The ATI RADEON 9800 and 9600 Series Discover how ATI varies a winning formula to create a family of graphics cards

The ATI RADEON 9800 and 9600 Series
Discover how ATI varies a winning formula to create a family of graphics cards

Introduction

After playing second fiddle to nVidia in the 3D graphics business for several years, ATI has become a winner with its Radeon 9800 series. Building upon the world-beating design of the 9700 series, the Radeon 9800 Pro is regarded by most observers as the best 3D graphics card currently available for desktop computers. However, at prices up to $450 for the 256MB DDR-II version, the Radeon 9800 Pro is too much graphics card for many users. Consequently, ATI has developed the Radeon 9600 series for mid-range customers.

In this article, youll learn about the major features of the Radeon 9800 series and how ATI simplified the 9800 design to create the 9600.

The R350, Son of R300, Graphics Chip

The Radeon 9800 series uses the R350 graphics chip, an improved version of the R300, which powers the Radeon 9700 series. The R350 has 110 million transistors, which is over 2.5 times the number of transistors found in a Pentium 4 processor! As you might expect from such a high transistor count, the internal design of the R350 is a complex one. I provide several diagrams in this article to help you understand the R350s design.

Here's a few fast facts about the R350 graphics chip:

• Fully supports AGP 8x
• Supports DDR and DDR-II memory
• Processes a vertex or triangle in a single clock cycle
• Fully implements all DirectX 9 features including version 2.0 vertex shaders
• Supports current and future OpenGL functionality

The Radeon 9800 series has three members: Radeon 9800 Pro, All-in-Wonder 9800 Pro and Radeon 9800. Table 1 compares the major features of each card.

Table 1 Radeon 9800-series Card Features Compared

*Includes personal video recorder software, TV listing software, translucent TV and DVD playback windows, remote control, dual TV tuner option, video editing software, optional support for HDTV output

Note that while all 9800-series cards support multiple displays, the AIW 9800 Pro lacks a second DVI-I or VGA port and thus cannot support two CRT or LCD displays.

With these differences noted, all three cards use the same architecture inside their R350 graphics chips. So let's take a look at whats inside.

Inside the Radeon 9800s Design

While the basic architectural diagram of the R350 chip at the heart of the 9800 series is impressive (see Figure 1), you need to look closely at each major component to see how ATI has used a divide and conquer strategy to handle processing complex image data more quickly and more accurately than ever before.

Figure 1

The basic architecture of the R350

To understand the amazing speed of the RADEON 9800, start with the AGP 3.0 (AGP 8x) interface. This interface enables the R350 chip to receive data at speeds up to 2.133MB per second, eight times the speed of the original AGP 1x specification, and sixteen times the speed of 33MHz PCI slots.

The Memory Subsystem

Support for AGP 3.0 (AGP 8x) alone isnt going to produce high graphics performance if the rest of the graphics subsystem isnt also designed for maximum throughput. The AGP port connects to a very fast four-way interleaved memory controller, which transfers data to memory through four 64-bit channels (see Figure 2). The memory controller uses a RAID-like striping methodology to transfer color data, Z-buffer data, and texture cache to the memory controllers. Just as RAID 0 speeds up hard disk access, this method improves read/write performance to and from video memory.

Figure 2

The 256-bit memory controller contains a switch to route data between memory channels and memory controllers.

The memory subsystem is just the first example of RADEON 9800s use of multiple pipelines to speed performance.

Programmable Vertex Shaders with DirectX 9.0 Support

The corners of each group of triangles that make up 3D objects are called vertices. Each of the four memory controllers passes vertex data to a separate programmable vertex shader pipeline (see Figure 3). The vertex shader pipelines are able to operate on 32-bit scalar data and 128-bit vector data at the same time.

Figure 3

The details of each programmable vertex shader pipeline are shown at left; there are four of these pipelines in the RADEON 9800s R350 Chip (right).

The programmable vertex shaders in the R350 chip are fully compatible with DirectX 9.0, which is the latest version of Microsofts 3D graphics, audio and gaming environment. After they complete their work, the vertices are passed to the triangle setup engine and onto the Hyper Z III technology.

Eight Programmable Pixel Shaders No Waiting!

Triangles without textures or colors arent very interesting to look at, so another important task the RADEON 9800 PRO must perform is to fill them with texture and pixel data. There are eight separate programmable pixel shader units which can process texture and pixel data at the same time (see Figure 4).

Figure 4

One of the eight programmable pixel shaders inside the RADEON 9800s R350 Chip.

One of the criticisms of the R300 chip used by the Radeon 9700 series was its inability to handle more complex pixel shader programs than those supported by DirectX 9.0. nVidias GeForce FX 5800 and 5900 go beyond DirectX 9.0s requirements in this area. However, the addition of the new F-buffer seen in Figure 4 enables the R350 to rerun color processing as many times as necessary to achieve the desired effect before passing the data through the rest of the pixel shader process.

Once the programmable vertex shaders and the programmable pixel shaders have munched on their respective data, they pass it to the triangle setup engine, which passes the data to the Hyper Z III+ section of the R350 chip for processing.

Hyper Z III+ A Smarter Z-Buffer with Shadow Effects

The Hyper Z III+ component of the RADEON 9800s R350 chip determines which parts of the 3D objects in a scene are hidden behind other objects and dont need to be rendered. Hyper Z III+s predecessor, Hyper Z III, introduced a feature known as Early Z (also called occlusion culling) to discard hidden pixels before the pixel shader engine processes them. Figure 5 provides a simplified explanation of how the Early Z feature used by Hyper Z III and III+ works.

Figure 5

Hyper Z III/III+s Early Z feature improves performance by discarding hidden pixels before they are processed by the pixel shader.

An object in the Z-buffer (1) is subdivided into blocks of pixels (2) by the Hierarchical Z component of Hyper Z. Other objects in the scene (3) are also subdivided into blocks (4). Any blocks that are behind other blocks in the scene (5) are discarded before Early Z renders the scene (6).

Previous Z-buffer methods didnt discard hidden pixels until after they were processed, so the graphics chip wasted time processing data youd never see. Other useful features of Hyper Z III+ include a lossless compression algorithm which shrinks data before it goes into the Z-buffer by a ratio of 2:1, 4:1, or as much as 24:1 when full-scene anti-aliasing (FSAA) is enabled. ATI chips, which use Hyper Z III, also support these features, so where does the Plus enter the picture? Hyper Z III+ features a Z cache, which is also optimized for real-time shadow rendering, which further improves the realism of 3D renderings. You can see a demonstration of real-time shadow rendering at the ATI website.

Improving 3D Image Quality

Rendering speed is only half the battle for a top-quality 3D graphics card; image quality is also paramount. The RADEON 9800 series features several improvements over its ancestors, including:

• Faster and better-looking FSAA with SMOOTHVISION 2.1
• Improved tessellation with TRUFORM II
• Support for ten bits-per-color modes

While previous versions of the ATI Radeon series also featured SMOOTHVISION FSAA to smooth on-screen objects, 3D rendering performance dropped drastically when the feature was enabled. SMOOTHVISION 2.x has the following improvements over the original version:

• it provides better user controls for the amount of FSAA (2x, 4x, or 6x)
• it supports combinations of supersampling (antialiases the entire image) and multisampling (antialiases the edges of triangles) for better image quality
• it supports anisotropic filtering with up to 16 texture samples per pixel to improve the appearance of objects at different distances from the viewer, such as textured walls or floors
• SMOOTHVISION 2.1 (new with the 9800) adds an adaptive algorithm, which uses the most efficient methods to achieve best display quality; can be adjusted for performance (bilinear) or quality (trilinear)

The performance of the Radeon 9800 still remains high when FSAA is enabled, so gamers no longer need to choose between high performance and high image quality.

TRUFORM II improves the appearance of complex 3D objects by subdividing each triangle in a scene into a series of smaller and smaller triangles formed from Bezier curves. This also enables more-realistic lighting effects. Compared to the original TRUFORM used in the first RADEON cards, TRUFORM II supports continuous tessellation for smoother curved surfaces and adaptive tessellation, in which closer objects have more detail and more distant objects have less detail. TRUFORM II also supports displacement mapping, a more advanced and realistic replacement for bump mapping.

Improving Video Quality

The R350, like the R300, features VIDEOSHADER, which combines programmable pixels shaders with video to improve video playback quality and add the capability for performing Photoshop-like transformations to video. VIDEOSHADER offers these benefits:

• streaming video de-blocking, which removes compression artifacts from Internet streamed video for a clearer picture
• improved de-interlacing for DVD and TV playback
• noise removal from captured video
• real-time special effects for video editing

These features enable the RADEON 9800 cards to provide excellent DVD playback.

Simplifying the 9800 - Creating the 9600 Series

The Radeon 9800 Pro cards beat nVidias top-of-the-line GeForce FX 5900 and FX5800 in most benchmarks, particularly when full-scene antialiasing is used. However, performance has its price. With prices for ATI-built Radeon 9800-series cards ranging from $400 to $470, the Radeon 9800 series is out of many users price range.

To provide a reasonable level of performance at a more competitive price, ATI has also introduced the Radeon 9600 series. Like the 9800 series, it has three members:

• Radeon 9600 Pro
• AIW 9600 Pro
• Radeon 9600

These cards range from $160 to $240, making them much more affordable. Table 2 compares the major features of these cards to each other.

Table 2 Radeon 9600-series Card Features Compared

As Table 2 makes clear, the Radeon 9600 series core and memory speeds are not that much different than the Radeon 9800 series. The differences can be found inside the RV350 chip at the heart of the 9600 series.

RV350 vs R350

The Radeon 9600 series uses a simplified version of the R350 chip known as the RV350. It is less expensive to make than the R300 because ATI gets a greater yield per wafer by using a 0.13 micron process with the RV350 than with the 0.15 micron process used by the R300. The RV350 is also cheaper to make because it has only 75 million transistors, compared to the R350s 110 million transistors. What did ATI leave out of the RV350?

The RV350 still fully supports DirectX 9.0 in hardware, but reduces transistor count (and performance) by using a 128-bit memory interface, using four pixel pipelines, omitting the new F-buffer, and using two vertex shaders. The RV350 also uses the older SMOOTHVISION 2.0 and Hyper Z III technologies. Table 3 compares these changes.

Table 3 Radeon 9800 (R350) Versus 9600 (RV350) Features

As a result of these changes, the Radeon 9600 series has considerably lower performance than the Radeon 9800 series in 3D gaming, although still on a par with mid-range cards based on the nVidia GeForce FX 5200.

9800 or 9600? Which Should You Choose?

If youre a hard-core gamer with a bulging bank account and you prefer ATI to nVIDIA, the Radeon 9600 series is probably not fast enough for you. Youll probably need to take a deep breath and spend the money for one of the 9800s. However, if youre a mid-range gamer who wants to be ready for DirectX 9 games when they start to show up, the 9600 series might be just right for you.

Another reason to consider the 9600 series is if you want the video and TV features of an All-in-Wonder card but dont want to sacrifice dual monitor output. The AIW 9800 Pro supports only one VGA or DVI monitor, but the AIW 9600 Pro features a dual VGA cable as well as the typical AIW video and audio input and output cables. If you want mid-range gaming and TV/video support the ATI way without sacrificing dual monitor support, the All-in-Wonder 9600 is your best choice.

More Product Choices? Dont Forget About Powered By ATI

This article has focused on the Radeon 9600 and 9800-series products that ATI produces. However, as I discussed in my article ATI Radeon Everywhere?, ATI has transformed itself into a major supplier of graphics chips for use by a wide variety of OEM suppliers while continuing to produce graphics cards under the ATI name. Since I wrote that article, ATI has added several new partners to the list, including Hercules, Visiontek, Connect3D, and others. For links to ATIs current partners, go to http://www.ati.com/support/partner/list.html

Some partners add different amounts and speeds of memory and other features on their versions of the Radeon 9800 and 9600 video cards, and prices might vary from ATIs own offerings. It pays to shop around!

For Further Study

ATI Resources:

Access the product pages for ATI Radeon 9800 and 9600 series cards at
http://www.ati.com/products/home-office.html

See real-time demos of the RADEON 9800s amazing rendering features at
http://www.ati.com/developer/demos/r9800.html

Take an animated tour through the R350 at the heart of the RADEON 9800 PRO at
http://www.ati.com/vortal/r350/static/demos.html

Third-party Resources:

Toms Hardware Benchmark Tests of RADEON 9600 PRO are here:
http://www20.tomshardware.com/graphic/20030416/index.html

Toms Hardware compares the RADEON 9800 PRO to the competition here:
http://www20.tomshardware.com/graphic/20030306/index.html

Anandtech Benchmark Tests of the RADEON 9600 PRO are here:
http://www.anandtech.com/video/showdoc.html?i=1812

Anandtech compares the RADEON 9800 PRO with its competition here:
http://www.anandtech.com/video/showdoc.html?i=1794

Get DirectX 9.0 from Microsoft here:
http://www.microsoft.com/windows/directx/default.asp

Anandtech uses the first major DirectX 9 game, Half-Life 2, to put both ATI and nVidia cards to the test. Find out what happened at:
http://www.anandtech.com/video/showdoc.html?i=1863

Copyright©2003 Pearson Education. All rights reserved.