Matrox Parhelia-512 : The Technology
	By CPU-zilla.
	Category : Graphics
	Approved by Jimmy Tang on Tuesday, 14th of May,2002

Parhelia Specifications Matrox Parhelia-512 Technical Specifications General Features & Characteristics 512-bit GPU 80 million transistors Based on 0.15-micron technology process 256-bit DDR memory interface Up to 256MB DDR unified frame buffer 10-bit Gigacolor Technology 10-bit per channel RGB rendering and output over one billion simultaneously displayed colors 10-bit precision for 2D, 3D, DVD and video 10-bit frame buffer mode for ARGB (2:10:10:10) 10-bit RAMDACs with full gamma correction AGP 8x with Fast Writes support 8-way parallel DMA streaming engine OpenGL 1.3 and DirectX 8.1 compliant 3D engine High Fidelity Display Engine DualHead-HF Display Technology Fourth-generation DualHead Dual integrated 400MHz 10-bit RAMDACs Dual independent RGB outputs Up to 2048x1536 @32bpp on each RGB output Support for two digital TMDS transmitters Dual independent DVI outputs (up to 1920x1200 on each output) Single dual-link DVI output (up to 2560x2048) Integrated 10-bit high-fidelity TV/video encoder NTSC/PAL output Direct encoding of native interlaced YUV TripleHead Desktop Support for 3rd RGB output Three display desktop at up to 3840x1024 @ 32bpp 10-bit gamma correction Per-layer gamma and color correction at full speed Dual independent, gamma correctable hardware overlays Support for true multi-display under Microsoft Windows 2000/XP Hardware accelerated multi-screen OpenGL support UltraSharp Display Output Technology Highest-quality analog, digital and TV output Ultra-crisp display quality at high frequencies Highest-quality design, electronics and filters 5th-order output filters Highest-fidelity frequency and transient response for optimal signal quality High signal-to-noise ratio (SNR) with super-low PLL Jitter High Fidelity 3D Engine Quad Vertex Shader Array Four vertex shader units (DirectX 8.1 and beyond) Highest sustained complex vertex shader performance Parallel processing of up to 16 vertices 512 instruction on-chip cache 256 constant registers Quad texturing per pixel, per clock cycle 64 Super Sample Texture Filtering Highest quality trilinear and anisotropic filtering 8-sample anisotropic and trilinear filering on 4 dual-textured pixels/clock 16-sample anisotropic filtering on 4 single-textured pixels/clock 36-stage Shader Array 4 pixel pipes 4 texturing units per pixel pipe 5 pixel shader stages per pixel pipe support up to 10 pixel stages per pass 4 pixels/clock throughput with quad texturing and 5 pixel shader operations Hardware Displacement Mapping Patent-pending Depth-Adaptive Tessellation for continuous level of detail (LOD) geometry Vertext Texturing for dynamic generation of geometry using texture maps Support of Bezier curves and N-patch (PN-triangle) evaluation Surround Gaming Support for games rendered across three displays Ultra-wide field of view (FOV) GigaColor Gaming 10-bit source texture support and precision High-precision ARGB (2:10:10:10) frame buffer 16x Fragment Antialiasing (FAA-16x) 16x supersampling quality on edge pixels only Avoids blurring of internal pixels Low performance overhead Support for Full Scene Antialiasing (FSAA) Texturing Support Support all texture formats including :- 32-bit source textures 10-bit per channel texture support All DXTC formats 2D, 3D (volume) and cubic textures Non-square and non-power-of-2 textures Planar and packed YUV textures Up to 2K by 2K source textures Support for projected textures Support for texture swizzling Render-to-texture support Other 3D Features include:- Depth acceleration unit for advanced Z processing 32-matrix Matrix Palette Skinning (MPS) Particle acceleration Full sub-pixel and sub-texel precision Environment Mapped Bump Mapping (EMBM) and DOT Product-3 Planar, cubic and spherical environment mapping Fogging, alpha blending and specular highlighting Flat and gouraud shading Independent intensity, Z and texture depths Antialiased 3D vector support High Fidelity 2D Engine GigaColor Desktop All drawing operations at extended 30-bit color (10:10:10) 10-bit per channel frame buffer High-quality dithering for lower bit depth output Glyph Antialiasing Hardware accelerated text antialiasing Programmable gamma correction Full acceleration of Windows XP GDI and DirectDraw functions GDI+ v2.0 ready Programmable, ultra-fast bliter at up to 16 pixels/clock True-color full-screen overlay plane with 8-bit alpha Alpha cursor support 32-bit ultra-fast VGA core High Fidelity Video Engine PC Theater DVD Playback 10-bit DVD playback 10-bit advanced filtering and scaling 10-bit DVD output via TV encoder Independent gamma and proc-amp controls Programmable overlay processor Video overlay with programmable proc-amp and independent gamma correction Video mixing engine in overlay processor High-quality horizontal and vertical scaling Up to 4x4 filter kernel with programmable filtering coefficients Full-speed bi-cubic filter Fully VMR-complaint front-end scaling Advanced de-interlacing with sub-pixel positioning VIP2.0 compliant video input port Compliance Operating systems - Microsoft Windows and Linux Platforms X86, X86-64 and IA-64 compatible AMD 3DNow!, MMX, Intel SSE & SSE2 optimized AGP 8x, 4x, 2x and 1x Compliance PCI 2.2, AGP 2.0 and AGP 3.0 PCI Bus Power Management 1.1 ACPI DirectX 8.1, PS1.3, VS1.1, VS2.0 OpenGL 1.3 DirectX VA, VMR, WDM Click the image above to see the full Parhelia chip block diagram.

DualHead to TripleHead When Matrox launched the Millennium G400, Matrox attracted many PC users as it gave new meaning to working efficiently. Dual displays were not common at that time and the G400 gave users a new dimension in computing with its ability to display graphics simultaneously on two monitor. However, all that is becoming too common today as other graphics manufacturers have, in one way or another, managed to integrate a secondary display in their new GPUs and graphics cards. The new Parhelia will break that trend now with its new TripleHead feature. The new feature does not only allow one to span their desktop display to three independent monitors, but it also let 3D games render graphics across all the three monitors at the same time. This new technology is called Surround Gaming. The TripleHead desktop in action. In Surround Gaming, 3D scenes will be rendered on two additional monitors, one on each side of the user. Giving users an ultra-wide field of view will greatly enhance and enrich game play with its peripheral viewing capabilities. Imagine playing a first-person shooter game (like Quake 3) while having the advantage of spotting the enemies hiding in the far corner of your eye. All this is possible with this technology. We see this as a terrific support for future 3D games as it not only brings a new level of realism to gaming, but it is here to set a standard in future gaming as well. Chip Hinnenberg of LucasArts' Entertainment Company LLC said, "The Matrox Parhelia-512's Surround Gaming feature gives games new life with its immersive peripheral view capabilities. This is the wave of the future and the way games should be played." Microsoft Flight Simulator 2002 in action. Click the picture to see the difference with and without Surround Gaming. Many other gaming developers agreed too and developers have pledged support for this cool new feature. "The Parhelia-512's Surround Gaming feature could significantly enhance the Massive Multiplayer experience by allowing the user to move many space-intensive HUD items off the primary rendering screen and onto one of the two alternate screens. Having a whole screen dedicated to chat, another dedicated to the 3D world view, and a third for maps, inventory and stats would be wonderful," Jeff Grills, technology director, Sony Online Entertainment, currently working on Star Wars : Galaxies. Quake3 Arena in action with three monitors. Click the picture to see the difference with and without Surround Gaming. Surround Gaming currently supports quite a number of games and its out-of-box support for a wide variety of popular games will make this feature widely acceptable. Developers need not be bounded by developing games specific for this feature. According to Matrox, it works on most games and true enough, they demonstrated TripleHead on popular games like Quake 3 Arena, and the recent Jedi Knight 2 : Jedi Outcast. Our GameAxis game reviewer (guess who?) showing off his skills in Jedi Knight II. With the TripleHead desktop mode, business users or professional graphics designers will once again enjoy an expanded view of their desktop. Let's just say the trend will no longer be a single display for multiple windows, but rather, multiple windows on multiple display screens. The TripleHead feature will produce clear and crisp video output at up to an analog resolution of 3840x1024 at 32-bit color. Of course, DualHead will still be supported and you can still get up to 2048x1536 of display resolution on each analog RGB monitor. If you intend to use digital flat panels, the Parhelia-512 will give you up to 1920x1200 of resolution on each digital display - that's more than enough to fit the next generation of ultra-wide LCD TFT screens. This is how you'll be wiring up all three monitors to the card.

Quad Vertex Shaders The new Parhelia-512 integrates four parallel DirectX 9 compatible vertex shader units into a single vertex processing array. With an array of four vertex shaders, the Parhelia-512 is able to sustain very high performance levels even when it is executing very complex vertex shader programs on high-resolution 3D models. The Parthelia is powered by a rather sophisticated control unit, with deep instruction and register caches. There are a total of 512 instructions in the cache and 256 constant registers with 16 vector input parallel registers. The Quad Vertex Shader array is measured to give up to 100+ Megatransforms/second. Compared with other cards currently available right now, the Parhelia boasts of the highest number of vertex shader units and it supports the latest Vertex Shader version 2.0. Other GPUs like the GeForce3, GeForce4 and Radeon 8500 supports only up to version 1.1. Vertex Shader Comparison with other GPUs Chip Vertex Shader Version # of units Parhelia 2.0 4 GeForce3 1.1 1 Radeon 8500 1.1 1 GeForce4 1.1 2 The Quad DirectX 9 Vertex Shader Array block diagram. Quad Texturing The Parhelia-512 is the first GPU to integrate quad pixel quad texturing feature (4x4 texturing). This simply means that there are now four pixel pipes (similar to both GeForce4 and Radeon 8500) and in each pipe, there are four texture stages per pipe. Other GPUs like the GeForce4 and Radeon 8500 only featured two texture stages in each pipe. Thus, the Parhelia-512 can render four textures on four pixels simultaneously in a single pass. Unlike other GPUs, when there are more than 2 textures to be rendered, another pass needs to be performed to complete the texture rendering. On the Parhelia, the rendering completes in one clock cycle - shortening the time by half. This greatly enhances performance throughput especially when the game requires the GPU to perform complex quad texturing. The Parhelia can render four textures in four pixel pipes in a single pass. 36-Stage Shader Array In each of the four pixel pipelines, the Parhelia-512 features four programmable texture stages and five pixel shader stages. This adds up to a total of 36 shader stages. According to Matrox, this gives the Parhelia the power to render some of the most demanding 3D images with unprecedented performance. Other GPUs could only offer up to 16 shader ops, less than half of what the Parhelia offers. However, the question is whether 3D games today would actually take advantage of this massive shader array. Maybe not today and you would certainly not see the difference in performance. Perhaps in the future when games get more realistic, the answer is a big YES. But before that happens, you can be pretty sure that the Parhelia-512 is armed with enough processing power to render next generation games. The 36-stage shader array in the Parhelia-512.

Hardware Displacement Mapping Perhaps one of the biggest feature introduced in the Parhelia-512 is Matrox's revolutionary Hardware Displacement Mapping (HDM) technology. As the first to develop the hardware implementation of HDM, Matrox has contributed elements of this technology to Microsoft for inclusion into DirectX 9's API as a standard feature. Thus, in the upcoming DirectX 9 release, you will be able to see games making use of this new feature and similar to Hardware T&L, it could very well become a standard requirement for games. To quote from the Matrox white paper, "Theoretically, displacement mapping is a technique that uses displacement maps to displace the vertices of a mesh. The resulting displaced mesh takes the shape represented by the displacement map." Thus, in order to displace a mesh surface, a 2D bitmap is used and the individual pixel values in the displacement map are height values defined by the shape of the object. When the 2D bitmap is mapped onto the mesh, the surface of the mesh is displaced to form the shape of the object. Hardware displacement mapping is divided into three stages in order to produce a smooth and detailed mesh that can be textured or shaded using vertex and pixel shaders. The Hardware Displacement Mapping process. Click for a bigger picture. In the first stage, a base mesh with low triangle count is sent to the GPU. The GPU then tessellates the base mesh to increase its triangle count. Increasing the triangle count will help smoothen the displaced surface and adds more detail to the shape. Matrox has also implemented an approach to tessellation and it's called Depth-Adaptive Tessellation. This is a patent-pending technique that allows varying levels of detail (LODs) tessellated across a mesh. By varying the LODs, much of the processing in the GPU can be relieved as there will be less triangles to process. Depth-Adaptive Tessellation explained. Click for a bigger picture. In the second stage, the position of each vertex of the tessellated mesh is displaced with the displacement map. The pixel values determined by the map will be translated to height information and the shape of the object can be easily defined with this displacement map. This displacement map can be applied to both static base meshes and dynamic, skinned base meshes. This gives the feature a lot more flexibility. In the third and final stage, the displaced mesh will be treated and rendered like any other vertex stream. This means that vertex shaders, textures and pixel shaders can be simply applied to the mesh to create the final rendered image. Alright, now that you know how hardware displacement mapping works, what's the advantages? The advantage lies in its versatility. Firstly, it is hardware accelerated. This simply means that complex geometries can be created without additional load to the CPU. In addtion, the technology helps to create surfaces that's much more realistic than bump-mapping. This will undoubtedly create 3D games with more breathtaking graphics than ever before. Already Matrox has confirmed that this feature has been adapted in one future game title. Amazing, even before Parhelia-512 is released. Displacement Mapping versus Bump Mapping. Secondly, Hardware Displacement Mapping works with static and skinned meshes. In other words, if used in skinned characters, very detailed 3D characters can be created. The beauty lies in its flexibility to use the same base mesh to render different objects based on different displacement maps. This makes HDM extremely easy to be implemented. Hardware Displacement Mapping can be applied to the same base mesh to create uniquely different characters through the use of varying displacement maps. As quoted from Randy Pitchford, president, Gearbox Software, "After seeing Matrox's Hardware Displacement Mapping technology in action, I got kind of giddy. The artists at Gearbox will go nuts for this technology!". Currently, Gearbox is working on Counter-Strike: Condition Zero.

16x Fragment Antialiasing (FAA-16x) Matrox's implementation of antialiasing comes with a different twist. Instead of fully applying antialiasing to the entire object, FAA-16x intelligently picks out the edge pixels of triangles and then applies 16x super-sampling on them. This leaves internal textures sharp and unblurred, thus giving the game a smooth yet sharp appeal. Since edge pixels only make up a small fraction of the total pixels in a particular scene, the FAA-16x rendering could be performed with very low performance penalty. Thus, the quality of 3D graphics offered by the Parhelia-512 would be very interesting indeed as this will surely add to the debate of who's better at antialiasing. No doubt each will have their own strengths and weaknesses. Without any antialiasing. With FAA-16x enabled. So, if you're uncomfortable with FAA-16x, you can still enable 4x FSAA (Full Scene Antialiasing). But judging from initial screenshots provided by Matrox, FAA-16x does seem to be pretty promising for now. According to Matrox, FAA-16x performance with anisotropic filtering far exceeds Quincunx, SmoothVision or Accuview. With FAA-16x, internal textures do not get blurred. Here's an example from Dungeon Siege. Click picture for a larger view. FAA-16x explained. Click the picture for a larger view. Glyph Antialiasing Glyph antialiasing is introduced not so much for the gamers but for those who wants to read text clearly. It is Matrox's answer to accelerating font rendering through the hardware. This is to take as much of processing away from the CPU as possible. Supported by modern operating systems from Microsoft such as Windows 2000 and Windows XP, Glyph antialiasing also supports independently programmable gamma correction for text. Glyph Antialiasing explained. Click picture for the full view. We think Glyph antialiasing will benefit TFT LCD monitors more as fonts will no longer look hard and cold on the LCD monitors. Antialiasing will soften the look of fonts and this will definitely make reading text on the monitor more enjoyable. Fortunately, only Matrox has actively addressed issues of text clarity and this will definitely make the Parhelia-512 more appealing for business use than any other cards.

Availability and Price Details about the availability of the Parhelia-512 are still a bit sketchy as there are still some more fine tuning to do. Skeptics say that it would be difficult for Matrox to produce the silicon since the die is potentially huge especially at only 0.15-micron with 80 million transistors. Skeptics aside, Matrox plans to ship the first version of the Parhelia-512 somewhere between June and July. They will come in a variety of boards with different memory sizes ranging from 64MB to 256MB. The exact product lines are still not finalized at this point of time but it seems that Matrox has plans to introduce products at different clock rates in different price ranges. Let's hope that Matrox will keep their product line cleaner than what the folks at NVIDIA or ATI did to their GeForce4 and Radeon respectively. With such features packed into the board, Matrox also plans to ship a separate set of boards which will target high-end 3D workstations. Looks like Matrox is all out to grab not only the gaming market, but high-end 3D workstation graphics as well. As for price, well, not much was told but they have indicated 128MB DDR versions to cost below US$400. The Parhelia-512 will certainly not be cheap, but if you're a serious gamer, the new Parhelia-512 does promise a lot more than just gaming performance. All in all, Matrox has once again surprised the graphics industry with a lot of fancy and powerful features. If they can pull through with enough yield to meet the demand, Matrox will definitely capture a large share of the gaming market next quarter provided that the card performs as well as the specifications promised. Today, Matrox is not only out to take the gaming market away from giants like ATI and NVIDIA, but Matrox is actively redefining the future of gaming. With Surround Gaming introduced in the Parhelia-512, there's no doubt that Surround Gaming will be the next big thing in gaming. Matrox deserves a good pat on the back for adding another level of depth and realism to our everyday computing life with TripleHead and Surround Gaming. Of course, the folks from the monitor industry should give Matrox an even bigger pat on the hind since each PC will be potentially accompanied by three monitors.

Top

 

	Table of Contents
	The Matrox Comeback Parhelia Specifications DualHead to TripleHead Performance Enhanced Hardware Displacement Mapping Antialiasing Revisited Availability and Price