgrade path. Upgrades occur either via an upgrade "kit" for DVD-Video, or by using lower-cost processors in conjunction with the coprocessors in very low-priced PC systems. So long as consumers demand more multimedia performance from mainstream PCs, there will always be a steady demand for coprocessors to deliver not only smooth media playback but also the kind of at-the-ready CPU performance reserve consumers expect.

Testing was performed on a Pentium II system running at 300, 350, and 400 megahertz, and a Pentium II Celeron processor running at 300 MHz. All systems had between 32 and 64 megabytes of RAM and ran Windows 98. A Microsoft developer tool, WinTop, was used to gauge processor / thread usage. All DVD-ROM drives used were second generation products. Software DVD-Video decode was performed by Quadrant, Inc.’s software, version 1.5.00.1052/0.9.06.1311. Hardware-based MPEG-2 decoding (the coprocessor’s task) was performed by a Sigma Designs Hollywood Plus PCI board version 4.06.96. Note the Sigma Designs solution decodes audio in software, but also offers an AC-3 passthrough for decoding by an external decoder.

The test result graphics within this document describe MPEG-2 DVD-Video playback with three variations of sound processing: Hardware AC-3 Downconvert, Hardware AC-3 Passthru, and Software AC-3 Downconvert. Downconverting, processed by both Quadrant and Sigma’s solutions, takes the audio stream from the DVD-Video title, be it AC-3, Dolby Digital, monaural or other, and converts it to 2 channel stereo for playback through the PC sound card. Hardware AC-3 Passthrough denotes the audio was not processed at all (not possible in software, which must decode everything, including AC-3), and is passed through a port on the back of the expansion board (Sigma Designs).

Coprocessors can also be considered media coprocessors, hardware assisting solutions that accelerate specific computing functions which would otherwise burden the host CPU, stealing processor performance normally expected by the user. These processors give the host the ability to calculate other intensive tasks in conjunction with the abilities of the coprocessor. This delivers a richer multimedia playback experience for both the consumer and professional consumer alike.

Coprocessors have been designed to take over such traditionally processor-intensive tasks as MPEG-2 decoding (Sigma Designs’ Hollywood Plus, others) and 3D acceleration (3dfx, Intel i740, others). Coprocessors handling such tasks have unique applications across both low-end and high-end PC markets. These solutions come at a minimal cost increase, but higher product value.

Low-cost media coprocessors are almost an OEM necessity to offer high-scale media management on modern low-end, low-cost systems, such as those using Intel Celeron processors and lower performance AMD and Cyrix CPUs. DVD Video playback on these devices is possible through software, although a large processor cache size is required for full frame rate playback. However, during DVD playback the CPU demand is so great that the system cannot process anything else, save a few OS calls and DVD-Video decode at the same time.

Coprocessors also have applications in other areas of the computing spectrum. High end, or Performance PCs, upgrade paths for PCs, and many dedicated consumer electronics devices already benefit from these dedicated coprocessors. Performance PCs already have powerful 3D acceleration hardware for spectacular graphics performance. Consumer electronics devices have dedicated hardware to handle graphics, modems and speech recognition capabilities. An MPEG-2 coprocessor enables that entire product spectrum to handle MPEG-2 decoding, an incredibly complicated decode task. This lets the host concentrate on other tasks. Other tasks may include audio decoding or calculating which corner the next ‘bad guy’ will be hiding behind in an interactive game.

Lending a helping hand to the host CPU enables many other tasks to take place. These additional tasks could include on-the-fly compression, decompression, encryption, decryption, Internet access, image refinement, intelligent agents and more. Intel has described these general background tasks with the term Constant Computing, which assuredly consumes central processor MIPS, but in general enables the user to be more productive by counting on these powerful services to always be available.

Unfortunately, these background tasks also need to share processor cycles, and combine to take away processor reserve power leaving users feeling their machine is sluggish. DVD-Video playback is gaining in popularity on PCs, and DVD-ROM drives will start displacing CD-ROMs in PCs in 1999. However, DVD-Video playback is a mathematically intensive task.

On low-end systems, DVD-Video processing on the host CPU tends to leave no processor power reserves left for background tasks. On higher end systems, such as 450 MHz Pentium IIs, about 40% of the CPU is left for other tasks. While a 450 MHz Pentium II or 333 MHz AMD K6 may be able to handle host-based processing, the host Windows operating systems are demanding more and more of the processor, making it possible that only 20% may be available in the next OS release or service pack upgrade.

Figure 1.

Indeed, Microsoft Corp. has been discussing new intelligent agent and natural language input technologies which are sure to consume every processor cycle and thread they can identify.

Game and edutainment titles are demanding more video playback performance responsiveness even as the host CPU capabilities in PCs become more powerful. The key force driving the need for quality MPEG-2 playback in these titles is Microsoft itself, urging developers to use the robust Motion Picture Experts Group worldwide digital video standard format at its many developer conferences.

A coprocessor therefore is the logical choice for offloading stress from the processor, enabling Constant Computing to occur, with a corresponding minimal Bill of Materials increase. As a result, a better experience is offered to the consumer. Consumers are happier if they can just "do more things" instead of having to shut off the DVD player function just to crunch some numbers or to surf the 'Net — a web seeking function many Hollywood studios try to build into their more interactive titles. Consumers generally have a "Ph.D." approach to new technologies — they should be "Push Here Dummy" with no compromises.

For the near term, host-based processing for consumers shouldn’t have to be disabled to "make the system fast again." A lack of performance can lead to expensive OEM PC maker tech support calls with consumers asking "Why is my system so slow?" A coprocessor can help prevent these phone calls and their post-manufacturing attendant costs. It could also shave total cost of manufacturing budgets and halt, or slow, complaints from dissatisfied PC customers.

Not all consumers will want to upgrade their old system to a new system. In this instance, coprocessors are their only choice if they want more "modern" capabilities, such as DVD-Video playback. A 150 MHz Pentium–regardless of whether it has MMX–cannot decode MPEG-2, nor can it process high-end 3D graphics in software. An upgrade card is the old PC system user’s upgrade path for now and the future, following users much as a printer can go from system to system. However, offering a low-cost PC with DVD-ROM and hardware-based DVD-Video playback may encourage purchase of a new system.

Low-cost PCs, also known as Basic PCs, made up more than half of the PCs sold in 1997, and consumers seem to be extending that trend in 1998 and beyond. OEMs know the profit margin on these systems is increasingly getting lower, as consumers demand more for less. It is sometimes painfully expensive to add additional capabilities to a system and remain profitable in the process.

Basic PCs generally use low-cost, generally lower performance processors, such as Intel’s Celeron, AMD’s K6, Cyrix’s MediaGX, IDT’s WinChip and Rise Technologies solutions. Basic PCs run up to 333 MHz today and are priced in the $399 to $999 price range. As DVD-ROM drives and DVD-Video movie capability become technology enabled across all PC platforms (including the low-end), the PC OEM will want to provide DVD-Video playback within those units.

Host-based processing is possible in these systems, and Intel claims a 266 MHz Pentium II can handle DVD-Video playback. Unfortunately, software DVD-Video playback uses up all available cycles in the lower end processor spectrum. Worse yet, host-based processing on low-end processors will, in many cases, drop frames as the operating system uses the processor for a variety of background tasks, or as the user interacts with the machine — expecting snappy performance. Users don’t want to wait for

their brand new PCs to take seconds longer to respond because of a movie playing in the background!

Performance issues such as these frustrate consumers, who wonder why their systems are so slow and

why they have to stop DVD-Video playback to get other tasks done. Having to stop one task to do another could very well damage the consumer’s faith in the ability of the PC OEM to deliver a quality product.

Offloading DVD-Video playback onto a coprocessor can alleviate these issues. For example, consumers can type documents in Microsoft Word and use its agent technologies (background tasks that "learn" about the user), spell checker, and grammar checker while watching their favorite DVD movie. They can also download email in the background, all without tying up the host processor. See Figure 2 for a CPU availability comparison on Basic PC processors.

With hardware-based MPEG-2 decoding for DVD-Video playback, a lower-cost processor, such as an Intel Celeron or an AMD K6, can be used very effectively and offer a compelling DVD-Video experience with just $25 worth of silicon in quantities of 10,000 units (for Sigma Designs’ Hollywood Plus). A third-generation DVD-ROM drive is $45 in quantities of 10,000 from Matsushita, quite price

competitive with CD-ROM. That DVD-ROM drive is capable of enabling many more applications to the consumer — a strong PC purchasing incentive. A CD-ROM in comparison is approximately $15-$20. (See Figure 3 for a rackup of various OEM component parts.)

The following chart shows the cost of 333MHz AMD K6, 300MHz Intel Celeron and 333 MHz Intel Celeron systems on an Intel 440LX equivalent motherboard with 16x CD-ROM, DVD-ROM and host-based (software) DVD-Video playback, and DVD-ROM with coprocessor-based (hardware) DVD-Video playback systems. Each system has 32 megabytes of memory, a 2-4 Gigabyte hard drive, 4 MB VRAM 3D graphics board, external 56K modem, sound circuitry and two speakers (based on Figure 3). Windows 98 and Microsoft Works (or other productivity software) licenses are extra. The DVD-ROM included in these systems is a third-generation unit capable of playing back DVD-RAM discs.

By offloading tasks that today demand too much CPU time, other host-based operations can be performed at a minimal cost. Such functions include software modems, audio processing, natural language input, and — as mentioned earlier — compression, decompression, encryption, decryption, and more.

For example, Sigma Designs’ Hollywood Plus chip decodes MPEG-2 in hardware, yet leaves the audio decoding to the processor, making it possible to enjoy Dolby Digital AC-3, Sony SDDS, THX, etc. decoding in software. This gives the user access to new audio options, such as Meridian Lossless Packing in DVD-Audio.

Processors of the future are sure to be faster and will have more processing power that can easily handle DVD-Video playback. Intel’s six-month price slashing schedule withstanding, the performance processors of today will seep into the low-end PC marketplace tomorrow. However, technology does not stand still, and as new technologies and more demanding operating systems come into the fray, more and more will be demanded from the host processor. Therefore, coprocessors will be necessary to handle ever more demanding consumer tasks in future systems.

Intel has defined a three-tiered PC computing platform that consists of Basic PCs (the low-end), Consumer PCs, and Performance PCs. The latter two are usually powered by Intel Pentium II processors. These systems usually run at 333 MHz or higher, come with a minimum of 256 KBytes cache (usually 512K), and have 100 MHz busses. Consumer PCs, which are in the $1299-$1799 price range, tend to have less bells and whistles than the Performance PCs, sometimes called "Enthusiast" PCs. Performance PCs generally have incredibly high performance graphics boards and very high storage capacities, bumping the price range to over $2000. These systems are fully capable of playing back DVD-Video titles and processing background tasks at the same time. However, only about 40% of the processor is left over for such extra tasks. Extra processor-intensive tasks such as additional multimedia (e.g., PowerPoint presentations with MPEG-2 and animation), high-speed file transfers, and graphics processing can suffer as they compete for processor cycles.

The second highest selling PC category is the Consumer PC. Not using the latest and greatest processor but using one-generation-behind CPUs (like a Pentium II 400 MHz rather than 450), these are still leading-edge, high power, high performance PC systems in the $1,200 to $1,799 range and have a higher profit margin. Adding inexpensive silicon for a better consumer experience can be used to effectively market and sell these higher margin systems to consumers choosing between a "Basic PC" and a "Consumer PC."

To provide the customer higher value in a multimedia consumer package, a coprocessor can be used to enable more tasks to be done simultaneously. With such a coprocessor, natural language input, compression, encryption, and other processor intensive tasks can be given the bandwidth they need to make the consumer more productive. Consumers get more bang for their buck by being able to do more, faster. Consumers also tend to form positive views of the PC brand they have just purchased since they have gained the ability to "Just do it" when it comes to an operation they want to perform.

Host-based processing of intensive tasks is not a pipe dream. Consider that today’s PC systems often do not leave enough CPU to be able to handle multiple high-end multimedia applications in software, even with the highest performance PC processors available today. Even in Performance PCs, not enough processor bandwidth is left for other compute-intensive processes. No consumer wants to spend $2,500 for a system that cannot play DVDs and handle other applications at the same time. Processors will get fast and capable enough, but not now, and not this year.

The vast majority of the installed base of PCs in the marketplace is powered by processors that are below Pentium II performance levels. There are many 166 MHz Pentiums, and over 70 million 133 MHz non-MMX system level performance PCs out there. Many of those who decide not to buy a new PC will desire DVD upgrades. Host-based DVD playback, let alone MPEG-2 decoding, is not possible for any of these installed PCs. A Pentium II running at 233 megahertz was the bare minimum for the Quadrant player tested, the lowest processor performance bar of any software decoder available for the PC today.

DVD upgrade kits rely on coprocessors to handle MPEG-2 video playback, just as games on older systems tend to demand better graphics boards and 3D acceleration. A coprocessor can affordably enable a 133 MHz machine with only 32 megabytes of memory to play back DVD-Video. Figure 6 illustrates CPU availability using hardware for MPEG-2 based DVD-ROM and DVD-Video titles on a legacy PC.

Older PCs can be upgraded only so far. Some consumers upgrade until their system is no longer either economically or physically feasible to upgrade, and then search for a new machine. If hardware assist has been beneficial to them as an attractive method of extending their PC purchase value in the past, then offering that feature in a new system becomes a purchasing incentive.

Upgrade kits enable legacy PCs to deliver full DVD-Video and MPEG-2 file playback performance while maintaining some host CPU reserves. However, coprocessors need not be used only in Windows 98 and NT systems. Coprocessors can also be utilized in consumer electronics systems that generally do not have processors capable of host-based DVD-Video playback. These systems generally run Windows CE or other low-overhead, minimal-footprint operating systems, and are generally low in cost. One example would be WebTV, and possibly a forthcoming Windows CE version of WebTV for Windows.

In the future, host processors will be able to handle tasks unimaginable on today’s systems. MPEG-2, DVD, real-time video encoding and more will be done on a host processor without breaking a sweat. However, this will not happen this year. Coprocessors are needed to perform today’s complicated media tasks, and will be needed for the future’s ever more demanding breakthroughs.

Moving forward, an OEM can provide benefit to its customers with a coprocessor. Furthermore, the increased performance leads to positive brand recognition and fewer customer support calls. Coprocessors are a bottom line incentive, increasing profits by offering better performance and attracting and retaining more customers to the same brand.

In today’s PC market, features and quality sell the first product, along with word of mouth. Customers pass their positive experiences to other potential buyers, who will look for "extra processors" to handle "high tech features" that new applications demand. For this year’s available processors, MPEG-2 decoding in a coprocessor is beneficial to low-cost PCs, whose profit margins are subsidized by increased sales volume. Slower processors incapable of handing MPEG-2 and processing other tasks simultaneously require coprocessors. As for those without the funds to buy new computers, coprocessors are their only option, and they will look for the same satisfying feature in a future computer purchase.

In essence, coprocessors for DVD decoding are a near necessity today in personal computers and will also be an enabler in computing appliances for a long time to come.