ATI Technologies

ATI Technologies Inc.

ATI logo
Company type	Acquired by AMD in 2006
Industry	Semiconductors
Founded	1985
Headquarters	Markham, Ontario, Canada
Key people	David E. Orton, CEO
Products	Graphics cards Graphics processing units Motherboard chipsets Video capture cards
Revenue	$2.222 Billion USD (2005)
Net income	$16.93 Million USD (2005)
Number of employees	3,469 (2005)
Website	ati.amd.com

ATI Technologies Inc., founded in 1985, is a major designer of graphics processing units and video display cards and a wholly owned subsidiary of AMD, as of October 2006.

As a fabless semiconductor company, ATI conducts research & development of chips in-house, but subcontracts the actual (silicon) manufacturing and graphics-card assembly to third-parties. The production-chain of a chip involves multiple third-parties: the foundry (UMC and TSMC) makes processor wafers, the test-house tests the dies for defects and sorts them based on performance-characterization and the packager seals individual dies in a hardened case. In terms of inventory management, ATI must place foundry orders months in advance of their planned sale, then hold the produced chips in a warehouse until final delivery. The long lead time between order and delivery, combined with market uncertainty, leads to occasional supply/demand imbalances.

On July 24, 2006, AMD and ATI announced a plan to merge together in a deal valued at US$5.4 billion. The merger closed October 25, 2006 (Press Release). The acquisition consideration included over $2 billion financed from a loan, as well as 56 million shares of AMD stock. [1]

History

ATI's Silicon Valley office

Founded in 1985 by three Chinese immigrants, China-born Kwok Yuen Ho [2] and Hong Kong-born Benny Lau and Lee Lau. ATI began as an OEM, producing integrated graphics chips for large PC manufacturers like IBM. However, by 1987 it had evolved into an independent graphics card retailer, marketing the EGA Wonder and VGA Wonder graphics cards under its own ATI moniker.

In 1997 ATI acquired Tseng Labs's graphics assets, which included 40 new engineers. In 2000, ATI acquired ArtX, the company that engineered the "Flipper" graphics chip used in the Nintendo GameCube games console. They have also entered an agreement with Nintendo to create the chip for the successor of the GameCube, named Wii. ATI was contracted by Microsoft to create the graphics chip for Microsoft Xbox 360. Later in 2005, ATI acquired Terayon's Cable Modem Silicon Intellectual Property cementing their lead in the consumer digital television market (press release).

Its current President and CEO is David E. Orton (formerly of ArtX). K. Y. Ho remained as Chairman of the Board until he retired on November 22nd, 2005.

As reported earlier, ATI has agreed to merge with AMD for $5.4 billion. [3]. The merger is approved by Markham, Canada-based ATI shareholders and U.S. and Canadian regulators and the combined entity will still operate under the AMD name.[4]

Products

In addition to developing high-end GPUs (graphics processing unit, something ATI calls a VPU, visual processing unit) for PCs, ATI also designs embedded versions for laptops (called "Mobility Radeon"), PDAs and mobile phones ("Imageon"), integrated motherboards ("Radeon IGP"), set-top boxes ("Xilleon") and other technology-based market segments. Thanks to this diverse portfolio, ATI has been traditionally the dominant player in the OEM and multimedia markets.

Currently ATI is the main competitor of NVIDIA. As of 2004, ATI's flagship product line is the Radeon series of graphics cards which directly compete with those boards using NVIDIA's GeForce GPUs. As of the 3rd quarter of 2004, ATI represented 59% of the discrete graphic card market, while its primary competitor NVIDIA represented only 37%, but the two commonly trade market share majority, for example 2nd quarter had NVIDIA at 50% and ATI at 46%.

As of 2005, ATI has announced that a deal has been struck with CPU and Motherboard manufacturers, particularly Asus and Intel, to create onboard 3D Graphics solutions for Intel's new range of motherboards that will be released with their new range of Intel Pentium M-based desktop processors, the Intel Core and Intel Core 2 processors. This ATI solution will effectively end Intel's range of entry-level desktop integrated graphics. However, high-end boards with integrated graphics will still use Intel integrated graphics processors.

Computer graphics chipsets

This list is incomplete. Major product families are shown below:

• EGA / VGA Wonder - IBM "EGA/VGA-compatible" display adapters (1987)
• Mach8 - ATI's first 2D GUI "Windows Accelerator" (IBM 8514/A clone) (1991)
• Mach32 - VGA-compatible enhanced feature-set 2D GUI accelerator (32bit "true-color" acceleration) (1992)
• Mach64 - refined 2D GUI accelerator with "motion-video" acceleration (hardware bitmap zoom, YUV->RGB color-conversion) (1994)
• Rage Series - ATI's first 2D and 3D accelerator chip. The series evolved from rudimentary 3D with 2D GUI acceleration and MPEG-1 capability, to a highly competitive Direct3D 6 accelerator with then "best-in-class" DVD (MPEG2) acceleration. The various chips were very popular with OEMs of the time. (1995-2004)
  • Rage Mobility - Designed for use in low-power environments, such as notebooks. These chips were functionally similar to their desktop counterparts, but had additions such as advanced power management, LCD interfaces, and dual monitor functionality.

File:Sapphire Ati Radeon X1900 XTX PCI-E 512MB Rev A 0552.JPG

A Radeon X1900 series graphics card.

• Radeon Series - Launched in 2000, the Radeon line is ATI's brand for their consumer 3D accelerator add-in cards. The original Radeon DDR was ATI's first DirectX 7 3D accelerator, introducing their first hardware T&L engine. ATI often produced 'Pro' versions with higher clock speeds, and sometimes an extreme 'XT' version, and even more recently 'XT Platinum Edition (PE) and XTX' versions. The Radeon series was the basis for many of ATI's "All-In-Wonder" boards.
  • Mobility Radeon - A series of power-optimized versions of Radeon graphics chips for use in laptops. They introduced innovations such as modularized RAM chips, DVD (MPEG2) acceleration, notebook GPU card sockets, and "POWERPLAY" power management technology.
  • ATI CrossFire - This technology was ATI's response to NVIDIA's SLI platform. It allowed, by using a secondary video card and a dual PCI-E motherboard based on an ATI Crossfire-compatible chipset, the ability to combine the power of the two video cards to increase performance through a variety of different rendering options.
• FireGL - Launched in 2001, following ATI's acquisition of FireGL Graphics from Diamond Multimedia. Workstation CAD/CAM video card, based on the Radeon series.

Console graphics solutions

• Flipper - The Nintendo GameCube contains a 3D accelerator developed by ArtX, Inc, a company acquired by ATI towards the end of development of the GPU. Flipper is similar in capability to a Direct3D 7 accelerator chip. It consists of 4 rendering pipelines, with hardware T&L, and some limited pixel shader support. Innovatively the chip has 3 MB of embedded 1T-SRAM for use as ultra-fast low-latency (6.2 ns) texture and framebuffer/Z-buffer storage allowing 10.4 GB/second bandwidth (extremely fast for the time). Flipper was designed by members of the Nintendo 64 Reality Coprocessor team who moved from SGI. The Flipper team went on to have a major hand in development of the Radeon 9700. Analysis
• Xenos - Microsoft's Xbox 360 video game console contains a custom graphics chip produced by ATI, known as "R420", "C1", or more often as Xenos. Some of these features include “Intelligent Memory” – a section of on-die memory that has logic built in (192 parallel pixel processors) to do features like anti-aliasing, thereby giving developers 4-sample anti-aliasing at very little performance cost. Another feature of the ATI Xbox 360 GPU is the “True Unified Shader Architecture” which dynamically load balances pixel and vertex processing amongst a bank of identically capable processing units. This differs greatly from contemporary PC graphics chips that have separate banks of processors designed for their individual task (vertex/pixel). Analysis
• Hollywood - Nintendo's next-gen gaming console, the Wii, will use a custom GPU by ATI.

Handheld chipsets

• Imageon - Introduced in 2002 to bring integrated 3D graphics to handhelds, cellphones and Tablet PCs. Current product is the Imageon 2300 which includes 3D engine, MPEG-4 video decoder, JPEG encoding/decoding, and a 2 megapixel camera sub-system processing engine with support for 2 MiB of ultra low-power SDRAM.
• In May 2006 ATI claimed it had sold over 100 million 'cell phone media co-processors,' significantly more than ATI's rival NVIDIA.

Personal computer platforms & chipsets

Early north bridge parts produced by ATI included the Radeon 320, 340 and 7000. Typically these were partnered with a south bridge chip from ALI. They sold in respectable volumes, but never gained enthusiast support.

In 2003 ATI released the 9100 IGP[5], with IXP250 southbridge. It was notable for being ATI's first complete motherboard chipset, including an ATI southbridge, admittedly light on features, but stable and functional. It included an updated Direct-X 8.1 class version of the 8500 core for the integrated graphics, based upon the 9100. Internally, ATI considered it one of their most important product launches.

The Xpress 200/200P is ATI's PCI Express-based Athlon 64 and Pentium 4 motherboard chipset. The chipset supports SATA as well as integrated graphics with DirectX 9.0 support, the first integrated graphics chipset to do so. The graphics is based on an X300 core integrated into the north bridge, with two pixel pipelines operating at a core speed of up to 350 MHz, each one having a single texturing unit.

In 2006, ATI Released the Xpress 3200, a true Crossfire solution. Where the XPRESS 200 (2x PCIe x8 or 1xPCIe x16) is not designed specifically for Crossfire, the Xpress 3200 (2x PCIe x16) is. Because both x16 slots are connected to one physical chip, ATI have been able to accelerate the link between both graphics card slots in order to compensate for the lack of a dedicated GPU-to-GPU interconnect.

Operating system drivers

ATI currently provides proprietary drivers for Microsoft Windows XP, Mac OS X, and Linux. Linux users have the option of both the proprietary (R200 and above) and open source (R480 and below) drivers. More details can be found on the Radeon page. In an interview with AMDs Hal Speed it was suggested that AMD were strongly considering open sourcing at least a functional part of the ATI drivers. [6]. However, at least until the merger with AMD is complete, ATI have no plans to open source their drivers:

Proprietary, patented optimizations are part of the value we provide to our customers and we have no plans to release these drivers to open source. In addition, multimedia elements such as content protection must not, by their very nature, be allowed to go open source.

— the company said in a statement, [7]

Incident with Richard Stallman on MIT campus

In April 2006, when an ATI representative was to speak at the MIT campus in the same building where the Free Software Foundation rents their offices, Richard Stallman organised a protest against ATI on the grounds that ATI does not release the documentation of their hardware, thus making it largely impossible to write free software drivers for their graphics adapters. For the duration of ATI's speech, Stallman was holding a sign that said "Don't buy from ATI, enemy of your freedom". Although Richard Stallman had no intention of disrupting the speech, and indicated that the sign was loud only visually, the organisers of the speech brought a police officer to the scene, although they have failed to provide the officer with a valid reason for his presence at the event, according to the FSF web-site. [8]

Market trends

ATI was founded in 1985, and in order to survive, initially ended up shipping a lot of basic 2D graphics chips to companies such as Commodore. The EGA Wonder and VGA Wonder families were released to the PC market in 1987. Each offered enhanced feature sets surpassing IBM's own (EGA and VGA) display adapters. May of 1991 saw the release of the Mach8 product, ATI's first "Windows accelerator" product. Windows accelerators offloaded display-processing tasks which had been performed by the CPU. (In fact, the Mach8 was feature-enhanced IBM 8514/A-compatible board.) 1992 saw the release of the Mach32 chipset, an evolutionary improvement over its predecessor.

Modern integrated chipsets

But it was probably the Mach64 in 1994, powering the Graphics Xpression and Graphics Pro Turbo, that was ATI's first recognizably modern media chipset. Notably, the Mach64 chipset offered hardware support for YUV-to-RGB color space conversion, in addition to hardware zoom. This effectively meant basic AVI and MPEG-1 playback became possible on PCs without the need for expensive specialized decoding hardware. Later, the Mach64-VT allowed for scaling to be offloaded from the CPU. ImpacTV in 1996 went further with 800x600 VGA-to-TV encoding. ATI priced the product at a point where the user effectively got a 3D accelerator for free.

ATI’s first integrated TV tuner products shipped in 1996, recognizable as the modern All-in-Wonder specification. These featured 3D acceleration powered by ATI's second generation 3D Rage II, 64-bit 2D performance, TV-quality video acceleration, video capture, TV tuner functionality, flicker-free TV-out and stereo TV audio.

However, while ATI had established a reputation for quality multimedia-capable cards popular with OEMs, by the late 1990s consumers began to also expect strong 3D performance, and 3dfx and NVIDIA were delivering. The first warning was seen with in January 1999 with the All-in-Wonder 128, featuring the Rage 128 GL graphics chip. While the basic 16 MiB version sold reasonably well, the improved but delayed 32 MiB version did not, because it lacked 3D acceleration appropriate for its price point. It became clear that if ATI was to survive, the company would have to develop integrated 3D acceleration competitive with the products NVIDIA was designing.

Improved 3D performance

ATI’s first real 3D chip was the 3D Rage II. The chip supported bilinear and trilinear filtering, z-buffer, and several Direct3D texture blend modes. But the pixel fillrate looked good only next to S3’s VIRGE cards, which were of very poor quality for the time, and the feature list looked good only next to the workstation type Matrox Mystique.

The 3D Rage Pro, released in 1997, offered an improved fill rate equal to the original 3dfx Voodoo Graphics, and a proper 1.2 M triangle/s hardware setup engine. Single-pass trilinear filtering combined with a complete texture blending implementation. The Rage Pro sold in volume to OEMs due to its DVD performance and low cost, but was held back by poor drivers. It was only in 1999, almost two years after the original launch, the drivers finally achieved their potential, delivering a 20-40% gain over the originals. Subsequently ATI learned to better prioritise driver development.

Work on the next-generation 128 GL was helped by the acquisition of the Tseng development team in 1997. Designed to compete with the RIVA TNT and Voodoo2, it was notable for its advanced memory architecture which allowed the Rage128 to run in 32-bit color modes with minimal performance losses. Unfortunately, at the time most games ran in 16-bit color modes, where NVIDIA's parts excelled.

The RIVA TNT2 came out with improved clock speeds, and the GL quickly became relegated to ATI's usual position: that of a strong OEM alternative to the market leaders, with outstanding DVD performance, attractive when priced low enough.

The part was updated in April 1999 with the Rage 128 Pro, featuring anisotropic filtering, a better triangle setup engine, and a higher clock rate. The Rage 128 Pro's MPEG-2 acceleration was far ahead of its time, allowing realtime MP@HL (1920x1080) playback on a Pentium III 600 MHz. ATI also ran an experimental project called "Project Aurora," marketed as the MAXX technology, consisting of dual Rage 128 Pro chips running in parallel, with each chip rendering alternate frames. Because the MAXX required double the memory, suffered from buggy drivers, and failed to deliver knockout performance, it was not a successful launch. As a result, ATI discontinued multiple chip development for mainstream products.

Radeon line

By this point the pattern seemed clear: ATI was good at producing low-end OEM-friendly parts with good 2D features, DVD acceleration, and rounded 3D feature sets. What they had failed to do was challenge effectively at the high end of the market. So, at the Game Developer's Conference in March 2000, developers were curious but generally somewhat skeptical about a new claimed sixth-generation graphics chip. This was a period when companies often announced products that they failed to deliver on time, or on spec. However, ATI subsequently demonstrated beta silicon behind closed doors at GDC, and named the product the Radeon 256.

Finally released in 2000, the original Radeon core (R100). ATI's new video card based on this core was originally named the Radeon 64 VIVO to emphasize its 64 MiB of DDR memory and video features, but was eventually renamed the Radeon 7200, reflecting its DirectX 7-compliant feature set. The R100 core established a number of notable firsts, such as a complete DX7 bump-mapping implementation (emboss, dot product 3, and EMBM), hardware 3D shadows, hardware per-pixel video-deinterlacing, and a reasonable implementation of many advanced DX8 pixel shader effects. Unfortunately, ATI used a 2 pixel pipeline design for the R100 with three raster units per pixel pipeline. NVIDIA's competing GeForce 2 chips had a four pipe design with two raster units per pipeline. Very few 3D applications at the time utilized more than two textures per pixel, and thus the third raster units in the Radeon were seldom utilized.

ATI proved the original Radeon had not been a one-off by following up with the second generation Radeon (R200) core in 2001, marketed as the Radeon 8500. The R200 raster pipeline arrangement matched the design of the NVIDIA's GeForce 2 series with four pipelines with two raster units per pipeline. ATI was shooting for 300 MHz core speed for the new 8500, but was unable to reach it. In fact, ATI retail boxes and literature describe the texture fillrate of the 8500 at the 300 MHz speed (2.4 GTexel/s), but the cards were only shipped at 275 MHz speed. NVIDIA quickly released GeForce cards with faster clock speeds. NVIDIA's top GeForce 4 Ti cards delivered greater raw power in terms of fill rates, but ATI started to open up a clear quality and shader performance advantage. In fact, many new games in 2005 still supported the DirectX pixel shader 1.4 of the R200, but not the less capable pixel shader 1.3 units of NVIDIA's latter released GeForce 4 chips

Challenging NVIDIA

During this period ATI also began to sell their core chip technology to third-party "Powered by ATI" board manufacturers, directly competing with NVIDIA’s business model. This change suddenly put NVIDIA on the back foot for the first time since the ill-fated NV1 project, to the amazement of the entire industry. Alongside the Radeon 8500 ATI released a die shrink version (RV200) of the original R100 core which was released as the Radeon 7500. This chip had an extremely fast core clockspeed for the time of 290 MHz with all the features of the original Radeon. ATI also sold a single pipeline version of the original Radeon as the Radeon 7000. Left over R100 chips were sold to third-party video card manufacturers and marketed as the Radeon 7200.

The Radeon 8500 proved popular with OEMs, partly because it offered wider motherboard compatibility than NVIDIA's offerings of the period. The 8500 finally established ATI as a serious performance and feature integrated chipset competitor to NVIDIA, in a period when other graphics card companies such as 3dfx were going out of business. However poorly-written drivers continued to be a notorious weakness of ATI products. ATI responded by introducing their unified "Catalyst" driver/application suite, which attempted to address the quality, compatibility, and performance concerns raised by the user community.

Performance leadership

The Radeon SE/VIVO and Radeon 8500 cards were warning shots for NVIDIA, demonstrating they could not take for granted their dominant market position. 2002 proved to be the decisive year for ATI, with an unexpected introduction of a new Radeon architecture. The third generation Radeon 9700, based on the R300 core, was designed from the ground up for DirectX 9 operation. Upon its release, it was easily the fastest consumer gaming video card available. Furthermore, ATI beat NVIDIA’s DirectX 9 chip to market by several months and soundly defeated it in almost every application. NVIDIA's "NV30" architecture, while innovative and forward-looking, suffered when advanced features were used, such as pixel/vertex shading, anti-aliasing, and anisotropic filtering.

Mainstream value

From then onwards, the challenge for ATI became holding onto their high-end advantage, while filtering their technology down to the mid and low end of the market, where the greatest volume sales are made. ATI decided to sell R300 cores with a reduced core clock speed, and half the pixel pipelines disabled, as a midrange product called the Radeon 9500. ATI's own Radeon 9500 Pro card was a R300 core with a 128 bit memory bus running at 275 MHz. This card proved to be just as fast as the fastest GeForce 4 cards and considerably faster than those cards in DirectX 9 applications.

The release of the R300 brought a great deal of interest in ATI from third party manufacturers. To meet the demand for new mid-range cards ATI even allowed manufacturers to sell some Radeon 9700 cards with half their pipelines disabled as Radeon 9500 cards. These differed from the 9500 Pro cards in that they had a 256 bit memory bus, however with only 4 working pipelines their performance was markedly reduced. Soon hardware enthusiasts discovered that it was possible and rather easy to unlock the disabled pipelines on these discounted cards. Eventually, the only thing required to turn these inexpensive Radeon 9500's into full Radeon 9700's was a hacked software driver.

For the low end, ATI released a new value chip (RV250) based on the Radeon R200 core with half the raster units per pipeline. This raster arrangement actually matched the original GeForce design, but the Radeon 9000 also had the same shader processing power and features as the 8500. This DirectX 8.1 capable part competed with NVIDIA's two pipeline, DirectX 7 GeForce 4 MX. Despite having half the fillrate of the Radeon 8500, the Radeon 9000 had very similar performance. ATI also allowed third party manufacturers to continue selling the original R200 cores as Radeon 9100's to reflect the slight performance advantage of the extra raster units. However, the situation was soon confused when the AGP 3.0 refresh of the RV280 was named the Radeon 9200, and when ATI named its new two pipeline integrated chipset the Radeon 9100 IGP.

ATI refreshed the 9700 to the 9800 Pro (R350) in 2003, featuring a small and relatively quiet cooling solution. The 9800 went on to become one of the most popular and best selling enthusiast cards to date. In the midrange market, the 9600 (RV350) was introduced with half the number of pixel pipelines of the 9800 Pro. Adding to the model naming confusion, this card was generally inferior in performance to the Radeon 9500 Pro (R300), but it was cool running. While the pixel fill rate of the Radeon 9600 did not exceed previous generation parts such as the Radeon 8500 (R200) and GeForce 4 series cards, it featured fast and power efficient shader support, offering excellent performance on DirectX 9 based titles. It was refreshed as the 9600XT, gaining another 100 MHz to 500 MHz, from an improved low-k manufacturing process.

Gaining market share

In 2004, ATI released the RADEON XPRESS 200 motherboard chipset, intended as a direct competitor to the more established nForce motherboard brand of chipsets from arch rival NVIDIA. The 9700 core trickled down into the low end market in the form of a cost-reduced 9600, the 9550, which was fabbed on 0.11u. Even at its core-clock of 250 MHz, the 9550 quickly overtook NVIDIA's 5200 as the favorite entry-level discrete OEM card. As a result, almost unnoticed, ATI completed one of the most surprising turnarounds in recent chip history.

According to data from Mercury Research, ATI Technologies' market share rose by 4 percentage points to 27% in the Q3 2004, while NVIDIA's share dropped 8 points from 23% to 15%. Intel's market share rose 1 point to 39% in the Q3 2004, holding on to the market number one position, although Intel only ships low performance integrated solutions.

In 2005, ATI began shipping the x800 XL PCI-E card, a 110 nm shrink of the x800 core (which originally shipped on a 130 nm low-K process.) This brought 16-pipeline cards closer to the mainstream. The x850 range marks the end of the old 9700 feature set/core as ATI's performance platform. The omission of SM 3.0 and FP32 permitted a more compact die size, allowing ATI to price the X800XL lower than comparable NVidia products.

x1000 series

The long-awaited Radeon X1000 series was ATI's first major architectural advancement since the 9700 series. The high end Radeon X1800 had been planned for a mid-2005 release, but the chip did not reach the retail market until October 2005. ATI's first foray into 90nm production was an unhappy one: a silicon-library bug reduced clockable speeds by 150 MHz, delaying R520 production for several months. The missed window of opportunity allowed NVIDIA's 7800 line to dominate the high-end market. As it turned out, the delay of the X1800 led to ATI's entire SM3 product-line launching at roughly the same time: entry-level X1300, mainstream X1600, and high-end/enthusiast X1800. The high end Radeon X1800 managed to maintain parity with the NVIDIA GeForce 7800 GTX. ATI retained slightly greater market share, though margins and profitability slumped.

In January 2006, ATI replaced the short-lived X1800 with the Radeon X1900XT and X1900XTX (R580). With 48 pixel shader units, R580 brought ATI's 3:1 pixel shader to pipeline ratio (first seen on the 12 shader Radeon X1600) to the desktop high end. This enabled ATI to regain the performance crown over the GeForce 7800 GTX 512 in the majority of situations and to an extent, even against the later-released 7900GTX. However, the R580's die-size suggests ATI's performance leadership came at a significant cost. The x1900 (R580) core contains roughly 384M transistors, in a die size of 352 mm². NVidia's 7900 (G71) core contains roughly 278M transistors, in a die size of 196 mm². As both devices are known to be manufactured on TSMC's 90nm low-K CMOS logic process, the raw per-die cost of the R580 core is estimated to be twice as much as the NVIDIA part. While the ATI part has a more flexible feature set, the difference in manufacturing cost points to ATI facing near term margin pressure.

On July 21, ATI announced the newest product to the x1000 series, the Radeon X1950XTX(R580+). The X1950XTX will come in two versions, the XTX and Crossfire version, and will feature GDDR4 memory, as opposed to the current GDDR3 memory used in the X1900s. The new core and memory clocks for the X1950s will be 650 MHz and 2 GHz, respectively, compared to the X1900XT with 625/1450 and the XTX with 650/1550. The X1950XTX was presented on August 23, 2006. It is available since mid-September and retails at US$449. On 17 Oct ATI introduced the x1950 PRO based on the new RV570 core, a die shrink to 80nm of the R580, with a reduced 12 pixel pipelines and 36 pixel shaders (the X1900GT had the same configuration, but was a R580 with one pixel quad disabled). A compositing engine is also integrated into the core, so that two X1950 PRO's can now work together, without the need for a mastercard or external dongle. At 230 mm2 with 330 million transistors[9], the new die is much cheaper to manufacture.

Stream / AVIVO

The R5x series has seen ATI introduce the concept of GPUs as 32-bit precision CPUs. That is to say the computation complexity of GPUs has moved so far from the fixed functions of Direct-X 7 requirements, that actual code programming is now possible, for specific application scenarios. The mass client project Folding@Home has reported 20-40 times[10] speed improvements using an R580 card, as opposed to a CPU. It is anticipated in the industry, graphics cards may be used in future game physics calculations.

See also

• AMD
• Comparison of ATI Graphics Processing Units
• Comparison of ATI Chipsets
• Graphics card
• Graphics processing unit
• NVIDIA
• Comparison of NVIDIA Graphics Processing Units
• Matrox, the other Canadian video graphics card company
• Stream processing

External links

• AMD's Graphics Division
• ATI: CorporateMilestones.pdf
• FiringSquad's History of ATI
• ATI headquarters at Google MapsI* Official AMD ATI Merger Site on ATI Website
• Official AMD ATI Merger Site on AMD Website