History of the Modern Graphics Processor, Part 3

With the turn of the century, the graphics industry has witnessed further consolidation.

In the professional market, iXMICRO completely abandoned the graphics function, while NEC and HP both produced the final products, namely TE5 and VISUALIZE FX10 series. Evans and Sutherland (Evans & Sutherland) also parted ways by selling RealVision series products, focusing on planetary stars and full dome projection systems.

In the consumer graphics market, ATI announced in February 2000 that it would acquire ArtX Inc. for approximately US$400 million in stock. ArtX is developing a GPU codenamed Project Dolphin (eventually called "Flipper") for Nintendo GameCube, which greatly increases ATI's profits.

Also in February, 3dfx announced

And quickly move to

Acquired the company's tile-based rendering IP for $186 million.

At the same time, S3 and Nvidia settled their outstanding patent litigation and signed a 7-year contract

.

VIA gained control of S3 between April and May, and the company itself has just completed the restructuring process of acquiring Number 9. As part of the S3 reorganization, the company merged with Diamond Multimedia, valued at US$165 million. FireGL, Diamond's high-end professional graphics department, was split into SONICblue, which was later sold to ATI for USD 10 million in March 2001.

3DLabs acquired Intergraph's Intense3D in April, and 3dfx's final action took place at the end of the year, although it started in 2000, as the long-awaited Voodoo 5 5500 approaches its debut in July, its prospects are even better. The latter ended the deal with GeForce 256 DDR and won the high resolution battle.

But 3dfx was once synonymous with original performance, and the advantage at this time is its full-screen anti-aliased image quality. Voodoo 5 introduces T-buffer technology as an alternative to transformation and lighting. It basically takes some rendered frames and aggregates them into one image. This produces a slightly blurred picture, which can smooth the motion of the animation when running in frame order.

3dfx's technology has become the pioneer of many image quality enhancements, such as soft shadows and reflections, motion blur, and depth of field blur.

3dfx's swan song Voodoo 4 4500 was released on October 19 after several delays-different from the 4200 and 4800 that were never released. The card was originally

As a competitor to Nvidia TNT2, but it ultimately runs counter to the company's iconic GeForce 256 DDR and the better-performance GeForce 2 GTS and ATI Radeon DDR.

On November 14, 3dfx announced that they were late

There have been rumors in their own brand of graphics cards for some time, but the discounts are huge. It is reported that the upcoming Pentium 4 motherboard will not support this processor.

Voodoo 5 series.

NVIDIA sounded the death knell for 3dfx a month later

Issuance of US$70 million plus 1 million common shares. Later, some Internet people pointed out that the 3dfx design team that moved to Nvidia finally gained revenge and realized its potential by providing the poorly performing NV30 graphics chip (powering the FX 5700 and FX 5800 cards).

Before Voodoo 5 arrived, ATI

As "the most powerful graphics processor ever designed for desktop computers."

NVIDIA was listed on April 25, and only 24 hours later, NVIDIA announced the launch of GeForce 2 GTS (GigaTexel Shader) to fight back. The latter includes Nvidia's ATI Pixel Tapestry Architecture version, called Nvidia Shading Rasterizer, which can apply effects such as highlight shadows, volume explosions, refraction, waves, vertex blends, shadow volume, bump maps and elevation maps to each pixel through the hardware basis.

This feature is believed to have been applied to the previous NV10 (GeForce 256) chip, but it was disabled due to a hardware failure. GTS also follows ATI's Charisma engine, allowing the GPU to support all transformation, cropping and lighting calculations. In other words, ATI went a step further in vertex skinning to make the polygons move more smoothly and perform key frame interpolation. Developers designed the start and end meshes of the animation, while the Charisma core calculated the middle mesh. .

ATI Radeon DDR finally began retail sales in August 2000. With excellent T&L implementation and support for the upcoming DirectX 8 function, Radeon DDR and GeForce 2 GTS ushered in the use of DVI output through integrated interface support. Enter the chip itself. DVI output is usually found on OEM cards, because retail varieties usually come with VIVO plugs.

One of the disadvantages of Radeon DDR is that the core and memory of the motherboard are down-clocked from the promised 200MHz and 183MHz respectively. In addition, the driver lost its best condition again when it was released. The VIA chipset has 16-bit color issues and compatibility issues, but this does not prevent the card from leading the way at resolutions higher than 1024x768x32. The 64MB version is priced at US$399, while the 64MB GeForce 2 GTS is priced at US$349-399. The latter has a profit margin of 10-20% in the benchmark test, helping ATI maintain its top share of Nvidia in the graphics market. One. .

Nvidia has not done anything bad for itself. The company reported net income for the fiscal year of 98.5 million U.S. dollars and a record revenue of 735.3 million U.S. dollars. This was largely due to its market segmentation strategy. In June, the company released a desalinated MX version of the card, and the Ultra model with a higher clock was released in August. The latter replaced Radeon in terms of performance, but it also costs $499. The Pro model arrived in December.

In addition to launching GeForce 2 cards at every price range from budget MX to professional Quadro 2 series, Nvidia also released its first mobile chip in the form of GeForce2 Go.

By the dawn of 2001, the PC graphics market had been monopolized by the discrete card duopoly. In addition to the vast majority of integrated graphics chipsets provided by Intel, these two products were exclusively owned by Intel.

As 3dfx got into trouble in November, Imagination Tech (formerly VideoLogic) and STMicroelectronics (ST Micro) tried to solve the high-budget market with the PowerVR series 3 KYRO. The price of the card is usually between US$80 and US$110, depending on the memory frame buffer. The card has a high cost performance for games at a resolution of 1024x768 or lower. If GeForce2 MX comes out later, or if its pricing is not so aggressive, about $110, it will become more popular.

KYRO II went on the market in April 2001. Its clock speed has been improved compared to the original clock, and it is manufactured by ST Micro using a smaller 180nm process. But the card once again faces fierce competition from GeForce 2 MX. Nvidia relabeled the card as MX200 and reduced its price by 40%, while adding a higher frequency MX400 card at the same price as Kyro II.

When PowerVR was unable to ensure the momentum for game development based on tile-based rendering, ST Micro closed its graphics business in early 2002. Imagination Technologies switched from desktop graphics to mobile graphics and used this expertise to develop System-on-chip graphics. They license the 5/5XT/6 series for ARM-based processors in the ultraportable and smartphone markets.

At the same time, Matrox and S3/VIA firmly grasp the profits of the traditional market.

Based on the rapid development of the GeForce 2 series, Nvidia launched the GeForce 3 on February 27, 2001, with a price of between US$339 and US$449. The card became the new king of the hills, but in fact it only really appeared at the (at the time) extremely high resolution of 1600x1200, it is best to apply full-screen anti-aliasing.

There was a problem with the original driver, especially in certain OpenGL titles. The new GeForce does bring DirectX 8, multi-sampling AA, plum blossom AA (basically 2xMSAA + post-processing blur), 8x anisotropic filtering and unparalleled processing 8xAF + tri-linear filtering ability and programmable vertex shader For tighter control of polygon mesh movement and smoother animation sequences.

It also supports LMA (Light Speed ​​Memory Architecture)-basically Nvidia's HyperZ version-for removing pixels that are ultimately hidden behind other pixels on the screen (Z occlusion culling), as well as compressing and decompressing data to optimize bandwidth usage (Z compression).

Finally, Nvidia implemented a load balancing algorithm, called the Crossbar memory controller, which consists of four independent memory sub-controllers, as opposed to the industry standard single controller, which can route incoming more efficiently Memory request.

Nvidia's product line later added NV2A, which is a derivative product of GeForce 3 with GeForce4 attributes. This product has been used in Microsoft's Xbox game console. At this point, Nvidia controls 31% of the graphics market, Intel 26%, and ATI 17%.

Nvidia NV2A built into Microsoft Xbox

As Nvidia supplements the GF3 series with the underclocked Ti 200 and overclocked Ti 500 models, ATI is anxious to increase the delivery of Radeon 8500. The card is built around the R200 GPU using TSMC’s 150nm process (the same as that used in GeForce 3’s NV20). . The chip was announced in August and has been eagerly anticipated since John Carmack of ID Software stated it.

As GeForce 3.

ATI official R8500 announcement

. However, once the card was launched in October, reality began to come into play, and was found to reach the frequency of the GF3 Ti 200 in the game. In the first round of evaluation of R8500, the unfinished driver and lack of viable Smoothvision anti-aliasing function had a significant impact on R8500. By the time the holiday season arrives, the second round of review shows that the driver has matured to a certain extent and has improved the performance of R8500 between Ti 200 and standard GF3.

The extremely competitive price and better feature set (2D image quality, video playback, anti-aliasing performance) make this card a strong competitor to GF3 and Ti 500.

ATI's annual sales fell to 1.04 billion U.S. dollars, and the company's net loss was 54.2 million U.S. dollars. The company began to grant board partners a license to manufacture and sell graphics boards, while refocusing its resources on design and chip manufacturing.

ATI Xilleon board

To complement Xilleon, ATI acquired NxtWave Communications for US$20 million in June 2002. The company specializes in digital signal processing and applications for set-top boxes and terrestrial digital solutions. ATI also launched Xilleon based on the set-top box for the first time, which is a development platform based on Xbox. The Xilleon 220 SoC provides a complete processor, graphics, I/O, video and audio for set-top boxes integrated into digital TV designs.

In order to keep up with the product release cycle, Nvidia released GeForce4 in February 2002. Three MX parts, three moving parts based on MX models, and two high-performance Titanium models (Ti 4400 and Ti 4600) form the initial product lineup-based on TSMC's 150nm process. GeForce 4 was effectively prepared to be released two months ago, but the postponement was to avoid affecting GeForce 3 sales during the holiday period.

The MX series cards were originally intended to be used in the budget segment, but because they are based on the old GeForce 2 architecture, they are still largely uninspiring. Added MPEG2 decoding, but as the early GF2 MX series, the card has been restored to DirectX 7.0/7.1 support. The price of $99-179 reflects the reduction of this feature.

On the other hand, the Titanium model performed well. In some cases, its performance was improved by more than 50% over the GeForce3 Ti 500. Ti4600 became the performance champion overnight, easily disposing of Radeon 8500, while Ti 4200 priced at $199 represents the best value credit card.

but then

And immediately entrust all other cards to a state that can still be used.

The ATI R300 GPU was developed by the team that originally formed the core of ArtX, delivered excellently, and delivered quickly. It is the first application to provide DirectX 9.0 support, and through extension, it is the first architecture to support shader model 2.0, vertex shader 2.0 and pixel shader 2.0. Other notable achievements: This is the second GPU series to support AGP 8x-SiS's Xabre 80/200/400 series is the first-and the first flip-chip GPU package is implemented.

Previous generations of graphics chips and other ICs used wire bonding to mount. Using this method, the logic block of the chip on the logic board is located under the metal layer, and its logic pads will be connected to the solder balls or pins on the lower side through thin wires arranged downwards around the edge of the chip. Flip-chip removes the wire assembly through the contact points directly on the "top" of the chip (usually soldered in the form of a ball grid array) and then turns it upside down or "flip" so that the solder joints directly contact the substrate or circuit board. Then, the chip undergoes local heating (reflow) to melt the solder, and then forms a connection with the lower-level contacts of the circuit board.

ATI supplemented its product line in October, adding a non-Pro 9700 for those who can’t buy the top model for $399 for $299. At the same time, the reduced prices of the 9500 Pro ($199) and 9500 Pro ($179) are segmented through the mainstream market, while the FireGL Z1/X1 fills the professional graphics bracket for 550-950 dollars. The All-In-Wonder 9700 Pro ($449) was also added in December.

When it is discovered that many cards can be switched to similar products with higher prices, ATI's sales may be hit. For example, you can use its reference board (with a complete memory trace) to convert a 9500 card to a 9700, or convert a 9800 Pro to its XT copy. For the latter, a driver patch is provided to check whether it accepts the mod. The patch includes soldering resistors or using a pencil to adjust the GPU and memory voltage control chip. The hard modification also includes upgrading various 9800 models to FireGL X2, and the patch/Omega driver can turn the $250 9800 SE 256MB into the $499 9800 Pro 256MB.

In addition to discrete graphics, ATI also introduced desktop integrated graphics and chipsets. These include A3/IGP 320 intended to be paired with AMD CPUs, RS200/IGP 330 and 340 for Intel chips, and mobile series U1/IGP 320M for AMD platforms and RS200M for Pentium 4-M. All these are supplemented by ATI South Bridge, especially IXP200/250.

SiS introduced the Xabre series between the GeForce4 and R300. At the same price point, these cards have always been slower than Nvidia and ATI products, and are limited due to the lack of a vertex shader pipeline. This translates to relying heavily on drivers and game developers to make full use of software simulation, so that SiS is at the edge of the desktop discrete 3D graphics.

The Xabre production line also implements "turbo texturing" technology, which increases the frame rate by greatly reducing the texture quality and lacks anisotropic filtering. None of these attracted commenters.

The Xabre product line is the last product line under SiS because the company divested its graphics division (renamed XGI) and merged with Trident Graphics a few months later in June.

Nvidia's first FX series came out on January 27, 2003. It came with the infamous "vacuum cleaner" FX 5800 and a slightly faster (but slower FX 5800 Ultra). Compared with the mainstream ATI Radeon 9700 Pro (and non-Pro), FX has a louder sound, its anisotropic filtering (AF) quality and anti-aliasing (AA) performance are both poor, and the overall speed is much slower . ATI is far ahead. The second-tier Radeon 9700 card, which was launched five months ago, has better performance than Ultra, and the price is $100 cheaper ($299 vs. $399).

The 3dfx design team handed over to Nvidia has both revenge and fulfilled its potential by delivering the poorly performing NV30 graphics chip on schedule.

The NV30 chip was originally scheduled to go on the market at the same time as the Radeon 9700 in August, but TSMC’s (TSMC) Low-K 130nm process has aggravated problems and high defect rates that discouraged Nvidia. Some circles also argue that the company’s engineering resources are limited, many of which are related to the NV2A Xbox console chip.

, And the motherboard chipset.

In order to move forward, Nvidia has undertaken a project to manufacture several FX series chips using IBM's more traditional fluorosilicate glass (FSG) low-K 130nm process.

ATI has updated its card series since 3800. Starting from 9800 Pro, this product has R350 GPU, basically R300, with some enhancements to Hyper-Z cache and compression instructions.

RV350 and RV280 were subsequently released in April. The first of these was found in Radeon 9600, using the same TSMC 130nm low-K process used by Nvidia. At the same time, the RV280 that powers the Radeon 9200 is nothing more than the relabeled Radeon 9000 RV250 (with AGP 8x support).

In the same month, ATI and Nintendo

It will eventually lead to the Hollywood GPU of the Nintendo Wii host. ATI added a second console coup in August, when Microsoft

Within three and a half months after the excellent debut of the FX 5800, Nvidia took another photo with the NV35 (FX 5900 and FX 5900 Ultra). The new Detonator FX driver has greatly improved AA and AF and almost matches ATI's solution in quality. But 5900 reached a level that 5800 could not. It beat ATI's Radeon 9800 Pro at the fastest speed, although it costs $499 each, few people actually use it.

As expected, ATI released the 9800 XT in September, thus regaining its bragging rights. Excellent driver support-mainly used in some DX9 games-also makes XT better than Nvidia's similar products, thus ensuring that ATI will win the performance championship by the end of this year. The 9700 Pro is still an excellent mainstream motherboard, and the FX 5700 Ultra, which is priced at $199, has won the price range of less than $200.

After losing US$47.5 million in 2002, ATI rebounded with a profit of US$35.2 million in 2003. A large part of this comes from the higher prices of the dominant 9800 and 9600 cards. At the same time, due to the popularity of FX 5200, Nvidia retained 75% of the DirectX 9 value segment.

The newly established XGI released the successor of Xabre in a staggered release between September and November. The graphics card series is renamed Volari, and the price ranges from the $49 V3 to the dual GPU Duo V8 Ultra. V3 is actually a rename of Trident's Blade XP4 and DX 8.1 parts, while the rest of the series (V5 and V8) were developed from the previous SiS Xabre and have DX9.0 support.

In most cases, except for the entry-level V3 which has the same performance as the GeForce FX 5200 Ultra and Radeon 9200, all other models are under-delivered. The price of Duo V8 Ultra is about 20% higher than Radeon 9800 Pro. 128MB, but still provides the same or lower performance as 9600XT.

XGI's Volari series coexisted with the 8300 at the end of 2005, more or less on par with Radeon X300SE / GeForce 6200, Z9 / Z11 and XP10. The company re-absorbed SiS in October 2010.

S3 is another company that has made a comeback to desktop graphics. After selling the graphics department to VIA for US$208 million plus the company’s US$60 million in debt, the reorganized joint venture focused on chipset projects.

The DeltaChrome desktop card was released in January, but the first S4 and S8 models did not appear in retail channels until December with the long-established S3 method. The new card has most of the new features in 2003; it supports DirectX 9, 16x AF, HD 1080p support and portrait mode display support.

Unfortunately, today's buying public usually regard desktop graphics as two horses, and S3 is not one of the two. While S3 seeks to remain competitive, ATI and Nvidia are pushing each other to achieve ever-increasing performance and image quality.

DeltaChrome was succeeded by GammaChrome in 2005.

Nvidia and ATI continued to introduce interlaced games in 2005. The former released its first GDDR3 card, the FX 5700 Ultra, in March, followed by the GeForce 6 series with the high-end 6800 series. The original lineup includes 6,800 ($299), GT ($399), Ultra ($499), and the overclocked version Ultra Ultra ($549) to counter ATI's X800 XT Platinum Edition. The latter is sold by a series of additional board partners.

The 6800 Ultra 512MB was added on March 14, 2005 and sold for an incredible $899-BFG added an overclocked version for $999. The 6600 series in September satisfies the needs of intermediate frequency well.

The feature set of the Nvidia 6000 series includes support for DirectX 9.0c, shader model 3.0 (although these cards will never fully utilize this feature), Nvidia's PureVideo decoding and playback engine, and SLI support-the previous multi-GPU performance multiplier IP from Obtained by 3dfx.

The implementation of 3dfx resulted in each processing unit being responsible for alternate line scans, while Nvidia handled things in several different ways. The company implemented Split Frame Rendering (SFR), where each GPU renders the upper or lower half of a frame, and Alternate Frame Rendering (AFR) so that the GPU renders the frames in sequence. In some cases, the driver just depends on whether The game supports this feature. The last feature is the initial failure of the driver development.

Although the technology was announced in June, it requires a motherboard with an nForce4 chipset to enable multi-GPU settings, and it didn't start entering retail channels until late November. To add fuel to the fire, the initial driver will occasionally (at most) be released into the second year.

Although Nvidia's SLI was released in June 2004, the required nForce4 motherboard did not appear in retail channels until November, and the initial driver did not appear until the following year.

The reviews at the time roughly reflected current performance, indicating that two lower-tier cards (such as the 6600 GT SLI priced at $398) are usually equivalent to an enthusiast card with lower resolution and image quality. However, in the case of the highest resolution and the application of anti-aliasing, the single-card setting still prevails. The performance of SLI and ATI's CrossFire was as unstable as it is now sometimes, running the entire color gamut from ideal scaling to completely inoperable.

Nvidia’s board partners immediately discovered the marketing opportunities brought about by the reinvented technology. Gigabyte provided dual 6600 GT SLI cards (

), then double 6600 (

) And 6800 GT (

). These cards require not only the nF4 chipset, but also a Gigabyte-branded motherboard.

Among high-end single GPU cards, 6800 Ultra and X800 XT / XT PE are fairly average in terms of price and performance. But they are not without problems. The latter arrived in May and was subject to supply restrictions throughout the production cycle, while Nvidia’s flagship product, the 6800 Ultra, arrived very late in August, and the card was also subject to supply restrictions due to restrictions in the distribution area. Because the card is only provided by a certain percentage of board partners.

The 6800 GT usually beats the X800 Pro at a price of $399, while the 6600 GT is eliminated at a price of $199.

The fierce competition with Nvidia that year did not adversely affect ATI's profits, because that year's profits increased from nearly 2 billion US dollars to a peak of 204.8 million US dollars.

A quirk associated with the acclaimed 6600 GT is that it was originally released as a PCI Express card, when PCI-E was an Intel-specific feature for motherboards designed for the Pentium 4 processor. The gaming performance of these chips usually lags behind AMD products, which of course use the AGP data bus.

Nvidia's 7000 series started rolling off the line before the 6000 series completed its product line. The 7800 GTX was delivered for five full months, and the reduced bill of materials (BoM) 6800 GS has just begun. The first generation of the 7800 series is based on TSMC’s 110nm process G70 GPU, but it quickly gave way to the G71-based 7900 series based on TSMC’s 90nm process.

Although the naming convention is changed from "NV" to "G", the latter is architecturally related to the NV40 series of GeForce 6000. Although only slightly larger than the 334mm² NV40-45, the G70 packs an additional 80 million transistors (a total of 302 million), adding a third of the vertex pipeline and 50% of the pixel pipeline. In most cases, G70 will be replaced within 9 months, while for GS and GTX 512MB, the figures are 3 months and 4 months respectively.

At the entry level, the 7100 GS continues to use TurboCache (the ability of the board to use some system memory), which was introduced in the previous generation GeForce 6200 TC.

On the other hand, the 7800 GTX 256MB was launched on June 22 with a manufacturer’s suggested retail price of US$599, although its actual retail price is higher in many cases. ATI regained the crown of single GPU with X1800 XT, but after 35 days, Nvidia launched a 512MB version of 7800 GTX and quickly won the championship.

Two months later, ATI launched the X1900 XTX and traded with Nvidia's flagship product. This special graphics performance competition resulted in both cards priced at $650. A derivative product of the card that moved to the 512MB frame buffer is that games can now be played on 2560x1600 via dual-link DVI, with 32-bit color and high-level image quality.

ATI announced its Doka Crossfire technology

The Xpress 200 Crossfire Edition chipset and X850 XT Crossfire Master board were launched in September to make them available. Due to the single-link TMDS, the resolution and refresh rate were initially limited to 1600x1200 @ 60Hz, but the dual-link TMDS for 2560x1600 will soon replace it.

Unlike the Nvidia solution where two identical cards communicate through a bridge connector, ATI has implemented a master card with a TMDS receiver, which accepts the slave card's communication through an external dongle and Xilinx composite chip enter.

Like Nvidia's SLI, CrossFire provides alternative frame rendering (AFR) and split frame rendering (SFR), but also provides a rendering technology called SuperTiling. The latter improves performance in some applications, but does not apply to OpenGL or geometry processing that does not support accelerated. Like SLI, Crossfire also faces driver-related troubles.

ATI plans to prepare R520-based cards from June to July. This is their first card using Shader Model 3.0, but it was delayed by 4 months due to errors found in the unit library.

The initial release includes X1800 XL/XT using R520 core, X1300 budget card using RV515 and a quarter of the graphics pipeline of RV530, and X1600 Pro/XT based on RV530, similar to RV515 but with higher shaders And the ratio of the vertex pipeline to TMU and ROP.

Due to the initial delay of R520, the GPU and its derivatives were only replaced by the R580-based X1900 series after three and a half months, which used TSMC’s new 80nm process. After the continued launch, half of the graphics pipeline resources went to the RV570 (X1650 GT / XT and X1950 GT / Pro), and the reduced RV530 became the RV535 that powers the X1650 Pro and X1300 XT.

ATI’s annual revenue reached a record US$2.2 billion, a record high in the company’s history, thanks to Xbox 360’s Xenos GPU shipments. However, net profit fell to $16.9 million.

At this stage, if there is no enthusiasm, all graphics cards received that are not based on Nvidia or ATI GPUs will receive a certain curiosity. This was the case when S3's overhauled graphics lineup debuted in November.

The high clock frequency of Chrome S25 and S27 guarantees excellent gaming performance, but mainly provides substandard products. The initial price is 99 US dollars (S25) and 115 US dollars (S27), making it compete with Nvidia's 6600/6600GT and ATI's X1300Pro/X1600Pro, but apart from power consumption, neither of these two S3 cards can be used with any The way of meaning stands up to competition. As ATI/AMD and Nvidia solve the problems of HTPC and entry-level market, this weak advantage gradually disappeared, effectively stifling the Chrome 400 and 500 series that S3 subsequently launched.

Another problem with S3 is that the cost of manufacturing the card has resulted in meager profits. The company needs to achieve high-volume sales in a market dominated by two suppliers. HTC will acquire S3 for USD 300 million in July 2012, a move that was initially seen as a leverage in the separate legal dispute between HTC and S3 and Apple.

Nvidia and ATI continued to maintain media coverage in 2006.

ATI acquired Macrosynergy, a design and engineering center in Shanghai with personnel working in California and formerly part of the XGI Group. Then in May, the company bought BitBoys for $44 million.

At the same time, after ATI, 3dfx and XGI, Nvidia first set foot in dual-GPU single-board products. The 7900 GX2 can sandwich two custom boards, which are basically equipped with a pair of down-converted 7900 GTX. However, Asus did not wait for Nvidia's dual GPU solution, but released its own product.

($900, 2000 units built), but paired with two 7800 GT GPUs.

This graphics card has aroused ASUS’s interest in limited-edition dual-GPU motherboards, and may consolidate Nvidia’s attitude towards motherboard partners, because ASUS products stand out from its reference models when they are released.

In the high-volume mainstream market, 7600 GT and GS both provide excellent performance and excellent service life, while ATI's X1950 XTX and Crossfire have become the benchmark for high-end enthusiasts with single GPU cards. The combination of X1900 XT and GeForce 7900 GT in high-end mainstream brackets is fairly even.

As an independent company for 21 years, ATI was acquired by AMD on October 25, 2006 for a total price of US$5.4 billion-including US$1.7 billion from AMD and US$2.5 billion borrowed from lending institutions. 57 million AMD shares and 11 million options/restricted stock units valued at $1.2 billion. At the time of the buyout, approximately 60-70% of ATI’s chipset/IGP revenue came from partnerships with Intel-based motherboards.

Two weeks after ATI's acquisition, Nvidia ushered in the era of a unified shader architecture for PC graphics.

As a large part of the Intel IGP chipset market moved to Nvidia, market share dropped sharply. The logic behind the acquisition seems to be a shortcut to adopting GPU technology, rather than spending $5.4 billion to develop AMD’s own IP and add licensing technology when needed. At the time, AMD’s goal was to quickly introduce

And related Fusion projects.

Two weeks after ATI's acquisition, Nvidia ushered in the era of a unified shader architecture for PC graphics. ATI’s Xenos GPU for Xbox 360 has introduced a unified architecture to the console.

In the fourth and last part, we will summarize the development of AMD’s Radeon products, the continued competition between GeForce and Radeon CPUs, the transition to stream processing, and the current prospects of graphics processors.

TechSpot is a registered trademark.

©2021 TechSpot, Inc. All rights reserved.