-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
- Latest in Graphics
Best Picks Best Graphics
by Chris Angelini
Slideshow The 30 Year History of AMD Graphics, In Pictures
by Michael Justin Allen Sexton
Slideshow The History Of Nvidia GPUs
by Michael Justin Allen Sexton
News Are FreeSync TVs On The Way?
by Derek Forrest![Nvidia]()
News Nvidia's Tesla P4 And P40 GPUs Boost Deep Learning Inference Performance With INT8, TensorRT Support
by Lucian Armasu
News Mixed Reality Is Going Hollywood: The Future Group Unveils Frontier Interactive Mixed Reality
by Kevin Carbotte
Introduction
Picture 1 of 23Matrox Impression
Picture 2 of 23Matrox's first graphics cards were introduced in the late 1970s, but they were relatively simple adapters. The company's first product capable of handling 3D images was the Impression, which had to be accompanied by Matrox's Millennium 2D accelerator. Its overall performance was lackluster, as the Impression wasn't really designed for gaming. Rather, it was intended for CAD work. As a result, the Matrox Impression didn't really, well, make an impression on enthusiasts.
Matrox Mystique
Picture 3 of 23Matrox introduced its Mystique in 1997, combining 2D and 3D acceleration on a single card. This helped pare back costs since you only had to buy one board instead of two. The Mystique was also a more competent gaming product thanks to features like texture mapping. The 64-bit graphics processor came equipped with a single pixel pipeline and one TMU.
It still lacked several key abilities, however, such as mipmapping, bilinear filtering, and transparency support. The Mystique shipped with anywhere from 2 MB to 8 MB of SGRAM. Memory on the 2 MB cards could be upgraded using add-on modules.
Despite improved performance, the Mystique's graphics quality was called into question. A later version, dubbed the Millennium II, included WRAM, which helped make the card faster than its predecessor.
Matrox G200 & G250
Picture 4 of 23Matrox's second 2D/3D graphics accelerator was far more successful; it offered several new features including full 32-bit color support, mipmapping, trilinear mipmap filtering, and anti-aliasing. The G200, released in 1998, used an internal 128-bit design with dual 64-bit unidrectional buses. The memory interface was 64 bits wide and accommodated between 8 and 16 MB of RAM. Although the G200 continued to employ a single pixel pipeline with just one TMU, Matrox's improvements allowed the card to far surpass its older Mystique. Because the company adopted 32-bit color at a time when the competition was mostly limited to 16-bit, the G200's image quality was often lauded for its lack of dithering artifacts.
The first Matrox G200 chips were built using 350 nm transistors, but Matrox later transitioned the design to a 250 nm process in 1999. This helped facilitate higher clock rates. Variants based on the 250 nm technology were sold as either G200A or G250. The G250 models offered the most aggressive frequencies, and thus the best performance.
Matrox G400
Picture 5 of 23The G400, released in 1999, was essentially a G200 graphics processor with twice as many resources. Instead of the 128-bit architecture complemented by two 64-bit buses, Matrox's G400 used a 256-bit design with two 128-bit unidirectional pathways. Matrox also added a second pixel pipeline with its own texture unit. The memory interface grew to 128 bits wide, and it supported twice as much RAM (from 16 to 32 MB). Clock rates increased significantly, while the processor's feature set grew to be DirectX 6-compliant. This was also one of the earliest cards to enable simultaneous video output to two displays.
Unfortunately, the G400 suffered from rather serious driver issues upon its introduction, limiting performance in the first reviews. Over time, the company resolved its issues with a stable OpenGL ICD and improved DirectX compatibility.
Matrox manufactured the G400 on a 250 nm process, but later transitioned to 180 nm transistors. Those smaller dies went into G450 cards, which were less expensive than the original G400. Matrox didn't increase the G450's clock rate, however, so its impact was fairly limited. A later version known as G550 added even more gaming-oriented functionality.
Matrox Parhelia-512
Picture 6 of 23Matrox's final enthusiast-oriented graphics card was the Parhelia-512, released in 2002. Its design was rather ambitious, featuring four pixel pipelines, each with a vertex shader and four TMUs. The GPU was fed by a 256-bit ring bus connected to 128 MB or 256 MB of DDR memory. Matrox claimed that the ring bus topology allowed the 256-bit memory interface to operate as if it was 512 bits wide. It typically operated between 200 and 275 MHz, was fully compatible with DirectX 8.1, and it had partial support for DirectX 9.
The Parhelia-512 far surpassed Matrox's older cards. However, it was slower than contemporary ATI and Nvidia GPUs. It was also rather expensive, and ultimately failed to gain much market share.
A less expensive version known as the Parhelia-LX was also produced, but it had half of the Parhelia-512's resources and wasn't particularly competitive. The full Parhelia-512 die moved to 90 nm manufacturing in 2007 and was re-released as a budget-oriented card with DDR2, but again, it didn't make much of an impact.
Following the Parhelia-512, Matrox exited the enthusiast graphics market. Though the company is still in business today, it focuses on more specialized applications.
Rendition Vérité V1000
Picture 7 of 23Rendition entered the market in 1996 with its Vérité V1000, which shipped with full 2D and 3D acceleration features. Back then, the Vérité was one of the fastest graphics cards you could find. It wasn't able to beat 3dfx's Voodoo. However, it did offer more value since the Voodoo card was limited to 3D gaming and required a companion 2D controller.
The graphics processor itself was based on a RISC design, and it connected to up to 4 MB of EDO RAM over a 64-bit bus. The card unfortunately ran into compatibility issues with several motherboards and its 2D performance trailed competing products. These factors prevented Rendition from taking a larger share of the market with its V1000.
Rendition Vérité V2100 And V2200
Picture 8 of 23Rendition's Vérité V2100 and V2200 were nearly identical to the V1000 at a __hardware level. The company did improve the V1000's design, however, and nearly doubled its processor's fill rate. These cards operated at slightly higher clock rates and used faster memory, yielding modest performance improvements. The V2100 and V2200 differed from each other in two key areas: the latter offered higher frequencies and came with either 4 or 8 MB of SGRAM, while the former was slightly slower and limited to 4 MB of SGRAM.
After the release of its Vérité V2100 and V2200, Rendition attempted to design a new graphics processor. But the design was delayed several times and Rendition was eventually purchased by Micron.
Image Credit: VGAMuseum.info
S3 Trio3D
Picture 9 of 23S3 entered the graphics card market in the early 1990s. It produced 2D accelerators for several years, culminating in the Trio series. Although that family mostly consisted of 2D cards, its final incarnation was the Trio3D, a 3D-capable board with a 128-bit processing engine. It was complemented by up to 4 MB of RAM.
S3 ViRGE
Picture 10 of 23S3 introduced its ViRGE line in 1995, a brand it continued to use until 1998. The ViRGE was capable of handling both 2D and 3D acceleration. Early models had access to either 2 or 4 MB of EDO RAM over a 64-bit bus, but later variants shipped with 8 MB. Unlike many graphics cards of this time, the ViRGE was a single-chip solution. This made it less expensive and easier to produce. But in order to fit all of the necessary __hardware into one die, S3 devoted fewer resources to 3D processing. As a result, the ViRGE wasn't a very fast gaming card.
Although its 3D abilities were unimpressive, the ViRGE had outstanding 2D performance and it came to be known as one of the fastest 2D accelerators you could buy.
S3 Savage3D
Picture 11 of 23The S3 Savage3D was introduced in 1998. It leveraged an entirely new design and possessed advanced features like trilinear filtering, a 24-bit Z-buffer, hardware-accelerated motion compensation and alpha blending for MPEG-2 video decoding, an integrated TV encoder, and texture compression. S3 implemented a dual-pipeline design. The pipelines were somewhat unusual in that one was used exclusively for rendering while the other was dedicated to texture processing. At most, the Savage3D could ship with 8 MB of RAM, which was less than competing cards.
Although the Savage3D was feature-rich, it was ultimately a failure. The Savage3D's performance lagged far behind its competitors, and S3 faced serious yield challenges that limited sales.
S3 Savage4
Picture 12 of 23S3 continued to improve on the Savage3D's design, which ultimately led to the creation of Savage4. Many of the architecture's inherent problems were fixed, and by transitioning to 250 nm process technology, S3 was able to manufacture the Savage4 with relative ease.
When the Savage4 shipped in 1999, it faced strong competition from Nvidia's Riva TNT2 and 3dfx's Voodoo3. It was significantly slower than those other cards, but it offered a strong feature set and was typically sold at budget-friendly prices.
S3 Savage 2000
Picture 13 of 23Later in 1999, S3 and well-known OEM Diamond Multimedia merged together to form a new company called SONICblue. S3 wanted to avoid the supply issues it faced with its other Savage graphics cards by using Diamond as a dedicated manufacturer. The two companies also wanted to increase profits from sales, which would ordinarily be split between S3 and several OEMs.
The joint venture's first product was the S3 Savage 2000. This new design featured dedicated transform and lighting hardware, which ultimately proved to be broken. It also used a faster quad-texture engine capable of placing four textures per clock cycle. The card featured numerous other enhancements to increase performance, and also came with 32 MB of SDRAM.
Performance tests showed that the S3 Savage 2000 was significantly faster than its predecessor. At times, it was even quick enough to catch Nvidia's flagship GeForce 256. The card had serious driver issues, however, and suffered from a few other defects that eroded its chances against more stable competition.
S3 Chrome
Picture 14 of 233dfx Voodoo
Picture 15 of 23The first 3dfx graphics card shipped in 1996 under the Voodoo brand, and it quickly became known as the fastest 3D accelerator available. One key advantage that 3dfx wielded was its Glide API. Because Glide was written exclusively to access 3dfx's products, it gave developers low-level access to the hardware, which naturally helped maximize performance. It also was less dependent on other platform components like host processing and memory.
Voodoo's only notable disadvantage was that it only accelerated 3D, meaning you needed a separate 2D controller as well.
3dfx Voodoo Rush
Picture 16 of 233dfx created its Voodoo Rush in 1997 by combining the Voodoo's 3D hardware with a third-party 2D controller on one card. 3dfx made deals with several companies that produced 2D accelerators, and as a result there was a lot of variation between Voodoo Rush cards.
Although the 3D hardware carried over from the original Voodoo's frame buffer processor and texture mapping unit, tests showed that the Voodoo Rush was typically slower. Part of this was attributable to the 2D and 3D subsystems sharing memory, which cut into the 3D accelerator's available bandwidth. But the Voodoo Rush also required more complex drivers, and an increase in overhead didn't help either.
3dfx Voodoo2
Picture 17 of 233dfx's Voodoo2 takes after the original Voodoo card, in that it is only capable of processing 3D graphic. The card used a single pixel pipeline the with two TMUs connected to between 8 to 12MB of vRAM. When it launched, the Voodoo2 was by far the fastest 3D accelerator on the market, and it would keep the performance crown for quite some time.
It should be noted that the Voodoo2 was also probably the most expensive graphics solution for gamers in 1998. The Voodoo2 was relatively expensive by itself, but it also required you to own a third-party 2D-accelerator for 2D games and Windows acceleration.
3dfx Voodoo2 SLI
Picture 18 of 233dfx Voodoo Banshee
Picture 19 of 23In 1998, 3dfx took another swing at the 2D/3D combo card, calling it Voodoo Banshee. The company pushed for greater integration. Its resulting single-chip solution contained a 2D engine developed in-house and most of the Voodoo2's hardware. The chip's 3D pipeline was limited to a single texture mapping unit for size reasons, but the other hardware units were exposed.
Greater integration made the Banshee more affordable than 3dfx's Voodoo2. And in single-textured tests, it kept up with the 3D-only card. More complex scenes weren't as forgiving though, and the Voodoo2's second TMU demonstrated its advantage. Still, the Banshee was regarded as a capable card.
3dfx Voodoo3
Picture 20 of 23In 1999, 3dfx acquired STB Systems and implemented a new strategy. Its idea was to use STB's considerable manufacturing resources to produce 3dfx cards exclusively. The company would then sell its boards directly, enjoying greater margins in the process. This was similar to what S3 tried, and it worked about as well. Following the acquisition of STB Systems, 3dfx lost its board partners to Nvidia.
The first graphics card produced by 3dfx following its change in direction was the Voodoo3, which looked a lot like the Voodoo Banshee. The biggest difference between them was the Voodoo3's second TMU. 3dfx managed to operate the card's rendering hardware at higher clock rates than the Voodoo2, however, and it typically shipped with 16 MB of RAM. These factors allowed the Voodoo3 to nearly match the performance of two Voodoo2s in SLI. It was still limited to 16-bit color though, and due to strong competition from ATI and Nvidia, it would be 3dfx's last successful product.
3dfx Voodoo4
Picture 21 of 233dfx Voodoo5
Picture 22 of 233dfx's VSA-100 was also used on the Voodoo5, which actually surfaced before the Voodoo4. The Voodoo5 5500 contained two VSA-100 processors connected via on-card SLI. This had the effect of increasing performance, but also made the Voodoo5 relatively expensive.
Only the Voodoo5 5500 made it to market, and there weren't many of them sold either. 3dfx planned to create more models, including a Voodoo5 6000 with four VSA-100 chips. But it was cost-prohibitive and never saw the light of day. Not long after, the company went bankrupt and was sold to Nvidia.
No comments:
Post a Comment