M I C R O P R O C E S S O R R E P O R T
`
`MPEG Choices for PCs Abound
`Soon to Be a PC Standard—Vendors Vie with Different Solutions
`
`by James L. Turley
`
`MPEG-1 recently took a major leap toward becom-
`ing the de facto standard for delivering compressed
`video on personal computers. Compaq’s decision to
`include MPEG-1 decompression hardware as a stan-
`dard feature on future models (see 0909MSB.PDF),
`coupled with Microsoft’s move to include MPEG-1 de-
`compression software with Windows 95, has pushed
`MPEG-1 past the point of critical mass and into the
`mainstream of the PC business.
`The accelerating adoption of MPEG-1 over compet-
`ing compression algorithms will ignite a boom in the de-
`velopment of software that includes color video and
`audio. Games especially will take advantage of these fea-
`tures, but multimedia reference titles will follow, with
`increasing reliance on the format to deliver extra content
`to business users.
`These developments are bad news for proponents of
`competing audio/video compression formats like Cine-
`pak, TrueMotion, and Indeo. With no clear industry
`leader, software vendors are divided among these codecs.
`But with new impetus from Compaq and Microsoft, multi-
`media developers will migrate toward MPEG-1, adopting
`it more earnestly as the installed base grows.
`Several solutions are available for MPEG-1 play-
`back on a standard PC. While software decompression
`and playback is now possible, hardware decoders will be
`a growth industry in the coming years.
`Let There Be Video
`Over the past few years, several methods have
`emerged to provide full-motion video on standard PCs.
`Uncompressed, even a still 256-color image at 640 · 480
`resolution requires 300K; animating such images at 30
`frames/s (the accepted minimum for smooth full-motion
`video) is beyond the capabilities of most PCs. Even one
`minute’s worth of this uncompressed video would fill a
`600M CD-ROM. Thus, compression and decompression
`are mandatory before video can be stored or played back
`on inexpensive computers.
`In the absence of any standard decompression or ac-
`celeration hardware, only the x86 instruction set and a
`VGA-resolution monitor are givens. In recent years, sev-
`eral companies strove to develop real-time decompres-
`sion algorithms that balanced playback quality with rea-
`sonable storage requirements, resulting in a number of
`lossy, low-resolution software codecs such as Cinepak,
`TrueMotion, and Intel’s Indeo.
`
`These delivered small, often jerky color video with
`unsynchronized sound or no sound at all. The MPEG-1
`standard (see 060803.PDF) was touted as a better- quality
`alternative, compressing 352 · 240 images at 30 fps into
`a 176K/second data stream, including stereo audio. But
`its more complex algorithm was too compute-intensive
`to deliver a reasonable frame rate on current 386 or
`486DX processors. Thus, Indeo and other codecs that de-
`livered poorer quality but ran on available CPUs gained
`popularity. But as CPU performance steadily in-
`creases—from 486DX2 to DX4 and Pentium—it becomes
`feasible to perform MPEG-1 decompression in some-
`thing approaching real time.
`Another drawback of Cinepak, TrueMotion, and
`most forms of Indeo is that they cannot encode video in
`real time. This makes them less attractive to software
`producers than MPEG, because the source video must be
`stored in uncompressed form in its entirety and com-
`pressed offline at a later time. MPEG, in contrast, can be
`compressed on the fly in real time. Although MPEG-1
`encoders are priced upwards of $1,000, this is a small
`price compared with the cost of massive amounts of disk
`storage required for uncompressed video.
`Software Provides One Solution
`Two companies, Xing Technology and Mediamatics,
`have developed software engines for decompressing and
`playing MPEG-1 video and audio on standard PCs. Both
`products run under Windows 3.x, avoiding the 64K dri-
`ver limit of MS-DOS. Microsoft licensed a version of
`Mediamatics’ MPEG Arcade Player to include in Win-
`dows 95. In an effort to improve graphics performance
`under Windows 95, Microsoft replaced much of the DCI
`(device control interface) graphics library in Windows
`3.x with DirectDraw. This change requires different
`binaries for applications that run under both versions of
`Windows. It has also been a significant setback to Xing,
`among others, which relied heavily on DCI functions for
`its software MPEG player.
`The Mediamatics decoder can play MPEG-1 video
`at approximately 24 fps on a system with a 90-MHz Pen-
`tium and a medium-priced graphics accelerator card
`(e.g., Diamond Stealth64). Like the Xing player, the
`Mediamatics software makes extensive use of the dis-
`play adapter’s accelerator chip to perform color-space
`conversion and scaling. These two functions comprise a
`significant portion of the MPEG decompression routine
`and are particularly time-consuming and inefficient on
`the x86 architecture. Although the software from both
`
`MPEG Choices for PCs Abound
`
`Vol. 9, No. 10, July 31, 1995
`
`© 1995 MicroDesign Resources
`
`Apple Exhibit 1019
`Page 1 of 4
`
`

`
`M I C R O P R O C E S S O R R E P O R T
`
`companies will run with an unaccelerated VGA adapter,
`their performance is much worse.
`Mediamatics estimates that its software running on
`a 120-MHz Pentium will provide full 30-fps playback.
`Even with a 120-MHz Pentium, however, sound quality
`is still sub-par. An MPEG data stream includes stereo
`44.1-KHz, CD-quality sound. The Mediamatics player,
`however, performs only a cursory demodulation of the
`sound stream, delivering 11-KHz stereo sound compara-
`ble to an AM radio. In taking advantage of a faster pro-
`cessor, Mediamatics chose to boost the frame rate rather
`than improve sound quality. Either way, 100% of the
`Pentium is devoted to processing MPEG-1 data.
`This is the biggest problem with software decom-
`pression. Although acceptable MPEG-1 playback is now
`available essentially for free (at least on systems with a
`90-MHz Pentium and Windows 95), it leaves no head-
`room for any additional processing. One could argue that
`additional processing is not needed; that video playback,
`by its very nature, commands the user’s full attention.
`Playing a video in the background behind a spreadsheet,
`for instance, makes little sense. Apart from those people
`who want to keep a miniature television running in the
`corner of their screen, any additional CPU cycles would
`be wasted anyway.
`While this may be true for “serious” MPEG applica-
`tions in education, business, or reference works, it is cer-
`tainly not the case for entertainment titles and games.
`Nor is it applicable to multitasking operating systems
`that allow file transfers, fax transmissions, or print
`spooling in the background. These all require significant
`amounts of additional processing at the same time the
`video and audio are playing. To avoid overwhelming the
`processor, a hardware assist of some kind is required.
`Adding a hardware MPEG-1 decoder frees the processor
`for other tasks and improves playback quality.
`Hardware Decompression Spares CPU
`In addition to its soon-to-be-popular software de-
`coder, Mediamatics also develops hardware. With a busi-
`
`ness model similar to that of ARM or MIPS Technolo-
`gies, the company has developed decompression cores for
`both MPEG-1 and MPEG-2, which it then licenses to sil-
`icon vendors. VLSI Technology has licensed both cores
`for use in its set-top-box decoder chips. Western Digital
`and a number of other hardware vendors have also
`begun development around the MPEG-1 core, with prod-
`uct announcements expected later in the year.
`As Table 1 shows, current MPEG-1 decoder chips
`are similar in many ways. They all include an internal
`processor of some kind and several additional function
`units, usually a special multiply/divide unit for per-
`forming IDCT (inverse discrete cosine transform) oper-
`ations. A DRAM controller for the minimal 512K buffer
`is also standard.
`Currently, every MPEG-1 decompression chip for
`PCs relies on an independent graphics accelerator chip
`for scaling, color-space conversion (from YUV to RGB),
`frame-buffer management, and an interface to the video
`DACs. Such features are now common in mid-range and
`high-end graphics chips and are often found on even
`basic PCs. Figure 1 illustrates a typical MPEG-1 multi-
`media board with separate graphics, video, and digi-
`tal/analog converters.
`Where some of these chips differ is in the way they
`receive their compressed data. The S3, Winbond, and
`C-Cube chips, for example, all have their own ISA or
`PCI bus interface. This allows the chips to be added as
`aftermarket multimedia accelerator boards, but it ne-
`cessitates irregular or nonstandard cable connections
`between the MPEG logic and the existing graphics ac-
`celerator. Alternatively, these chips can be integrated
`onto one board with the accelerator chip, sharing the
`latter’s bus connection.
`The recently announced chip set from S3 (see
`0909MSB.PDF) offers two roads to MPEG integration.
`The Scenic/MX1 includes a PCI interface that it uses to
`communicate with the company’s general-purpose accel-
`erator, thus allowing it to be used on an aftermarket PCI
`add-in board. The nearly identical Scenic/MX2 version
`
`Decompression
`Chip
`
`C-Cube
`CL450
`
`C-Cube
`CL480PC
`
`SGS-T
`STi3430
`
`Winbond
`W9910IF
`
`Winbond
`W9920IF
`
`Winbond
`W9920CF
`
`S3
`Scenic/MX1
`
`LSI
`L64002
`
`Hyundai
`HDM8211
`
`MPEG-1
`MPEG-1
`MPEG-2
`MPEG-1
`MPEG-1
`MPEG-1
`MPEG-1
`MPEG-1/2
`MPEG-2
`MPEG-1
`PCI
`PCI
`32-bit
`16-bit
`ISA
`ISA
`16-bit
`Bus interface
`16-bit
`ISA
`No
`Yes
`Yes
`Yes
`Yes
`Yes
`No
`Audio?
`Yes
`No
`Yes
`Yes
`Yes
`Yes
`No
`Yes
`Yes
`Video?
`Yes
`Yes
`512K
`512K
`512K–2M
`512K
`None
`512K
`512K
`DRAM required
`1M–2M
`512K
`16-bit
`16-bit
`64-bit
`16-bit
`N/A
`16-bit
`16-bit
`DRAM data bus
`64-bit
`16-bit
`36 MHz
`28 MHz
`27 MHz
`30 MHz
`36 MHz
`40 MHz
`40 MHz
`Frequency
`32 MHz
`36 MHz
`PQFP-100
`PQFP-128
`PQFP-160
`PQFP-120
`PQFP-80
`PQFP-128
`PQFP-128
`Package
`PQFP-208
`PQFP-160
`5 V
`3.3 V
`3.3 V
`3.3 V
`5 V
`3.3 V
`5 V
`Voltage
`3.3 V
`5 V
`0.6 m
`0.65 m
`0.5 m
`0.5 m
`0.8 m
`0.65 m
`0.8 m
`0.65 m
`0.6 m
`Mfg process
`$20
`Price (10K)
`$35
`$45
`$41
`$9
`$35
`$36
`$23
`$78
`Table 1. MPEG-1 and MPEG-2 decoder chips are available from a number of sources, with various feature sets. Older chips separate audio
`and video decode functions, while newer designs integrate them. (Source: vendors)
`
`2 MPEG Choices for PCs Abound
`
`Vol. 9, No. 10, July 31, 1995
`
`© 1995 MicroDesign Resources
`
`Apple Exhibit 1019
`Page 2 of 4
`
`

`
`M I C R O P R O C E S S O R R E P O R T
`
`drops the PCI interface in favor of a dedicated 8-bit bus
`between itself and the accelerator chip. This approach
`keeps video traffic off the host PCI bus. The Scenic/MX2
`could also be offered as an end-user upgrade, possibly as
`a daughtercard on an S3-based graphics board.
`Although the interchip bus is proprietary, the S3
`chips can configure the interface for compatibility with
`existing digitizers from Philips, C-Cube, IIT, and others
`that output 4:2:2 YUV data.
`The MPEG-1 standard specifies an output resolu-
`tion of 352 · 240, but in PCs this results in a small pic-
`ture, which is often scaled to a larger window. The
`method used to “upsample,” or replicate pixels to achieve
`the larger size affects the perceived quality of expanded
`video windows. Some chips interpolate pixels, using a
`smoothing algorithm similar to those commonly found in
`laser printers. Others simply replicate the pixels, result-
`ing in blocky, albeit larger, video images.
`Audio, Video or Both?
`Another feature separating MPEG chips from one
`another is their treatment of the audio data stream. At
`the low end, MPEG chips either do not decode the audio
`stream at all or, like Winbond’s W9920IF, require a sep-
`arate audio chip. For the lowest-cost implementation,
`omitting audio support has price advantages. But with
`both audio and video playback now available purely
`through software, the market window for these very low
`end devices is rapidly closing.
`Many chips that include audio functions still rely on
`extra support, using device drivers running on the host
`CPU to demultiplex the audio from the video stream.
`This task is not extraordinarily complex, and it allows
`the system to handle nonstandard or encrypted MPEG
`data streams. S3’s Scenic chips fall into this category.
`However, synchronizing the audio and video when they
`are handled by separate chips is a nontrivial task.
`Standard Microprocessors Need Not Apply
`Interestingly, although most vendors supplying
`MPEG decoder chips hold licenses for embedded micro-
`processor cores, only one company has chosen to use it
`for its MPEG products. Hyundai, which acquired a
`SPARC database through TI in 1994, is the only vendor
`to use a standard microprocessor instruction set.
`LSI Logic, for example, opted for a custom logic
`design rather than apply a MIPS core to the problem.
`Winbond bypassed PA-RISC, and IBM, which has an
`MPEG-2 decoder, did not include PowerPC. C-Cube de-
`signed its own core, while IIT, which has x86 experience,
`chose to buy the MIPS-X core developed at Stanford (see
`MPR 10/30/91, p. 1). Cirrus Logic, while it has not an-
`nounced any MPEG chips to date, is unlikely to exercise
`its ARM license, relying instead on the CompCore Multi-
`media design it licensed in 1994 (see 0811MSB.PDF). The
`
`C-Cube and Hyundai chips are the only ones to actually
`execute external code, the others relying on internal
`microcode ROMs.
`The widespread decision to develop special MPEG
`cores highlights many important aspects of this market.
`First, price is a critical factor. Cost requirements for
`mass-market peripherals can’t support CPU licensing
`fees. Second, none of the current 32-bit microprocessor
`architectures is well-suited to the task of performing dis-
`crete cosine transforms or other routines necessary for
`video decompression (witness the need for a 120-MHz
`Pentium to keep up with a 40-MHz CL480). Finally, soft-
`ware compatibility is not an issue for a chip that will ex-
`ecute only a carefully optimized device driver for one
`specific task. On all counts, mainstream microprocessors
`come up short.
`Furthermore, video-output tasks gain very little
`benefit from the conventional data cache designs found
`on general-purpose microprocessors. Unlike computer
`applications, there is little or no locality of reference for
`video data; once it’s processed, it’s gone.
`At this time, there are no chips that combine MPEG
`decoding and conventional graphics acceleration. All the
`MPEG-1 decoders are separate devices and require ei-
`ther their own connection to the system bus or a private
`interface into the video subsystem. For the time being,
`end-user MPEG upgrades are often awkward and re-
`quire add-in cards or ribbon cables between boards.
`Today’s graphics-chip vendors argue that the dual-
`chip approach is not likely to change within the next
`year, for reasons both technical and commercial. Al-
`though the adoption of MPEG-1 will be widespread, and
`the demand for MPEG-1 accelerators will increase, sin-
`gle-chip multimedia devices are not imminent, they say.
`An MPEG-1 decompression engine requires approxi-
`mately 400,000 transistors; graphics accelerators are not
`much smaller, and few functions can be shared between
`
`Video
`DAC
`
`Frame
`Buffer
`
`MPEG
`Buffer
`
`Audio
`DAC
`
`VGA
`Controller/
`Accelerator
`
`MPEG-1
`Decoder
`
`PCI
`
`Figure 1. A typical MPEG-1 decompression solution for PCs adds a
`decoder to an existing GUI accelerator.
`
`3 MPEG Choices for PCs Abound
`
`Vol. 9, No. 10, July 31, 1995
`
`© 1995 MicroDesign Resources
`
`Apple Exhibit 1019
`Page 3 of 4
`
`

`
`M I C R O P R O C E S S O R R E P O R T
`
`the two. The MPEG-1 decoder market is too uncertain
`for chip vendors to gamble on what is for them a very
`large die. By 1997, however, sub-half-micron technology
`should allow these functions to be economically merged
`in a single die, sharing a PCI bus interface and a single
`DRAM buffer.
`
`Whither MPEG-2?
`With current processor performance, MPEG-1 de-
`compression can be performed either by software or in
`hardware. However, a software decoder for MPEG-2
`(which delivers 4· resolution over MPEG-1) may be
`many years away. Even with the P6, Mediamatics claims
`that acceptable MPEG-2 performance is not possible
`without a significant hardware assist.
`MPEG-1 will offer sufficient resolution and sound
`quality for PC users for several years to come. It has to—
`MPEG-2’s storage requirements are better suited to
`broadcast media, where storage is not an issue. MPEG-1
`was developed to address both the capacity and band-
`width limitations of original CD-ROM technology. A
`modern 4· CD-ROM drive can read data quickly enough
`to play back MPEG-2 streams, but the discs still do not
`have the capacity to make it worthwhile. MPEG-2 on a
`typical 4· CD-ROM is sweet and short.
`Several major consumer-electronics giants are
`striving to change this situation. Both the Sony/Philips
`alliance and the Toshiba/Time Warner sodality are seek-
`ing to establish their respective high-density CD-ROM
`formats as industry standards. So far, both factions are
`accumulating licensees at a brisk pace, leading to a po-
`tential industry split reminiscent of the VHS/Beta battle
`of a generation ago. This would obviously be destructive
`to the industry as a whole, so hopefully some agreement
`on a common standard will be reached soon.
`If one (or both) of these “quad-density” CD formats
`becomes popular, it will then be feasible to deliver
`MPEG-2 content on a static storage medium suitable for
`PCs. Fortunately, MPEG-2 is a superset of MPEG-1, so
`earlier titles should still play on an upgraded machine.
`Until that time, MPEG-1 will be the de facto standard
`for PCs, while MPEG-2 is relegated to the higher end of
`the market for broadcast encoding. The proposed MPC-3
`(multimedia PC) standard, in fact, specifies MPEG-1 at
`30 fps as a minimum requirement.
`Adoption and Outlook
`The MPEG-1 standard is sufficient for compression
`and playback on PCs, where video quality is not crucial
`and storage space is limited. Moreover, video playback
`on a PC is currently more of a novelty than a require-
`ment, and most long-time computer users are thrilled
`that their PC can play small, grainy images with AM
`radio–quality sound at all. Except for a few games
`(which always seem to push the limits of PC hardware)
`
`For More Information
`For additional information about the MPEG decoder
`chips and software mentioned in this article, contact
`the vendors listed below.
`C-Cube Microsystems (Milpitas, Calif.) 408.944.6400;
`fax 408.944.6314. Cirrus Logic (Fremont, Calif.)
`510.226.8300; fax 510.252.6020. Hyundai (San Jose,
`Calif.) 408.473.9200; fax 408.473.9800. LSI Logic
`(Milpitas, Calif.) 408.433.8000; fax 408.433.8989.
`Mediamatics (Santa Clara, Calif.) 408.496.6360; fax
`408.496.6634. S3 (Santa Clara, Calif.) 408.980.5400;
`fax 408.980.5444. SGS-Thomson (San Jose, Calif.)
`408.452.8585; fax 408.452.1549. Winbond (San Jose,
`Calif.) 408.943.6666; fax 408.943.6668. Xing Technology
`(Arroyo Grande, Calif.) 805.473.0145; fax 805.473.0147.
`
`and multimedia encyclopediae, none of today’s software
`is near the 74-minute video limit of a standard CD-ROM.
`For as long as mainstream PC applications do not stretch
`the capabilities of the standard, MPEG-1 will continue to
`satisfy the majority of users and OEMs.
`Including MPEG-1 decompression software (along
`with Cinepak and Indeo decoders) with Windows 95 will
`give millions of casual PC users their first taste of MPEG
`video, much as QuickTime did for Apple customers.
`Those who run Windows 95 on their new Compaq Pre-
`sario or other PC with hardware MPEG support will be
`happier still. The widespread availability of software
`MPEG players will spur the purchase of MPEG-1 accel-
`erator boards as more and more PC owners seek to sat-
`isfy an appetite whetted by Windows 95. Microsoft’s
`move, in fact, has overjoyed the makers of 1996’s “games
`accelerator” cards. Users who are unsatisfied with the
`performance of their software decoder will likely take a
`big step, paying perhaps $150 for a hardware accelerator
`that delivers noticeably better performance.
`As an interesting side note, Apple will start includ-
`ing MPEG boards in low-end Macintoshes while its high-
`end systems rely on software decompression—because
`MPEG chips are cheaper than a faster PowerPC, which
`can perform MPEG-1 decoding at 30 fps with full audio.
`The demand for cheap MPEG chips at the low end
`will come from motherboard makers and OEMs that
`need to offer MPEG as a checklist item. Like PCs with
`accelerated graphics cards today, these will become com-
`petitive necessities in 1996, but they do not have to offer
`competitive performance in the customer’s system.
`That leaves a great middle territory with few po-
`tential buyers. Many of today’s chips fall into this cate-
`gory. Separate audio and video decoders are too expen-
`sive to include on a motherboard and not fast enough to
`satisfy performance-hungry upgrade customers. A battle
`is brewing for the high ground in PC MPEG-1 chips, and
`several vendors are girded for combat. ¤
`
`4 MPEG Choices for PCs Abound
`
`Vol. 9, No. 10, July 31, 1995
`
`© 1995 MicroDesign Resources
`
`Apple Exhibit 1019
`Page 4 of 4