In the Forums...
Posted: April 11, 2000
Written by: Tuan "Solace" Nguyen
Double Knock Out
Here's the actual Rage Fury MAXX reference card:
ATI takes one AGP4X PCB and slaps on two new Rage 128 Pro chips to double the fill rate and throughput performance. This is indeed raw muscle.
With two full speed Rage 128 Pro graphics processors on one board, each processor requires 32MB of independent local memory to provide 3D resolutions up to 1600 x 1200 @ 32-but color in entertainment apps. This is where the 64MB memory spec comes from. This technique is similar to what 3dfx did with their Voodoo2s, SLIing 2 x 12MB for a total of 24MB. Voodoo2 SLI has 16MB of effective memory. The frame buffer memory is additive, it's only the texture memory which is duplicated. Since a single 12MB Voodoo 2 card has 4MB of frame buffer and 8MB of texture memory the pair have 8MB of frame buffer and (effectively) 8MB of texture memory. This however, is not the case with the Aurora. The effective total memory of the card is 32MB. This meant that the card has to dublicate textures and frame buffer in both of its Rage 128 Pro's memory and is the biggest drawback in the Multiple ASIC technology.
Simply slapping two Rage 128 Pro engines together doesn't denote an automatic double performance gain. ATI still had the difficult design task of getting the two engines to work as a pair in the most efficient and effective method possible. Here's the how they did it.
Scan Line Interleave
The first popular method, called Scan Line Interleaving (SLI), was the most widely used method.
3dfx did this very successfully with their Voodoo2 architecture. Plugging two Voodoo2's together would give double the fill rate performance. In SLI, this is achieved by the two cards rendering independently of each other. One card would render the odd lines, and the second card would render the even lines.
Though bringing excellent performance boosts to the table, there was a drawback in using SLI: you needed two exact cards (with the same amount of memory and vendor) and two slots. With the Voodoo2, you had to have either a PCI or AGP slot for your primary card, and then two more PCI slots just to do SLI. That's a total of three slots. In our opinion, this was pretty wasteful slot usage. 3dfx's upcoming Voodoo5 6000 however does 4-way SLI on one single board. It seems that SLI will make a comeback soon.
Then there was Metabyte, and its solution was a bit different but of the same concept.
Parallel Graphics Configuration
This second method introduced by Metabyte consisted of two cards. The solution was to pair up an AGP card with a PCI card. These cards were both matched together by the manufacturer to simultaneously render two independent halves of the screen. This method used Metabyte's own home-grown drivers, which allowed cards on different buses to dish out two different images that were put together by the main card into a single image.
PGC came with a load of problems. First, many vendors did not want to risk developing costs to bring this technique to mass audience. Secondly the load balancing between the two cards rendering the two halves of the image could shift as far from an equal 50/50 to a freaky 70/30 under certain circumstances, with one card doing a disproportionate amount of the rendering workload compared to the other card.
Another caveat in PGC came with the problem of configuring both cards to display the same image quality. Adjusting contrast, brightness and gamma among other things became too much of a hassle and problem for anyone to take on the project.
The PGC technique eventually disappeared. This is where ATI jumps in with Multiple ASIC Technology.
Editor's Note: Information on the Voodoo2 was gratefully given by: Nicolas Bryant @ MSWTech.com. Thanks Nic.