ATI RADEON 256 Preview

by FastSite
04/25/2000 | 12:00 AM

In March we posted an early preview of a new GPU from ATI known under the code name Rage6. Today the curtain of mysteryhas been finally raised and we are ready to introduce in all its glory a newly born offspring of a well-known Canadiancompany - ATI.<%BANNER[article]%>

As we have expected, a new chip from ATI got a totally new name - RADEON 256. Well, ATI gave up the idea of continuingthe Rages family, and we really approve of this decision. In fact, there have been the whole bunch of chips launched underthis trade mark, both - successful and not quite, which pushed their way through the market. Moreover, RADEON 256 can boastquite a lot of absolutely new features that is why a change of a brand name is hundred percent justified. Of course, you canargue whether ATI made the right thing when they called their baby RADEON 256 and mock at the logo, but nothing to do aboutit: the name is chosen and the logo is drawn. The best thing we can actually do is to put up: anyway, we will get used tothem one day...

We decided not to edit anything in our early RADEON (Rage6) preview, so that you could compare our today's info withwhat we had posted before. Besides, this approach will help us avoid repetitions and concentrate mainly on the mostinteresting and worthy details.

No doubt, all of us long for a modern graphics accelerator, which will serve us faithfully for a year or so. To meetthese requirements the graphics processor should have a certain set of features, some of which are just a must for now.First of all, it should demonstrate high performance in 32bit color in the today's games as well as in the developingones. It is typical of all contemporary (and future) games that they work with an enormous amount of polygons duringthe surface tessellation (i.e. complex objects are built with a set of polygons, which means that the more polygonsare used, the more lifelike is the shape of the object), support scenes with high depth complexity, make good use ofmultitexturing and various lighting technologies. In the first place, we should mention the rendering Charisma Engineand Pixel Tapestry rendering architecture. Besides the graphics part, the developers of RADEON 256 also focused on videoplayback and LCD monitors support, which is so popular today. RADEON 256 features an integrated TMDS transmitter and hencesupports monitors with digital interface. As for video, RADEON 256 boasts not only HDTV formats playback but also a uniquevideo image improving technology - Adaptive de-interlacing.

So, let's take a look at the key features of RADEON 256 GPU:

General Characteristics:

• 256bit graphics core made of two 128bit units
• 0.18 micron technology, which s going to change to 0.15 micron and then - 0.13 micron
• 100sq.mm core size
• 30 million transistors
• 2 rendering pipelines, with 3 texturing units each
• 200MHz core working frequency (as far as we know, the core architecture allows up to 400MHz)
• Up to 200MHz memory bus working frequency
• The supported memory types include: DDR SDRAM (DDR FCDRAM is likely to be supported in the future) and a standard SDR SDRAM/SGRAM
• 8-128MB local graphics memory
• Supports HyperZ technology, which allows increasing graphics memory bus bandwidth by over 20%. HyperZ -Color/Depth Mixed mode 12/32, offers you the ability to select the depth buffer size independently of the colordepth. For example, in 32bit color you could select a 16bit Z-buffer. Leads to performance increase.
• 6.4GB/sec memory bus bandwidth (200MHz peak working frequency), with HyperZ technology - up to 8GB/sec
• Pixel fillrate: 400 million pixels per second
• Texel fillrate:
  • 1 texel per 1 pixel: 400 million texels per second
  • 2 texels per 1 pixel: 800 million texels per second
  • 3 texels per 1 pixel: 1200 million texels per second
• The chip has a texture, pixel and vertex caches
• 350MHz RAMDAC integrated into the graphics core
• Max display resolution - 2048x1536 at 75Hz
• Integrated TMDS transmitter supports digital monitors and the resolutions up to 1600x1200
• MAXX technology support, i.e. there can be two RADEON 256 chips installed onto one and the same card
• External bus interface: full AGP 4x/2x (including SBA, DME and Fast Writes) and PCI 2.2 (including Bus Mastering).AGP 4x Fast writes allows direct data transfer between CPU and GPU at up to 1GB/sec avoiding the PC system memory. Itshould undoubtedly have positive influence on the general performance and help to somehow reduce the system memory bus utilization.

3D Acceleration Features

• Charisma Engine: hardware Transformation and Lighting, clipping (deletion of all the scene details beyond the endframe), vertex skinning, keyframe interpolation, perspective divide, triangle setup
• Supports 4 skinning matrices used for vertex blending
• HW TCL capacity (peak): about 30 million textured polygons per second
• 8 source hardware lighting for the whole scene (directional aka infinite and point Lights aka local)
• Full OpenGL and DX7 support - Transform and Lighting, Cube environment mapping, projective textures and texture compression
• Pixel tapestry rendering engine, which allows using 3 texturing units in each pixel rendering pipeline
• Supports Cubic, Spherical and Dual-Paraboloid texturing methods
• Rendering in 16 and 32bit color depth mode
• Hardware bump mapping support. Embossing, Dot Product3, EMBM
• Supports the textures of 2048x2048 @ 32bit
• Programmable texture blending modes
• Supports 3D textures and hence some special 3D effects, such as fog or dynamic lighting, e.g. fire burning in the fireplace
• Supports hardware effects, such as: Motion Blur, Depth of Field, Full-Scene Anti-Aliasing (FSAA), etc. via D3D8
• 8-bit stencil buffer
• 16/24/32bit Z-buffer
• Shadowing for each separate light source with the help of a special Priority Buffer
• Full Direct3D lighting model support
• Table and vertex fog

Video

• Supports hardware decoding of all HDTV formats
• Supports all ATSC resolutions including 1080i
• Due to YPrPb support allows connecting HDTV displays directly
• Supports Adaptive de-interlacing. This unique technology from ATI allows playing high quality videowithout any artifacts or blurred effect
• 8bit alpha-blending for video and graphics (can be used for subtitles or animated menus)
• Fully compatible with ATI Rage Theater companion chip

And now let's dwell on a couple of most interesting details of a new chip and new ATI's solutions. Thescheme below shows RADEON 256 architecture:

With the help of this picture you can easily learn everything we could miss out.

Charisma Engine

The geometric Charisma Engine possesses the following features:

Hardware vertex lighting support. The performance drop is expected to be less tangible compared to GeForce 256 in caseof more than one Directional Lights (aka Infinite; when the light source is defined as a point located at an infinitedistance from the visible scene objects).

Hardware transformation of 3D vertex coordinates of the modeled scene into 2D screen coordinates taking into accountthe distance correction. This operation is called perspective transformation.

There is also hardware Clipping support: deletion of those polygons, which are invisible to user in the end scene.

Although GeForce 256 supports hardware Clipping, the name of its geometric coprocessor is usually abbreviated to HW T&L.Today, more and more often you may come across the following abbreviation used for geometric transformation and lighting unit -HW TCL, which stresses the hardware clipping, although the former abbreviation "HW T&L" is not in the least worse.

The promised performance of RADEON 256 HW TCL unit should reach 30 million polygons per second. If we keep in mind that theperformance of GeForce 2 GTS is promised to be around 25 million polygons per second, then we have every right to regard thegeometric unit of RADEON 256 as the fastest of all in computer industry till autumn (at least theoretically).

Charisma Engine also supports some special operations, such as Vertex Skinning and Key-Frame Interpolation.

Hardware Triangle Setup support implies transformation in the frame buffer of floating point triangle vertex coordinatesinto integer coordinates as well as setting other vertex parameters, such as textures linking.

Vertex Skinning

This method deals with correct transformation of geometric grid in model joints, i.e. we work with pure geometry and not withtextures. If the skeleton is correctly transformed, the textures will cover it properly. In order to ensure correct transformationof the object's geometry, especially in case of joining and bending parts (take for instance all joints of a human body) vertexskinning is applied (single-skin effect). Note that vertex skinning is a single case of a more general vertex blending technology.In fact, vertex blending works in the same way as alpha-blending, however, it deals with vertices instead ofpixels (V = V1*alpha + V2*(1-alpha)). We would like to stress once again that the textures will be laid properly over skeletaljoints only if the triangle vertices are correctly located, which is possible due to vertex skinning. Vertex Skinning is usedto provide smooth and natural picture especially if the object is moving. Vertex blending is carried out with Skinning Matrices.

D3D supports 2- and 3-matrix vertex blending. In most cases 2-matrix blending is more than enough. However, when a neck andtwo arms are jointed to the body, 4-matrix blending will be more preferable, and in case of a body and two legs - 3-matrix one.Nevertheless, you will be able to notice the quality improvement only if you attentively examine a full-screen man. GeForce 256,for instance, supports only 2-matrix blending, while Rage6 will also have 3- and 4-matrix blending. If you are looking for someillustrations of 2-matrix blending, take all the living creatures from DMZG demo, which makes full use of this effect.

Keyframe Interpolation

This technology allows changing the object's appearance on the hardware level by setting only the initial, the final anda few key transitional images: all the other transformations are carried out automatically. In other words, Charisma Engineinserts the required amount of interpolated frames between the key ones, which changes the character's facial expression in avery lifelike way. Take a look at the animated picture on the right. This is good example of keyframe interpolation:

These screenshots were taken from a special demo program. The parameters for the changing facial expression were provided inan ordinary text file and left a lot of room for experiments. The initial picture is provided with a few sets of facial fragmentsin TGA format. Actually, keyframe interpolation is rather easy to use that's why it is very likely to appear in real gamesquite soon.

You will hardly argue that TCL operations as well as Vertex Skinning and Keyframe Interpolation require huge computingresources. However, RADEON 256 seems to have no problems with that. As a result, game developers will be able to use morepolygons for tessellation and apply the whole bunch of various special effects, which will make the 3D gaming world muchmore lifelike.

So, game developers will be able to use the uninvolved CPU resources for AI calculations, for example. Just for a clearerpicture: the games without HW TCL usually have about 5% of the CPU resources for AI, while new games supporting HW TCL will beable to boast 30-40%, which will make your virtual enemies much more inventive and cunning (we can hope at least that it willbe so). Besides, the freed CPU resources can be also utilized to calculate complex physical effects, such as gravitation andatmosphere phenomena. We should also say that if a game supports HW TCL but does not utilize the available CPU resources, youwill still get very fast graphics of very high quality, although you won't notice any other improvements.

Pixel Tapestry

Besides the possibility to speed up via hardware the geometric stage during image rendering, RADEON 256 can undoubtedlyaccelerate rastering as well. This is exactly the thing pixel rendering architecture - Pixel Tapestry engine - is aimed at.This rendering engine is built into RADEON 256 and serves to provide proper functioning of all three texture units of eachrendering pipeline. RADEON 256 features two rendering pipelines with three texturing units each, which allows:

• blending and filtering up to three textures per pixel at full speed
• supporting on the hardware level versatile texture transforming methods (cubic environment mapping, projective texturing, etc.)
• executing single-pass Embossing, Dot Product 3 and Environmental Mapped Bump Mapping (EMBM)
• modeling accurately the reflective properties of materials (water, metal, wood, etc.)

Besides, Pixel Tapestry provides hardware support of:

• Priority Buffer used for realistic shadow mapping from individual light sources
• 3D textures creating complex dynamic lightmaps, as well as volumetric fog, smoke and fluids and simplifying handling theobjects with deformable geometry

You know that the more textures are laid over one pixel per clock, the more realistic and detailed turns the final image withoutsacrificing the general performance. Textures are used for laying lightmaps, shadow maps, specular maps, gloss maps, bump maps, etc.

We should also point out EMBM support, which is now featured only by G400 chips from Matrox but is already applied in a few tensof games.

A very important feature of RADEON 256 is Cubic Environment Mapping supported in D3D 7.0. This method is viewpoint independent,but uses 6 textures instead. Besides that, RADEON 256 also supports Spherical method (1 texture, viewpoint dependent) andDual-Paraboloid method (2 textures, viewpoint dependent).

There is a special technology used to make a scene look more lifelike and to create scene depth - Shadow mapping. Verylifelike shadows depending on the distance between the object and a light source can be obtained with RADEON 256 due to aspecial Priority Buffer. This buffer is made of the local graphics memory resources and accumulates information about theobjects casting shadows. Priority Buffer helps to define the order of calculations that is why the shadows start formingfirst for the closest object to you. This method seems to be not only much easier but also much more efficient than theordinary stencil buffer.

RADEON 256 supports Range-Based Fog, i.e. the fog density depends on the distance between the object and the player.

One of the newest functional features of RADEON 256 is 3D textures support. 3D Texture is a 3D pixel array and not a 2D oneas usual. This is a totally new thing for OpenGL 1.2. Some time ago Permedia3 from 3Dlabs was the only one supporting thetextures of this type. 3D textures are perfect for voxels and spherical textures used to create dynamic globs of light. 3Dtextures simplify the use of deformable geometry. Parts of objects can be chipped or cut away without having to updatetextures.

RADEON 256 is already prepared for a couple of innovations of the near future, which will be supported in the upcoming D3D8from DirectX 8.0 expected to be announced by the end of the summer. Here we should certainly mention Motion Blur and Depth ofField, as well as Full-scene Anti-aliasing. Sounds familiar? Of course! In API D3D8.0 we will see 3dfx T-Buffer effectssupport.

Frankly speaking, RADEON 256 press release says the following thing about FSAA: "Comprehensive support for full scene,order-independent anti-aliasing". In fact, we should be ready to experience a quite tangible performance drop (the performance drops at least twice)with FSAA enabled. However, it is still up to users to decide: to use or not to use...

In fact, the situation with Pixel Shaders is not quite clear. Pixel Shaders is a method of developingmultitexturing effects "manually". It will be supported in DirectX 8.0. The matter is that some time agotexture blending was allowed only according to certain previously defined schemes (as a rule they includedone or two operations). But now Pixel Shaders serves as a "program" of textures blending manually defined(it includes the following operations: +, -, /, * and a couple of other more exotic ones; all of them canbe combined).

Pixel Shaders implies that the work on 3D graphics is carried out at the level of pixel modelinginstead of polygon modeling. It looks as if there were no special programmable and previously definedmultitexturing effects at the pixel level in RADEON 256. However, it is possible to control the blendingof texel units results. It seems quite enough for the following announcement in the press release: "First withadvanced DX8 pixel shader effects".

Besides the graphics part, which is of course of great importance to us, RADEON 256 also boasts some additionalintegrated features, which can be referred to video field and are known as Video Immersion. First of all, it impliesATI's traditionally beautiful hardware support of DVD video decoding including Motion Compensation and iDCT. RADEON256 represents ATI's best achievements in this field intended mostly for set-top box stations and electric appliances.That is why RADEON 256 will be easy to use in our everyday life for DVD, DTV, digital recorders, etc. And the possibilityto decode all HDTV formats and to connect analogue as well as digital displays without any extra effort, makes RADEON 256an excellent solution for HDTV watching via desktop PC.

But this is just the beginning. RADEON 256 also supports a unique Adaptive de-interlacing technology from ATI, whichallows high quality video playback with clear-cut images and absolutely no artifacts. Just take a look at the screenshots,and you'll get the point without any commentary:

        
From left to right: Bob De-interlacing (Blurred Text, Edges), Add Field De-interlacing ("Feathered" Image) and finallyAdaptive De-interlacing (Sharp text, edges No feathering)

Looks impressive, doesn't it? So, let's wait for the first cards and a corresponding ATI DVD player to come out and thenwe will be able to see everything with our own eyes.

Moreover, with a new Rage Theater companion chip installed onto RADEON 256, the user will be able not only to play thevideo but also to capture some fragments of the video stream and to encode them into various video formats.

Among the most pleasant features of RADEON 256 we should list the built-in TMDS transmitter, which allows connectingdigital monitors to a card built on this chip.

As far as we know, graphics cards on RADEON 256 will be announced in summer already and their mass supplies will startright after that. Besides, we should be ready for a family of graphics cards including a piece with two RADEON 256 chips onboard. You are probably wondering why they postponed the graphics cards launching? We believe that the root of the problemlies in the drivers and not in hardware, as you could think at first sight. ATI administration probably preferred postponinggraphics cards announcement for a month after the official launching of their chip rather than announcing the cards now butstarting the sales only in a month or maybe even later. Well, this was the right step to make: ATI seems to have alreadylearned its lesson from the previous experience.

On the photo on the right you may see a RADEON 256 based graphics card version. You should be able to notice that thereis Rage Theater companion chip already preinstalled onto the card and hence you can also see all the video Ins and Outs.Moreover, you can also find two transformers, which are responsible for stable power supply and protect the users againstthe problems of poorly made mainboards. Furthermore, the memory chips marking is also quite clear: DDR SDRAM (4Mx16) byHyundai. This memory is supposed to work at 183MHz. Nevertheless,we shouldn't forget that the official RADEON 256 based cards specs haven't been announced yet.

Of course, everybody can't wait to learn how expensive these cards will be and what performance they will show in today'smost popular games. As for the prices, we can only guess, what it will be like. However, by the time mass production of RADEON256 based cards starts, those on GeForce 2 GTS are expected to get tangibly cheaper. Nevertheless, we can easily predict thatATI will pursue a really aggressive pricing policy, which is only to our, users', advantage.

As far as the performance is concerned, all the previous products by ATI give us every reason to believe that this timethe chip will also prove beyond our expectations in 32bit color and will definitely make us happy with 60fps at 1024x768.

Conclusions

Well, now it's high time we drew the final conclusion. A new GPU RADEON 256 from ATI features the whole lot of reallyinteresting functions, some of which are just unique. Charisma Engine is also promised to show impressively high performancein geometry calculations. And if we assume that about 50% of the coolest games of the year 2000 are expected to supportHW TCL, then RADEON 256 will hardly remain undemanded. Fast memory used, namely DDR SDRAM, will let graphics cards basedon RADEON 256 turn a really threatening competitor in the today's gaming market. Pixel Tapestry architecture guaranteeshigh quality 3D graphics. Besides, bump mapping (EMBM) support is also a valuable thing to mention, because thanks toMatrox this technology is inevitably spreading all over the gaming world. All in all, there shouldn't be any problemswith 3D graphics quality, and now everything depends on the driver developers. Honestly, RADEON 256 meets all therequirements for a number one gaming graphics chip.

According to our sources, RADEON 256 core architecture is intended for up to 400MHz clock frequency (the frequency ismost likely to rise over 300MHz only in case they shift to 0.15 or even 0.13 micron technology), which will allow ATI tolaunch a new chip version by the end of the year. However, we shouldn't disregard MAXX technology as well. So, altogetherthe future looks quite promising. ATI will hardly sit idle this year.

Perfect DVD video decoding support and an optional Rage Theater companion chip will let you turn your ATI RADEON 256into a universal multimedia combo-machine. Actually, not all people on Earth are crazy about playing Quake-like games.Some guys may be fond of home video archives and the like.

So, summing up, we can say that ATI RADEON 256 seems to be a well balanced gaming solution for the year 2000. If RADEON256 based graphics cards appear this summer and are available in sufficient amounts, every single user will be able to getone of those for his home monster. Of course, this will be true only if ATI asks for a reasonable price. And if ATI looksforward to winning a bit of NVIDIA's empire, they will have to pursue a really aggressive marketing policy.

As for us, the best thing is to have patience and wait... wait for cards on RADEON 256 to make their first steps towardssuccess.