Review: VideoLogic's Vivid! Graphics Card

The Vivid! graphics card is based on KYRO, which is a PowerVR Series3 graphics processor manufactured by STMicroelectronics, but designed by VideoLogic (div. of Imagination Technologies). KYRO is the first in a series of chipsets based on the PowerVR Series3 architecture. An update to the architecture called KYRO 2, was released with the exact same graphical features as KYRO 1, but running at a clock rate of 175 MHz. More details on that, can be found here.

Click to Enlarge Click to Enlarge

GPU core clocked at 115 MHz 32 MBytes* of single data rate SDRAM clocked at 115 MHz 128-bit memory data path 1.85 GB/s memory bandwidth 230 megapixels/sec or 230 megatexels/sec 20 million polygons per second (peak) 12 million transistors STMicroelectronics 0.25 micron process fast 128-bit 2D engine AGP 4X/2X 270 Mhz RAMDAC *Note: There are also KRYO based boards that contain 64 MBytes of single data rate SDRAM.

Click to Enlarge

PowerStrip program showing Vivid!'s core clock rate and memory clock rate. You can download PowerStrip yourself from EnTech's website.

3D Features

• Effective fill rates increase with 3D complexity (around one gigapixel at a depth complexity of four) 
• Flat and Gouraud shading 
• Perspective texturing and shading 
• Specular highlights 
• D3D Environment Mapped Bump Mapping 
• DirectX Texture Compression (DXTC) 
• 8 Layer Multitexturing support 
• 32-bit Z/Stencil buffer 
• Internal True Color™ 32-bit ARGB internal rendering and layer blending· Full tile
• blend buffer 
• Z load/store mode 
• Table and per vertex fog 
• Palletised textures 
• 16-bit textures 
• 32-bit textures 
• YUV and DXT textures 
• Point, bilinear, trilinear and anisotropic filtering 
• Full range of OpenGL and D3D blend modes 
• Alpha test 
• Pixel-perfect full scene anti-aliasing (FSAA)

Besides the Vivid! video card, you get two CD-ROM's:
 

1

CD-ROM

Driver CD contains the Vivid! card's drivers and also contains:

Game Demos 

• Sierra's SWAT 3 
• Eidos Interactive's Tomb Raider: The Last Revelation
• EA Sports FIFA 2000
• Attention to Detail's Rollcage Stage II
• Eidos Interactive's Urban Chaos
• Neversoft's Tony Hawk's Pro Skater 2

Technology Demos

• Fortune - demonstrates fill rate with lots of overdraw present using a stack of playing cards
• House - illustrates PowerVR Series3's 8-layer multitexturing by applying lots of layers to a house
• Dragonfly - the wings of a dragonfly are used to illustrate motion-blur effects
• Scanner - illustrates modifier volumes effects that can be implemented by use of stencils
• Wheel - shows bump mapping
• Vase - shows one of the several uses of multi-texture blending
• VillageMark - showing how the Vivid! card handles lots of depth complexity in a demo setting by rendering a village

Applications:

• E-Color's Colorific - gives you precise, predictable color matching between your screen and color peripherals such as printers, scanners, and digital cameras by creating an accurate profile of your monitor
• 3Deep - provides colour and lighting correction to optimise your 3D gaming environment. 3Deep eliminates dark or washed out graphics.
• True Internet Color - delivers accurate color for all your online images
• Microsoft's Netmeeting Videoconferencing - provides the link between your Internet connection, a video capture card and a video camera to turn your computer into a video phone
• Winzip 8.0 - compression software

InterVideo's WinDVD 2000 software player that provides support for MP3, VCD, MPEG1, and MPEG2. The WinDVD 2000 software uses the KYRO chipset hardware MPEG2 decode acceleration with motion compensation; 4:2:0 Overlay support; sub-picture blending; X, Y interpolated scaling and color keying. This software was not tested, due to not having a DVD-ROM drive.

2D Core

The KRYO chipset has an excellent 2D core as these results show at Tom's Hardware Guide.

PowerVR Technology Overview
 

PoverVR Technology is quite different than traditional rendering architectures, and is more or less best defined as a rendering architecture that only renders what the viewer can see. A traditional graphics chip renders all polygons in a scene, even those polygons sitting behind other opaque polygons. The traditional way of rendering wastes a lot of fill-rate on hidden surfaces.

Infinite Planes
 

The key to PowerVR's rendering efficiency is the use of an algorithm to determine all front facing polygons and transform those surfaces into "infinite planes". It's this procedure that eliminates all the hidden surfaces in a scene. As you can see in the example on the left, there is quite a lot of hidden surfaces that should not be rendered. The PowerVR way of rendering a scene will get more important with time, as games are getting more complex with more objects in any given scene, and each surface in a scene is increasing in the amount of multi-texturing, shading, and lighting used. Another important advantage of infinite planes, is that a Z-buffer is no longer needed, thus giving a savings in memory space and memory bandwidth. This becomes more important as the resolution increases.

Tile Rendering
 

The PowerVR architecture also breaks up scenes into "tiles", and renders each tile in a buffer on-chip. This allows for more efficient use of the memory bandwidth, as the tile can then be written to memory more efficiently in blocks.

Important Graphical Features

Environment-mapped bump mapping

Click to Enlarge

KYRO provides efficient Direct3D Environment Mapped Bump Mapping (EMBM), which increases the realism of 3D rendered scenes by creating an illusion of depth on a surface. KYRO’s EMBM uses per-pixel operations to represent the greater detail on texture surfaces. As KYRO's EMBM is fully Direct3D compliant, developers are able to use it to create a wide range of surface texturing effects.

Full Scene Anti-Aliasing (FSAA)
The KYRO chipset is capable of 2X and 4X full-scene anti-aliasing (FSAA). It's the on-chip mini-framebuffer (tile size: 32x16 pixels) that allows for 2X and 4X full-scene anti-aliasing (FSAA) with no additional overhead. KYRO's 4X FSAA, means that the source image is four times the size of the display image. It's the downsampling the gets rid of the annoying artifacts like sharp contrast pixel edges (jaggies), texture shimmering, and moire patterns resulting in a smoother more pleasant looking display.

Internal True Colour
KYRO does all internal color calculations at 32-bit color depth, even if the source is 16-bits. When many layers of color information is blended togeather in a 16-bit pipeline, you lose information due to lack of accuracy, but with KYRO's Internal True Color feature, the best possible image quality is retained no matter how many layers have to be blended togeather. This is very important for transparencies, as rendering explosions, fire, etc., can result in many transparency layers that have to be blended togeather. It is also important for multi-texturing on opaque surfaces. 

Here is a page at Sharky Extreme's KYRO review that shows the difference between KYRO's Internal True Color 16-bit output, and 16-bit output from a Geforce2 MX chipset.

8 Layer Multi-texturing
KYRO allows 8 layers of texturing in a single pass (multi-cycle though). Information below is from the technology overview section at powervr.com on 8 layer multi-texturing: 

Performance Increase due to Reduced Bus and CPU Loading
Multi-pass rendering on traditional hardwares will require the polygon data to be sent to the HW several times. KYRO's multi-texturing only requires the polygon data to be sent once. As a result system bus bandwidth and CPU loading are much lower with KYRO's multi-texture hardware.

Performance Increase due to Reduce Z-buffer and Framebuffer Access
Multi-pass rendering on traditional 3D accelerators involves blending the current primitive onto the previous primitive contained in the Z-buffer and framebuffer. Additional Z-Buffer Read/Write and Framebuffer Read/Write memory accesses are necessary, reducing performance and consuming fill rate. KYRO's multi-texturing support renders interim stages into the internal tile buffer eliminating the need for external memory accesses until the tile is complete and ready to be written to the framebuffer. 

Includes Windows 9X/ME, NT4.0 and Win2000 display drivers.

Vivid! driver used: version 1.00.04.0044 dated 10/10/2000 (drivers that came with the  card)

Installation of the drivers was an easy affair after the installation of the card in the AGP slot. Both OpenGL and D3D drivers that come with the card are extremly stable, and run every game that I threw at them over the past month. 2D performance is very fast also, with good color presented. Gamma correction is also provided.

Excellent support is provided by VideoLogic by providing a method off of the Windows Start menu to check for new drivers as shown below.

Vivid! Start Menu Options

The above clearly shows an impressive set of utilities and support options for this card!

Benchmarks

System 1

• AMD K6-2 500 MHz
• EPoX EP-MVP3G2 (VIA MVP3 100MHz) motherboard with 2X AGP slot
• 128 MB 66 MHz SDRAM
• Vivid! 32 MB AGP video card
• XWave QS3000A (Yamaha XG) sound card
• Windows ME with DirectX 7.0

• AMD Athlon XP 1600+ (1.4 GHz) and AMD Duron 600 MHz
• Elitegroup (ECS) K7S5A (SiS 735) motherboard
• Micron 256 MB 133 MHz DDR SDRAM, PC2100
• Vivid! 32 MB AGP video card
• AC97 Sound Codec on motherboard
• Windows ME with DirectX 8.0a

Note: sound was on for all benchmark tests, and used the driver that was released with the card.

Quake III Arena (demo version)

A lot people are tired of seeing this game as a benchmark, but it is a good benchmark, as it has quite high polygon and fill-rate requirements. The benchmarks popularity allows these results to easily be compared with other CPU's and graphics cards.
 

Quake III Arena 640x480 32-bit High Quality FSAA

Run benchmark "demo001" by:
1) In game options, turn on everything except "Sync every frame"
2) Open console by pressing "~" key
3) Enter: timedemo 1
4) Enter: demo demo001
 

As you can see with Quake III the Vivid! is great for a K6-2 500 MHz or Duron 600 MHz, but clearly is fill-rate limited on a Athlon XP 1600+ (1.4 GHz).
 

Full Screen Anti-aliasing (FSAA) was on for both horizontal and vertical axis. FSAA was only tested at low resolution, as aliasing errors are less noticable at high resolutions. Performance is quite good with FSAA on, but my personal preference is playing at 1024x768 with it off.

FSAA cleans up more than aliased edges (jaggy edges) as it gives the whole scene a smoother look, and also gets rid of flickering pixels (usually white pixels). This makes it worth while to use FSAA for resolutions 1024x768 and above, but of course only if the video card can sustain reasonable frame rates at those resolutions.

Q3A Conclusion:
Well adding an Athlon XP 1600+ (1.4 GHz) does very little to Quake III over a Duron 600 MHz as clearly the video card is becoming the limit of the system.

Notice also that the frame rate drop from low quality to high quality is very small, yet the visual quality difference is huge, due largely to the texture size differences. I see no reason to play in low quality mode, in order to save a few frames per second. Here is comparison between the two modes:
 

640x480 32-bit Low & High Quality FSAA

Quake III Arena is very good for showing Vivid!'s 16-bit quality, as it looks as good as 32-bit quality. Other graphics chips would be hard pressed to make that claim.

Dagoth Moor Zoological Gardens
 

DMZG 640x480 32-bit FSAA

Dagoth Moor Zoological Gardens is a technology demo and a benchmarking tool developed by WXP to originally demonstrate the GeForce 256 3D accelerator card. It is not a game, nor a demo of a game. This technology demo runs on DirectX 7.0, and as you can see runs just fine on the Vivid!.
 

At low resolution, the demo shows a huge increase between the processors, but at high resolution very little increase between the Athlon and Duron, as the video card is limiting the frames per second with lack of fill-rate.

The fact that this demo shows a huge increase between the Athlon and the Duron at low resolution compared to Quake III, is because the game is very CPU intensive with it's huge amount of polygons.
 

Full-scene anti-aliasing (FSAA) at 1280x960 means that the game was rendered at 1280x1920 internally. The reason why the X axis is not doubled is because the KYRO driver that is currently available turns off FSAA for any axis that has a resolution over 1024. So only FSAA on the Y axis here. 640x480 with no FSAA versus 1280x960 with Y-axis FSAA is an 8 times difference in load (pixel writing).

Interesting results as the Duron score is lower than the K6-2. Not much improvement when the Athlon is added to the mix, as the card is more or less hitting it's fill-rate limits with the K6-2. Not a very practical benchmark, as I doubt anyone will really care about running this demo at 1280x960 with FSAA on, but interesting to see, and will be used for comparison purposes with future KYRO cards, like maybe KYRO 3.

3DMark2000

This benchmark has been heavily criticized by a lot of people as being too heavily CPU dependent, as it puts more emphases on polygon transformation than fill-rate. You have to consider that this benchmark was not designed just for testing video cards, but was designed to test your system as a whole, as games are not just increasing in fill-rate, but are also increasing in polygons.
 

The criticism mention above is fully justified, as the benchmark is very much CPU limited at lower resolutions, but at higher resolutions, the video card is holding the score back, even though it is still showing an increase between each CPU, but just a smaller increase. The continual increases at either resolution between the different CPU's shows this to be a very good benchmark, as it is not CPU limited or even fill-rate limited yet.

3DMark2000 is available at MadOnion.com. It is a very nice benchmarking tool, which offers a lot of options and detailed reports that can be used to compare against other systems online at MadOnion.com's site.

3DMark2001

Decided to add 3DMark2001 benchmark to the review, as it is a popular benchmark. Can only include the Athlon XP 1600+ (1.4 GHz) scores, as that is my current system. These scores can be used to compare with other systems at MadOnion.com.
 

Those are half decent scores, but what is not listed, is that the Vivid! card is showing it's age a bit, as there are parts of this benchmark that will not run, as they require DirectX 8 hardware, like Pixel Shaders. Of course that is not really an issue with games yet, as even games that use DirectX 8.0 hardware features, run just fine, as the card is DirectX 8 compliant.

Conclusion

Well the card is still holding it's own, after a couple CPU upgrades, and is more than adequate for the number of games that I play on my computer. With fill-rate heavy games like Quake III, you are not going to see much an improvement with a processor over a 600 MHz Duron, but with games that require lots of CPU power, the card is doing just fine.

This card comes with solid OpenGL and D3D drivers (download DirectX 8.0a), and provides about the best 16-bit quality/performance output around.

Been playing a lot of Tribes 2 (demo), and Serious Sam - The Second Encounter at 800x600 with all graphic features set to max, and plays very smooth with no graphical abnormalities.

Other Cards

The Vivid! 32 MB AGP video card by VideoLogic is not the only card using the KRYO 1 chipset, as there are a number of others, and that can be had for well under a $100 US. You can find other cards listed at www.pricewatch.com. Just type "kyro" in their search engine.

Links

VideoLogic's site
Imagination Technologies site
PowerVR.com website
STMicroelectronics KYRO section