3D --- Media Creation --- Shared Space
Published by 4th WAVE, Inc.
Issue #615 12/6/96
CONTENTS
Microsoft Melts Down With DirectX 5.0
Microsoft's Meltdown, a week of testing and debugging held every summer to test game software and system hardware, is having a special session from Feb. 11-13 to test DirectX 5.0. The summer session will be held from July 9-11. Once more info becomes available, it can be found at their Web site:
http://www.microsoft.com/hwdev/meltdown.htm
Last Monday (Dec. 2), Paradigm Simulation released Vega 3.0, the next generation successor to their SGI development tool. Using LynX, a X- Windows/Motif point-and-click environment, Vega is a "simulation software environment for real-time visual simulation, virtual reality, and general visualization applications" for building real-time 3D environments. Vega 3.0 offers several new features including immediate rendering and feedback, clip map support, directional horizon glow, and dynamic video resolution support. Vega 3.0 is available now with prices starting at $3,495.
http://www.paradigmsim.com/vega.html
This week at the Interservice/Industry Training Systems and Education Conference (I/ITSEC) in Orlando, Intergraph announced several new products in their 3D hardware line, including the newest member of their RealiZm graphics accelerator line (only available in their Windows NT- based TDZ and StudioZ workstations), the RealiZm V25. It offers a whopping 64 MB of dedicated texture memory and a frame buffer of 32 MB, along with advanced features including translucency and stenciling, along with a geometry accelerator option that performs 840 MFLOPS.
The RealiZm V25 was demonstrated along with the new TDZ 410 and TDZ 610 RAX rack-mountable workstations, which will utilize dual or quad 200 MHz Pentium Pro processors along with up to 1 GB of memory. The TDZ RAX workstation modules, which will begin shipping in January (and will include the RealiZm V25 accelerators), range in price from $16,700 to $31,600.
Also previewed was Intergraph's new InterView 28hd96 28-inch high- definition aspect ratio (16:9) multisync monitor for wide-screen viewing. The InterView 28hd96 Color Monitor will be available starting in May 1997 for $9,995.
http://www.intergraph.com/ics/
3Dfx Announces New Accelerator
This week 3Dfx Interactive unveiled the latest high-performance accelerator based on their Voodoo Graphics chipset, the Obsidian XS-100. The XS-100, the highest-end accelerator based on Voodoo Graphics yet released, if targeted at the professional markets such as military and civilian visual simulation applications and LBE system integrators. It offers an impressive performance of up to 100 Mpixels/sec (tri-linear filtered, texture-mapped), along with 3Dfx's scan-line interleaving and texture-streaming architectures which provide over 2.4 GB/s of dedicated graphics memory. The XS-100 offers a 4MB frame buffer and 8MB of texture memory, with display resolutions of either 640x480 or 800x600. The Obsidian XS-100 supports both 3Dfx's own Glide API as well as Microsoft's Direct3D, and is supported by many key development tools such as Datapath Realimation, Gemini OpenGVS, Hybrid SurRender, Sense8 WorldToolKit as well as others. The XS-100 will be available later this month through systems integrators, VARs and OEMs for $2,500 MSRP; initially, there will be about 10 dealers/VARs/integrators, mostly in the vis-sim market, which can be found by contacting 3Dfx directly (will also soon be listed at their Web site).
http://www.3dfx.com/products/obsidian.html
ATI Announces 2m 3D Chips Shipped
At the AGP conference ATI announced it had shipped 2m Rage and Rage II chips.
Check out:
"An Incomplete Guide to Programming DirectDraw and Direct3D Immediate Mode" at
http://www.wksoftware.com/publications/d3dim.html
For an interesting assessment of Direct3D compared to OpenGL check out:
"Direct 3D Analysis" at
http://www.sgi.com/Technology/OpenGL/direct-3d.html
Looks like we missed one: On October 29, Trident announced the first 3D geometry engine for consumer-level gaming applications. Available as an option for the new Trident 3DImage accelerators (see WAVE #612, 11/8/96), the Trident TGC335 is a quad DSP-based device with set-up engine capable of transforming 750,000 polygons per second. A reference board will available by the end of the year, with pricing at $30 each in 10,000 per month quantities.
ActiveMovie Moves Forward With Truevision
On Tuesday, Microsoft announced that they have selected Truevision Inc. to extend the ActiveMovie 2.0 API for the growing desktop digital video editing market (see WAVE #610, 10/11/96). Truevision, a leader in the desktop video market since 1984, will "contribute to the design of the ActiveMovie 2.0 specification as well as furnish certain component modules to the Microsoft code base." ActiveMovie 2.0, which is still in development and was previewed by 25 independent developers at a recent meeting at Microsoft, stands poised to make professional video editing on the PC a reality.
http://www.microsoft.com/imedia/default.htm
615.2 Conference Report - AGP/3D Developers Conference; December 3-4, 1996 by John Latta
The Accelerated Graphics Port (AGP) is an Intel initiative to bring a new bus to the PC with speeds up to 500MB/sec which is to enable significant improvements in 3D performance. This, the second AGP conference, was a milestone event which was restricted to only the members of the AGP Implementor's Forum (IF). There were 200 attendees from 60 companies who represented a cross section of the PC industry. There are now 100 companies in the forum.
Mike Aymar, VP and GM of the Desktop Products Group, made the now familiar Intel argument that in spite of the fact that the PC has increased in performance by 2X to 3X every few years that 3D platform should increase by 5 to 10X in 2 years. Consistent with Andy Grove's comments at COMDEX (see WAVE #614, 11/25/96), Mike stated "Graphics is what will get people more interested in PCs." He characterized today's software performance at 120Kpolygons/sec while in late 1997 with AGP this should rise 4X to 600KP/s. By 1999 there should be another major jump to 1 - 2 M p/s with advanced filtering and even procedural effects. Mike referred to the E&S demonstration at COMDEX.
The standard for quality is that a good AGP implementation should exceed the quality of today's best arcade games. He cited Tomb Raiders running on the 3Dfx card with 6MB of texture as a case in point.
Mike set as the AGP target the P6 (Pentium Pro) processor to be used as the geometry engine, a graphics controller for 3D rendering and the core logic chip set which provides control for the complete system including memory management.
Companies which have made commitments for AGP parts include: 3DLabs, Trident, ATI, Cirrus Logic and S3.
By late 1997 he expects that there will be volume production of PCs with AGP. At this time there would also be first titles available and the AGP platform would be introduced. In the first 1/2 of 1998 there will be volume AGP products and this will be matched by AGP penetration into the value line of PCs in the 2nd 1/2 of 1998. 6 months from now an extensive marketing campaign will begin and Intel is expected to have ads in the 2nd 1/2 of 1997.
It was also announced that an AGP Implementing Forum (IF) marketing council is being established to guide the IF. There will be separate requirements for membership. Applications will be available on the Web site and these are due in January. No other details were provided.
In a surprise move, Intel announced that aftermarket cards are recommended to ship with large amounts of texture memory onboard. The reason being that AGP is part of a finely tuned system design and that there is no assurance that adequate memory bandwidth will be available for all aftermarket cards. One of the purposes of AGP, as advocated by Intel, is to use system memory for texture storage. This would allow for cost reduction over the more conventional 3D implementation approach which has its own separate, and costly, texture memory. Thus Intel's aftermarket recommendation for onboard texture memory could severely impact the potential viability of the aftermarket for cards, especially in the low end of the market.
Other Intel presentations provided additional details and these will be summarized. Two documents were included in the materials handed to the attendees and are on the web site: "AGP Platform Design Guide" 1.0 (10/23/05) and "AGP Interface Specification", 1.0 (7/31/06).
The first details of the connector were provided. This part will initially be available from AMP, Frametone and Molex.
Intel laid out a roadmap for the use of AGP in mobile computers. This is consistent with the industry view of what goes on the desktop must move to mobile. Again corresponding with the introduction of P6 for the mobile markets Intel sees mobile AGP happening in the first 1/2 of 1998. The market justification for AGP includes Intranet/Internet (VRML), data visualization, presentations, and VR product models. The key mobile issue is power. What AGP and the 3D chip does is add 5W to an already power constrained platform. Thus, the design tradeoff comes because power scales with 3D performance. A number of recommendations were made to keep the power within 3 watts.
Also for the first time the mechanical details of the add-in card were shown. This prototype card is based on fitting into an Intel motherboard, and given the dominance which Intel has in the motherboard market, the card details are a de facto standard. The card is only 6.6" by 3.25" with a slot of 1.5" by 1" cut out of the back of the card where the connectors normally are. This might seem rather mundane - not so. First, the card size by most PC standards is small and a number of companies we spoke to expressed concern that this limits the amount of texture memory for aftermarket products. Given that aftermarket cards will want to emphasize performance and onboard texture memory is one way to accomplish this the mechanical constraints limit the ways in which 3rd parties can differentiate themselves. The second impact of the design is that there is only room at the back of the card for the standard display cable connector because the notch takes away so much rear card space. Thus, by only using alternate means of connecting to the card will it be possible to provide for video input - an increasingly important display card requirement. WAVE spoke to a number of companies who expressed strong concerns about the limitations that this will have on the development of a strong AGP market. Others saw this as an example where Intel had not thoroughly thought out the system ramifications of AGP and the implementors are left with the consequences.
Microsoft - Major Announcements on DirectX and Direct3D
Carl Stork, General Manager for the Windows Platform and Ty Graham the Technical Evangelist for the Windows Platform presented important new information about AGP, Microsoft's 1997 roll out plans and DirectX.
Microsoft support for AGP is planned to synchronize with two major product events in 1997: "Memphis" the next retail upgrade to Windows 95 (Windows 9X) and NT 5.0 the next major upgrade to Widows NT. Both are due to beta in the H1 1997. Memphis is farther along than NT 5.0 by several months according to Microsoft. There will be no AGP support in Windows 95 or earlier versions of NT.
There will be no Direct X 4.0 and the plans for this version have been integrated into Direct X 5.0 which is to come out with both OS releases. Microsoft made the decision that to accomplish Direct X 4.0 while trying to accomplish the integration of Direct X into the OSs would have been too disruptive. Thus, Direct X 5.0 became the central objective. Direct X 5.0 is to ship June 1997.
Ty Graham announced for the first time that Microsoft acknowledges that Direct3D has not had the performance which proprietary APIs are showing. Microsoft intends to correct this and is in the process of performance tuning. Other changes which are being made include progressive meshes, retained mode enhancements, better documentation, HAL enhancements, DLL tuning and possible API changes. To show the impact of some of this a demonstration of Monster Truck Madness was shown on a 3Dfx card. The performance was stated to be 2X to 3X the frame rate of the current Direct 3D and with an increase in image quality.
Carl Stork gave an overview of the a number of Windows OS efforts. A major objective of these OS updates is to bring the hardware support of Windows 9X and NT in synchronism. A key to this is the Win32 driver model (WDM). WDM brings the concept of Class Drivers and Mini Drivers. A class driver from Microsoft could handle the general requirements of video cameras for example while the 3rd party would write a mini driver for a specific camera. In some cases the Class Drivers will be sufficient that hardware will work without a mini driver. A critical issue for WDM is that Microsoft will address driver performance issues, such as latency in Windows NT, for real time OS support and time critical applications such as multimedia. With WDM manufacturers will only have to write one driver for both Windows 9X and NT. Other areas of hardware support parity is in Plug and Play and power management.
As an illustration of the significance of these upgrades Carl gave a demo of DVD. Using early software he showed how it was possible to play a DVD video and switch between language tracks with the PC. Carl emphasized that the complexity of DVD support in the OS is so great that end users should not count on any DVD functionality until Windows 9X is released. Thus, those early to purchase DVD in the first half of 1997 will be sorely disappointed until the next version of Windows arrives.
Both Cary and Ty stressed multiple times that integrating AGP into both Memphis and NT 5.0 is very tight. There is no time pad left. Microsoft has done considerable work in simulating AGP but they need hardware immediately to begin testing. This includes systems with AGP parts. These tests cannot be restricted to just Intel hardware but Microsoft cited the need to have 100+ different systems to do adequate testing.
Hardware Developer Information:
http://www.microsoft.com/hwdev
Microsoft Developer Information:
http://www.microsoft.com/devonly
The second day of the conference was the OEM day and a number of companies spoke of their AGP plans. Noticeable by their absence were: 3Dfx, S3, Rendition and Real 3D.
ATI laid out its roadmap for AGP and its intent to be a major player during the Christmas 1997 season. Two parts were announced: AGP-133 for each of the workstation and professional market and AGP-66 at the low end the mainstream corporate and consumer desktop. Alpha samples would be available in January with production samples in March and production in May.
3D Labs - AGP Product Plans Defined
GliNT-AGP and Permedia, both at 66MHz, will be sampled 3/96. Delta-AGP will sample Q1 1997. No technical details were disclosed.
Cirrus Logic - Texture Map Management Technology Outlined
A strength of the Laguna 3D chip is its use of Texture Jet (tm) technology for hardware texture management. This includes: texture caching, on-chip address translation and pre-fetching of textures. That same technology is being extended to its AGP part. In addition, Cirrus Logic stressed the importance of video integration including capture and MPEG playback. Samples of the AGP chip are due Q1 1997 with production Q2 1997.
Trident - 3D Roadmap Disclosed
Trident has 6 3D parts in its plans and 3 of those are AGP parts: 3DImage 985, 3DImage 985 DVD and 3DImage 985+. Trident also stresses its support of video including DVD and television out. Their 3D technology is called rCADE3M (tm) which includes on chip: tri-strip backface culling, setup sub- pixel positioning, and rendering. They see that tradition chip packaging is inadequate and will be using BGA packages.
AGP Conference - Points to Ponder
There is little doubt that AGP is very important in enabling 3D graphics on PCs. AGP is also part of the transition to making MMX enabled Pentium Pros the baseline processor on the desktop and in portables. Coupled with Microsoft's release of significant OS upgrades, AGP is one step in another major feature, function and performance boost in personal computing. However, a few details in execution are extant.
At WAVE we came away with a number of uncertainties which could impede, at least in the near term, the realization of AGP's potential. The schedule for 1997 is very aggressive. Many events have to fall in place at virtually the same time: quantities of hardware from many vendors to Microsoft in time to test, Microsoft completing its software development in both Direct X and the OSs, Intel having systems with AGP parts in time for OEMs, 3rd parties shipping chips to OEMs, OEMs building systems with AGP and content developers with games, at least, ready to exploit AGP's potential. A very tall order.
Many developers expressed concern that Intel has not disclosed many details required to develop tuned AGP designs. There is much latitude in the specification and details about memory tradeoffs, buffer sizes and timing were left open. Developer's described the presentations as being an overview of what was already in the specification. Another document "Graphics Performance User Guide" was hinted at that is only available to a select group of developers who have signed a special NDA with Intel. Thus, we see two groups in the AGP development process. A first which will have 1st generation optimized designs and an outside group, made up of most of the industry, which builds to their best estimate of how to achieve competitive performance. The critical issue arises if the aggressive schedule outlined above is not met. From a consumer platform perspective, the window of opportunity passes to the next Christmas, 1998, and the playing field begins to level between the inside and outside groups. In the process another 3D Christmas opportunity has passed.
The target markets for AGP, outside of consumer, remain unclear and it is more than just a marketing issue. Many developer's expressed concern about the mechanical constraints which limit what can be put on add-on cards. Most card buyers will be performance focused and with the combination of limited card area and one back connector the number of options for add-on card differentiation are limited. For example, we see that AGP could well accelerate the adoption of texture compression as a means to get greater use of the limited amount of texture memory which can be included in add-in cards. Intel's surprise recommendation that add-on cards have on board texture memory only reinforces the sensitivity about the difficulty of integrating high performance 3D into millions of PCs. The last concern Intel wants to have is a bow wave of consumer calls when the AGP not working - an AGP black eye will hurt the whole 3D industry.
If the 3D market is to grow it must enter the business desktop. Here again, much was left open on how AGP will support these needs. As 3D performance continues to increase the use of main memory for texture storage is less likely, especially as the depth complexity rises. AGP takes on a different role of providing a high bandwidth port to the CPU. Much design optimization remains to be done to optimize the PC 3D workstation. Hints came from the floor about the next version of AGP which is faster but given all the issues the chip and OEM companies face making 1.0 AGP work few seemed concerned.
As we described in our COMDEX Points to Ponder significant shifts are taking place in 3D. AGP is at the hardware cutting edge but much remains to be done to optimize designs which fully utilize the potential of 3D in all of its potential markets.
615.3 Company Profile - Axial Systems: Making VRML Come Alive by John Latta
Axial Systems was the first VRML company to receive venture capital funding and at WAVE we wondered why. Situated in the heart of Silicon Valley in Saratoga, CA, this 21 person company brings a fresh approach to VRML. In modest 5,000 sq. ft. offices the company promotes an open atmosphere in both its offices and how it works. For example, only on Monday and Thursdays are meetings held and the other days are reserved for "outward focus" and coding. Much of the company is in development with 15 in engineering, QA and documentation. According to the President, John Ison, the out-of-date view of VRML is that it is slow, limited in capability and part of a 3D cult. Simply it is "a standard in search of an application." With compelling content and tools to create it Harry Vitelli, VP of Marketing, believes that there is a "huge market for interactive 3D on the desktop." What Harry means is accelerated 3D and Axial intends to provide their own software solutions to make this happen.
Axial intends to be the leading 3D interactive solutions provider - both with a player and tools. They believe that the success of VRML will be judged by the quality of the 3D experience it brings to users and that real time is a critical element in shaping that experience. This requires that it is easy to create content with tools to match and that content be pervasive.
Two key components which Axial has already developed are its intelligent scene management software for passing triangles to the rasterizer which are only visible in the scene and a software rasterizer - polyman. Of these two components the scene management software is critical in providing their performance advantage.
WAVE was shown several examples which demonstrated Axial's strengths. A Doom world and a set of Quake characters were both passed through a translator so they could be written to a VRML 2.0 file. It is Axial's view that if VRML cannot approximate the Quake experience it is not credible. In this scene, with approximately 1,500 to 2,000 polygons and a display area of 640 X 480 the update rate was 15 f/s. We saw the play rate as being quite realistic. A conventional VRML browser ran at only 3f/s. The 3DLabs 500TX reference card netted only 4 f/s. The key difference between the slow results and Axial's is their intelligent scene management software.
A second application was a commercial one to sell tickets for a sporting event. The intent here is that the buyer could purchase a ticket by location - the 3D interface would allow the customer to experience their seat view by going to various locations in the stadium. If one included a Java application one could also even go to a specific seat and purchase the ticket while in the virtual seat. This demonstration was also fast well beyond most VRML examples we have seen. Axial states that their scene management technology excels in large worlds and this was supported in the stadium demonstration.
Axial has broken the perception that VRML is a performance dog. We can expect to hear much more from this company which is bringing a fresh approach to desktop 3D with leading edge tools and end user products.
615.4 DTV On the Way? by David Lohse
After almost 10 years of squabbling, last Monday (November 25) an agreement was finally reached between industry groups on the long-fought Digital TV (DTV or HDTV) standards. Discussions over DTV began in 1987, and after 8 years, in November 1995 it appeared that a standard would finally be agreed to when the so-called Grand Alliance standard was presented to the FCC by the ATSC (Advanced Television System Committee). However, before the proposed standard could be adopted by the FCC, the computer industry (led by Microsoft) stepped in in protest, claiming that the ATSC standard was not "computer friendly."
Earlier this fall, action began towards an agreement when Susan Ness, an FCC commissioner, urged the industry groups to resolve the issues surrounding DTV so that the FCC could ratify a standard by the end of the year. Over the following four weeks, the industry groups, including the Broadcasters Caucus (led by NBC), the Consumer Electronics Manufacturers Association (CEMA) and the Computer Industry Coalition on Advanced Television Service (led by Intel), met and worked out an agreement which they presented to the FCC for approval on November 25.
While the broadcast and computer industries could agree on most of the ATSC proposed standard, the major point of contention was with the proposed scanning format: the broadcast industry's proposal endorsed the interlacing format for video scanning, while the computer industry backed the format known as progressive scanning format(also endorsed by the film industry due to its currently higher quality). Interlacing, which is the scanning format currently used by television monitors, today provides a resolution of 525 lines/screen, while progressive scanning, which is used by computer monitors, attains resolutions of 720 lines/screen. However, the interlace format, as used with DTV, will provide resolutions of 1080 lines/screen, significantly higher than is currently possible with progressive scanning techniques. It is the format endorsed by Gary Shapiro, head of the Consumer Electronics Manufacturers Association (CEMA), who told WAVE that the reason computer manufacturers were against a mandated interlaced format was due to the cost of adding interlace decoders to the monitors, which "costs about $30 now, and in two years will cost less than $15 to add."
Under the agreement, most of the original ATSC's proposal (which can be obtained at their Web site) is included with the exception of the transmission format: all 18 of the original transmission formats (14 of which were progressive and 4 of which were interlace) have been dropped, leaving each industry free to use whatever format it wishes, and leaving the eventual standard to be decided by market forces. In the near term, what this means is that televisions will probably accept both interlaced and progressive-scanned broadcasts (as well as others), while computers will continue to support only progressive formats, although Mr. Shapiro believe that eventually PCs "will have no choice but to support interlace as well."
If the standard is adopted by the FCC before the end of the year as expected, it will open the doors for the era of digital TV. The next step, which is expected to get underway early next year, is the allocation of digital channels to broadcasters. As explained to us by Mr. Shapiro, "the spectrum for HDTV are like holes in the swiss cheese of the current analog spectrum," with broadcasters each receiving a digital channel to simultaneously broadcast in both digital and analog formats. This dual broadcasting approach is expected to remain in place for 7 years, after which the analog spectrum will be returned to the government and auctioned off, at least according to the Clinton/Gore administration (although many in the industry, including Mr. Shapiro, feel that the concept of auctioning off the spectrum is outdated and won't happen again).
Both Digital Television sets as well as computers equipped with DTV receivers are expected to start appearing as early as spring 1998, at which time digital broadcasts will also begin, although according to Mr. Shapiro, "it will be at least 15 years until we have a totally digital [TV] world."
615.5 3DLabs - Notes from the Prospectus by John Latta
As of 30 September the company had an accumulated deficit of $4.1m. From its incorporation on 1 April 1994 to 31 December 1994 it lost $3.197m, for the year ended 31 December 1995 it lost $1.358 m and in the nine months ending 30 September 1996 it had net income of $485,000. For the same nine month period Texas Instruments, ELSA and AccelGraphics accounted for 19%, 19% and 13% of its total revenues. Again, at that same time, the company had 54 employees, 35 in engineering, 12 in sales and marketing and 6 in administration; 10 are in the U.S. and 44 in the United Kingdom. When the company was formed Creative Technology invested $3.5m and after the stock offering owns 17.6% of the company. Creative paid a $1m license fee for GLiNT technology, a $1m royalty in advance and $1m for the ASIC core (Gigi - 3D Blaster). Creative also has an obligation to pay a minimum of $10m in royalties through 2000. Texas Instruments paid $2m in license fees for PERMEDIA technology. At the same time the TI executed its license it made an investment of $5m. TI has the right to co-sell certain Permedia derivatives developed by TI through the year 2001. Intel made an investment of $2m in October 1996. As of 30 September 1996 the company had 15 US patents and 20 pending patents.
615.6 Multi-User VRML Proposed by David Lohse
Since it burst onto the scene in 1995, VRML has become one of the most hyped technologies around. VRML 1.0, which allowed static 3D worlds to be built on the Internet, was limited due to its lack of interactivity, and was seen by many as neat but somewhat boring. VRML 2.0 solved those problems by adding interactivity and animation to the standard, along with a slew of other new features. However, the VRML community at large realized that the most important key component of the VRML vision was still missing: multi-user capabilities.
Several companies have been tackling the multi-user problem by developing their own proprietary methods, including companies such as Worlds, Black Sun, and Oz Interactive, but such efforts have been mostly run in parallel instead of cooperatively. In order to move the technology forward and to create an open solution, on November 15, Mitsubishi Electric Research Laboratory, in conjunction with Chaco Communications and Velocity Inc. announced the proposal for a new multi-user VRML standard: Open Community. Open Community, which is based on the Spline (Scalable Platform for Large Interactive, Networked Environments) software architecture developed by Mitsubishi Electronics, provides a Java API for both the network and applications to create distributed data communications, allowing many interesting applications. As Dan Greening, the President and CEO of Chaco told WAVE, "Using the Open Community standard interface, you can create virtual worlds that "hyperlink" to each other in interesting ways. For example, you can look through a window in one world at buildings in another world, and you can hit a baseball 'out of this world' and have it land in another world."
In addition to proposing a fully open industry standard, Open Community has been designed to integrate fully with the other proposed standards currently under review, including Living Worlds (see WAVE #612, 11/8/96) and Universal Avatars. As explained to us by Mr. Greening, "Open Community and Living Worlds are compatible standards focusing on different things. Living Worlds is a VRML API for virtual worlds, Open Community has a Java API for general multiuser communication. Living Worlds requires 3D in some form, whereas Open Community considers 3D at the same level as 2D (and incorporates Living Worlds into its view of 3D). Living Worlds allows companies significant leeway to provide proprietary APIs in Living Worlds, whereas Open Community tries to standardize as many important features as possible."
The proposal, which can be found at MERL's Web site, is up for public scrutiny and will follow the usual "wait-an-see" approach taken by proposed VRML standards.
615.7 Company Profile - WK Software - Making 3D APIs Work by John Latta
Located in Sunnyvale, WK Software is both the consulting firm of Brian Hook and the culmination of 8 years programming experience in 3D. Brian has worked at Trident Microsystems, 3Dfx Interactive, Nvidia, and Silicon Graphics. As the architect of 3Dfx's Glide API Brian has considerable experience in making 3D work on the PC. Brian was interviewed by WAVE and provided some blunt views on the state of the 3D industry.
WK Software is at the center of an explosion of demand for 3D API programming skills. Areas where it works include porting of 3D titles and the design and implementation of 3D applications or APIs, with a focus on maximum performance and robustness. The goal of the company is to provide robust, fast, and, when applicable, portable software solutions. Some of the WK software projects include: Parallax Software Descent 2 port to Voodoo Graphics, design and optimization of the software rasterizer for the Windows 95 version of CosmoGL and the development of the T3D Hardware Interface Library, for Trident's family of 3D graphics accelerators.
Brian puts today's 3D chips into two camps - those with fast rendering and the rest. He calls the good chips, such as, PERMEDIA/Delta, Rendition Verite, and 3DFx Interactive as those which stress fill rate. At the other end of the spectrum at the "less expensive chips such as the ATI, S3, and Cirrus Logic offerings [which] tend to skimp on gates for faster triangle setup, so they have very low triangle throughput." One of the reasons that the Voodoo chip dominates the fast rendering segment of the market is because it is 2X to 4X faster than the others.
When asked when software only rasterizers will be a thing of the past Brian Hook felt that it would take 2 years. He also feels that "developers are getting tired of writing rasterizers and want to move on to working on The Next Big Thing instead of reinventing a faster wheel every 18 months."
AGP is very important to the future of the next generation of 3D according to Brian. Some of the chip features will be "... 2D/VGA/3D on them, and they will have performance near that of today's top performers. Key features will be included, including Z-buffering and bilinear filtering." He also comments on the realities of today's market forces "...since we have Darwinistic capitalistic forces working on the industry - - lack of a single item that singles you out as being less than full featured can be a killer in the marketplace. This affected Nvidia, which lacked Z-buffering and triangle support in their NV1, and Matrox, which lacks bilinear blending in their Mystique."
Brian feels that for a 3D API to be successful it must be easy to use and widely available. He has strong reservations about Direct3D because of its weaknesses in ease of use and robustness. He also feels that "...it is not intuitive, it is very error prone, it does not possess much in the way of help to debug faulty execute buffers, it lacks a conformance test for drivers, the software emulation layer is very minimal, it forces capability bits on developers, and...it doesn't offer manufacturers the ability to innovate with extensions." Brian showed concern about the documentation when he stated that "...no easy way exists to digest sample code, and it takes hundreds of lines to do even basic initialization. The capability bits maze is daunting and very error prone, and by its nature nearly untestable." (See the story on Microsoft's efforts to improve Direct3D which were announced at the AGP Conference - in this issue)
He reserves some of his strongest comments about benchmarking. Benchmarks fall into two categories: those which measure performance of an application or are "application like" or those which measure one exact performance parameter. In the latter case one must isolate the one aspect of performance and measure it closely. The problem is that "different manufacturers optimize for different things. So if you do a rasterization fill rate benchmark, Company A may love it but Company B will think it's 'unfair' because it doesn't test CPU load, overlap, texture download speed, or it tests some feature they don't support well, such as bilinear filtering, z-buffering, RGB lighting, etc." He continues his assessment "So in the end everyone but the winner thinks the benchmark is bogus. The benchmark Andy Bigos and I developed for Game Developer magazine tests pure rasterization performance, and companies complained about things ranging from 'this doesn't test overlap', 'Z-buffering isn't indicative of real world apps', 'you're not testing the features people will use tomorrow', 'white lighting is far more prevalent than RGB lighting', 'you're not testing texture download speed', 'you're not testing CPU load', 'you don't test geometry', 'this isn't indicative of a real game's, etc., etc."
In spite of his enthusiasm for 3D and involvement in the industry Brian is reserved on its impact on the PC industry as a whole "...because I haven't seen a wide market 3D application except games...When I start seeing 'killer apps' that leverage 3D I'll believe it's important, but until then I think it's primarily important for games and vertical market applications, not computing in general."
For the latest article by Brian Hook look at Game Developer, December 1996/January 1997 issue: "3D Hardware Acceleration Demystified, Part I."
615.8 Apple's Enabling Technologies Move to Windows by David Lohse
Apple's QuickDraw3D, the perennial underdog in the 3D API wars, has made a major comeback with the recent release of version 1.5. Having received a "Best of COMDEX" award at last month's show, QD3D 1.5 is supported on both the Macintosh and Windows platforms. With 99% of the function calls on the two platforms identical, QD3D offers perhaps the most well-rounded cross-platform functionality found in a 3D API yet.
With growing industry enthusiasm behind it, QD3D's main strength's lies in its open architecture and inherent extensibility (for a reference diagram of the QD3D architecture on Windows, check out our page: http://www.fourthwave.com/apple.htm), which allows developers to use plug-in rendering engines and shaders. The architecture includes the promising RAVE hardware abstraction layer, which provides 3D hardware acceleration as an integral part of QD3D or as a stand-alone engine. QD3D also brings a unique approach to 3D geometry by offering support for over 24 geometry types beyond the standard polygon-based API, including six new types in version 1.5 including ellipsoid, cone, cylinder, torus, trimesh and polyhedron.
Under the promise of a potentially superior technology, industry support for QD3D on the Windows as well as Macintosh platform has been steadily growing, and it has been endorsed by more than 25 hardware and software vendors. The major software vendors that have shipping or announced QD3D for Windows products include Specular, with Infini-D 3.5 the first application to support QD3D on the Windows platform (see WAVE #609, 9/27/96), Strata and Macromedia,. Hardware support has also been growing with a number of leading vendors, including 3Dfx, 3D Labs, ATI, S3, S-MOS and Cirrus Logic, also currently have or will soon offer QD3D-based hardware.
QuickDraw3D 1.5 for both Macintosh and Windows is available from Apple's QD3D Web site, as well as a QD3D Cross-Platform Parser.
Apple is also making a major move forward in the PC multimedia arena with QuickTime 2.5 as a part of their QuickTime Media Layer (QTML) initiative (see WAVE #614, 11/25/96). Already an important client technology on Windows, especially in regards to the Internet (it's the second most popular plug-in downloaded currently, after Shockwave, which Macromedia claims has been downloaded over 12 million times), Apple hopes to establish QuickTime as the premiere platform for multimedia development as well, making it "the first multi-platform, high-performance open architecture for manipulating and distributing high-quality digital video on Windows 95 and Windows NT."
In the spirit of creating an open architecture, Apple is working closely with 8 companies as co-developers of QuickTime 2.5 for Windows, including Adobe, Broderbund, Cinebase, Equillibrium, Intergraph, Macromedia, Media 100 and Truevision. These companies are collaborating on the key components of QT 2.5, including video frame-grabbing, NT multi-processor support, memory management, media compression, and decompression and high-performance playback. In addition, 47 hardware and software vendors have endorsed QT as a multimedia development environment on the Windows platforms.
Work on QuickTime 2.5 for Windows was begun approximately two months ago, and is now in alpha testing.
615.9 E&S Debuts Universal 3D Architecture by David Lohse
This week Evans & Sutherland rolled out their "Universal 3D Architecture" at the I/ITSEC trade show, at which they demonstated Harmony, the high end of the family (at the desktop end of the family is REALimage, which was introduced in August at SIGGRAPH and is an integral part of the forthcoming 3DPro from Mitsubishi/Vsis). According to E & S, the technology brings a standard 3D architecture across multiple hardware platforms from real-time image generators to NT workstations (with the NT platform being particularly targeted). The Universal 3D Architecture is based on industry-standard technologies, including four key components: the Intel Pentium Pro processor, the Windows NT operating system, the OpenGL API and the PCI standard.
Harmony, which E & S describes as "a sophisticated real-time image generator" for the Windows NT platform, is targeted at high-end applications such as flight and ground vehicle simulation, digital television and content creation. Features of the Harmony system include texture sharpening on a pixel-by-pixel basis, Phong shading, bump mapping, an optimized depth buffering architecture, and advanced video output. Harmony also boasts very impressive performance: E & S claims that it can achieve peak rates of 10M polygons/sec and 1.2B pixels/sec (with sustained fill rates of 720M pixels/sec).
Residing at the high end of the market, Harmony also commands a high-end price point, with systems *starting* at $150,000 (compared to REALimage boards, which are expected to retail for around $2,000). Harmony systems will be available by mid-1997, and other members of the product family are expected to be released throughout the year.
Copyright 1996 4th WAVE Inc.
May be redistributed in full for individual readership and posted to newsgroups, Web, and FTP sites. May not be reprinted or redistributed for profit. Short quotes are permitted but must be attributed to the WAVE Report on Digital Media.