Please consider a donation to the Higher Intellect project. See or the Donate to Higher Intellect page for more info.


From Higher Intellect Vintage Wiki
Jump to navigation Jump to search

Dual Pipelines

"Multi-texturing" is a technique used by many of the latest game engines to apply two or more textures to a single pixel - such as a dynamic lightmap illuminating a textured wall. Traditionally this requires a graphics chip to make multiple passes, redrawing the entire screen several times per frame.  Not only does this hinder performance, but if the framebuffer is using 16bit color visual quality can suffer severely as well. The lack of precision in the frame buffer between passes results in greater and greater distortions as every new layer is applied. The Rendition RRedline has two internal texturing pipelines, allowing it to render a dual-textured pixel in a single clock cycle with a single drawing pass. Internal precision is maintained at 32 bits-per-pixel to maintain the maximum possible visual quality.
This is a similar technique to that used by a well known 3D-only graphics chipset.  But unlike that product, the RRedline's dual pipelines can also work independently - rendering two single textured pixels per clock, doubling performance of games and applications not written with multi-texturing in mind. The RRedline's second pipeline never gets left to waste.
And unlike other multi-texturing capable chips that only support dual-texturing while bilinear filtering, the RRedline actually can support dual-textured trilinear filtering. There is no need to sacrifice the beauty of trilinear filtering to get the performance and quality advantages that multi-texturing gives. The RRedline can do it all.  The texture units of the RRedline are both full featured. Unlike other architectures, the second unit is not a trimmed down version of the first but is also fully capable of trilinear filtering and more. Each texture unit can be independently working on input data of any format, and the output can be blended together in a very flexible, programmable fashion. This allows the RRedline to draw single-textured bilinear or trilinear filtered pixels, dual-textured per-pixel mipmapped pixels, dual trilinear filtered pixels, or even to have one texture that is trilinear filtered and another that is bilinear mipmapped. This allows rendering a bilinear filtered lightmap onto a trilinear filtered wall in one pass!
Multi-texturing in hardware is a central feature of Microsoft's DirectX 6, and it is already found in Quake 2 and Unreal engine games. To get the best possible quality and performance out of future games and applications, you will need hardware that can support multi-texturing. And the RRedline will do it better than anybody else.


Two Heads Are Better Than One

You can never have enough screen real-estate. We've evolved from 640x480 up to 1024x768, and even 1280x1024. Some have even pushed to 1600x1200 (if you can afford the monitor, or like squinting). But it  isn't enough - we still hunger for more space to shuffle our windows around in. One monitor just can't satisfy. So why not have two? The Rendition RRedline is the first mainstream graphics accelerator to provide built in support for two displays. The RRedline's dual independent CRT controllers each have a 250 MHz RAMDAC, allowing it to drive two displays without the need for multiple graphics boards or any special external electronics. No other chip is so flexible. The RRedline can support the following board configurations:
        1. Dual Monitor
        2. Monitor and TV
        3. Monitor and LCD Screen
        4. LCD Screen and TV
        5. Dual LCD
            And so on...
Unlike other boards which support TV output - the CRT controllers of the RRedline are truly independent. You can display the same image on both screens simultaneously, or you can use Windows 98's multi-mon feature to extend your desktop across both displays. The potential is awesome! The resolution, refresh rate, and color depth of each CRT controller is totally independent. You can have one monitor displaying your desktop at 1600x1200x16bpp at 85 Hz, and the second monitor playing back a DVD full screen at 740x480x32bpp at 60 Hz. Or why not go for the max usable area - you can crank both screens up to 1600x1200 (or more!) for a total desktop size of 3200x1200. That should satisfy your craving for real-estate for a while.
The RRedline opens up new realms of flexibility and power for a wide range of applications. Just imagine:
Games -  Imagine the killing power of having a game's main view on one screen, and a tactical map or rear view on the other. Both screens can be fully 3D accelerated!
Video Production - You can redirect a window on your desktop to be full screen on the second display - imagine being able to view a playback window from your video editing software simultaneously in a window on your desktop and full screen on a TV!
Programming - Keep your editor window open and handy while you step through your app on the second display.
3D Modeling - Imagine how much easier model design would be if you could have more 3D views open on the screen at once.
Web Surfing - Surf in style with two screens worth of windows open. Or keep one screen surfing while you work in the other.
PC TV - With a TV-tuner card, you can keep your on-screen TV-guide on the monitor, while watching your favorite show on an attached TV. Or keep up with team stats while watching the game!
Word Processing - Keep your word processor unobscured on one screen, while you do your research on the other.
And More! - Having two screens can totally change the way you use your computer. Let your imagination run wild!

2x64 > 128

Wider is better, right? The width of the bus that attaches a graphics chip to its local memory has been swelling over the years. First 32, then 64, and now 128 bit busses have become common. Obviously, a wider memory bus means that more data can be pumped per memory clock cycle. This is good. But getting fat isn't the only way to improve memory bandwidth.

The Rendition RRedline  features a true 128 bit wide memory bus, but the RRedline takes things a step further.  By breaking this wide bus down into dual independent 64 bit wide channels, the RRedline can achieve real-world memory performance 50% faster than a 128 bit bus alone.

To understand how this is possible, consider the memory demands of a typical 3D game. Five separate areas of memory need to be accessed at once:

Front Buffer - Accessed by the RAMDAC to draw the screen

Back Buffer - The target for 3D rendering

Z-Buffer - For checking and updating depth information

Texture0 - A texture map

Texture1 - A lightmap being dual-textured onto Texture0

A single 128 bit bus is left dizzy by these demands, constantly having to reverse itself between reading and writing. It is only able to get at one piece of memory at a time. The RRedline on the other hand can read in texture data from one channel while simultaneously writing data to the frame and z-buffers in the other. There is much less need for time consuming bus turnarounds.

The RRedline's dual 64 bit channels present another advantage as well. With the current graphics memory DRAM's available on the market, a 128 bit bus is limited to two logical banks of local memory. But when split into two 64 bit channels, standard memory configurations give the RRedline access to four banks. Having more memory banks lets the RRedline avoid many of  the "page misses"   that slow memory access down. Memory access after a page miss is 10x slower than right before, and by having more memory banks to work with the RRedline can avoid these better than other chips.

A third advantage of the RRedline's bus configuration comes from what is know as the "triangle edge effect." When a 128 bit accelerator draws a triangle with 16 bit pixels, it has to write 8 pixels at a time. But if any of those pixels are off the edge of the triangle, up to 7 pixels of that drawing bandwidth are wasted. With a 64 bit channel, the RRedline can work with a granularity of only 4 pixels - losing much less bandwidth to the edge effect. The RRedline has an inherent electrical advantage as well. Because each 64 bit channel has separate address and control lines as well as separate data paths, the electrical load on the bus for a given memory configuration is half what it would be on a 128 bit bus. This makes it a lot easier to design boards that support extremely fast memory speeds. So not only will a RRedline chip use memory more efficiently than the competition, it will also be able to run it at a faster clock too.

The RRedline's dual channel architecture lets it get at two areas of memory at once, allowing it work on tasks in parallel during the time that a less advanced chip spends spinning in circles trying to figure out which way to go next. The RRedline also avoids many page misses that cripple other chips, and its finer granularity keeps it from wasting memory bandwidth on triangle edges. This explains why, despite their sums being equal, 2x64 is in reality much greater than 128. Throw in the ability to support faster memory speeds, and the RRedline's advantages only increase.

Forget 'wider is better'. The RRedline will show the world that 'better is better'!!


Bargain Matinee

Video is cool. Very cool. But despite all the talk of 'convergence', computers still are not very good at dealing with it. A movie almost always ends up looking better on a $99 discount VCR and cheap TV than it does on even the most top of the line DVD-equipped PC and snazzy monitor. Expensive DVD kits and decoder cards usually fall way short too - either using a passthru scheme that degrades your desktop image, or an overlay that relies on the often poor video scaling of the primary graphics chip. Sure, video on your desktop is cool, but its not worth paying extra for such poor implementations. But what if you could get exceptional quality video playback and DVD decoding, built in to your graphics accelerator, essentially for free?

The Rendition RRedline gives this power to you. It can decode MPEG-1 (video CD) and MPEG-2 (DVD) encoded videos in hardware, and display them with superb visual quality on screen. No need for expensive, clunky decoder boards.  No need for chewing up all your CPU cycles with Soft-DVD. The RRedline has everything you need built in, making 'convergence' fun again.

DVD video playback is a difficult task, requiring the following steps to occur after the data is decoded and the audio stream split away:

iDCT Transformations   Inverse Discrete Cosine Transforms are the computationally intensive portion of MPEG-2 decoding. This transform is used so as to exploit the spatial redundancy of pixels within blocks of an image, helping achieve MPEG-2's incredible compression. Only dedicated MPEG-2 chips and a very few graphics accelerators can handle this. But thanks to the RRedline's programmable RISC core, it can work these transformations with ease.

Motion Compensation   MPEG-2 uses pixels from past and/or future frames to reconstruct an image, thus exploiting the temporal redundancy found in moving images. This significantly enhances the compression achieved in the encoded video bitstream by allowing a moving object on a mostly still background to be compressed by encoding relative motion, rather than the entire frame. Reassembling video encoded with this technique is called motion compensation, and only a few of even the most recent graphics chips provide this acceleration. (The Rendition V2200 was among the first to do so)

Scaling, Filtering & Color Space Conversion  The final step is expanding the image to be full screen (or shrinking it into a window), filtering to make the scaled image smooth, and then converting the YUV colors of the MPEG-2 stream to display properly on an RGB monitor. Many graphics chips handle this portion of the DVD stack, but only a few can do it with style - keeping the image looking good. The RRedline keeps it looking great!

The RRedline's powerful RISC core along with its specialized video processing logic can handle all of these tasks in hardware, with horsepower to spare. This lets the RRedline handle over 80% of the work required to play back a DVD, leaving your CPU free for other tasks.  Chips that  can only handle motion compensation take only 30% of the load off the CPU. And accelerators without even basic motion compensation leave the CPU to do all the hard work, sweating and dropping frames.

You are going to want to have good video support on your PC. Not just to watch movies - video enhanced DVD encyclopedias, travel software, MPEG-2 filled games, and more are coming fast. If you don't have great video and DVD support built in, you're going to be left  wanting later.

Don't settle for spending all your CPU cycles or all your cash just to get your PC outperforming a cheap TV. With the RRedline you can save your dimes, and still get great video.  That's what a bargain matinee is all about. So grab some popcorn, and get ready for action!


The More The Merrier

You can never have enough RAM in your computer. What seemed like more than enough memory yesterday is barely enough today, and will surely feel confining tomorrow. Remember hearing "no one will ever need more than 640K" just a few short years ago? Those days are long gone, but the trend continues. Today 32 MB is already starting to feel real tight, and only the brave dare multitask with just 16 MB installed. Modern apps and operating systems are memory hungry beasts, and seemingly getting hungrier by the year. At this rate 128 MB will be standard in systems before we can blink! The situation is no different on your graphics card. We've quickly gone from 1 MB to 2 MB to 4 MB, and today if you want to be able to play the latest texture rich games you'd be foolish to consider buying a card with anything less than 8 MB on board. How long before even that much isn't enough?

The RRedline from Rendition is designed with the future in mind, and can support a full 32 MB of graphics memory on a single board, twice the maximum most other new 3D accelerators can address.

The RRedline supports the following memory configurations:

SGRAM: 2 MB, 4 MB, 8 MB, & 16 MB

SDRAM: 8 MB, 16 MB, & 32 MB

This wide range of supported configurations allows the RRedline to fill every nitch from the most cost conscious mainstream motherboard implementation on up to the ultimate hard-core gamer's rig. The workstation market is well addressed too - imagine how useful 32 MB would be on a board driving a modeling app in high-res simultaneously on two monitors? The workstation world will never be the same.

More memory on your graphics card lets you support higher resolutions, deeper buffer depths, more local textures, and more. Even 2D desktop applications get sped up, as the graphics drivers are able to treat the local memory as a fast graphics cache - letting windows zip around the screen even at the highest resolutions.

The more megabytes you have, the merrier your graphics  will be. The RRedline is ready to handle the demanding memory needs of tomorrow.  Make sure you are.

Go Deep

The more bits, the better. Rendition has known this from the beginning. Even our two year old V1000 worked with 32 bits of internal precision, and could work (albeit slowly) with true color 32 bit textures and frame buffers. This is just part of the reason behind Rendition's award winning (and continually top rated) visual quality. Nobody knows quality rendering like we do. The RRedline will continue the Rendition tradition of superb industry leading visual quality. In addition to further refined and improved algorithms and internal precision, dual texturing support, and the most versatile blend unit - the RRedline also has the speed to handle 32 bit rendering with ease, and supports a 24 bit deep Z buffer to eliminate annoying Z glitches even from tomorrow's most demanding games. The competition has just woken up to the importance of visual quality, billing true color rendering and increased internal precision as 'new' features in their latest chips. Yawn. Rendition has already been there, done that. While the other guys are just starting to figure out what quality rendering is all about, you can count on the RRedline from Rendition to set the standard for both speed and beauty. Hey - with a name like 'Rendition', the image better look good!

(image missing)

A Z buffer stores depth information about a frame as it is being rendered. The Z buffer is checked whenever a triangle is being drawn, to make sure that only the parts of the triangle that would not be obscured by previously drawn triangles are rendered to the screen. If there is not enough Z buffer precision,  pieces of a surface that should be obscured can pop annoyingly to the front, as seen here in this demo showing two overlapping rotating cubes.

(image missing)

Most of today's 3D games involve running around in mazes and tunnels, but tomorrow's games open up wide vistas that really begin to stress the limits of a traditional 16 bit Z buffer. Annoying jagged edge glitches like these will be common in accelerators that don't support 24 bit Z.

See Also