Sharky Extreme's exclusive preview of the Metabyte TNT SLI has the scoop on plans to offer SLI-mode TNT accelerators (two cards used in tandem) to offer increased performance, and from the report it sounds like this is a breakthrough that will offer SLI mode for any card they choose to implement this with. Here's an excerpt to give a feel for what this is all about:

Metabyte's engineers have been working on an undercover technology that allows not only nVidia chipsets to be SLI'ed in tandem, but virtually any future chipset from any vendor that hits the market.

The internal code name for the Metabyte project is "Wicked3D Stepsister" (don't ask us, no one we talked to remembers where this name originated from) and it heralds a new dawn as far as linked high-powered video options go. Our Metabyte sources answered "Yes, yes, and yes" when asked if it was possible for the Wicked3D Stepsister program to be adapted to work with the S3 Savage 4, 3Dfx Voodoo3, TNT2, or another brand new chip architecture. Although it's unlikely in Metabyte's mind that they'll do a Savage4 or Voodoo3-based SLI product, the TNT2 is a whole different enchilada. Later in this article we'll break the news on two possible TNT2 SLI solutions, but for now let's look at the Step Sister technology itself.

Here's the way the system will work:

Metabyte is going to sell both PCI and AGP versions of their TNT SLI capable cards separately for between $125 and $150 for each 16MB card. Much like the Voodoo2 SLI setups that we all know and love, the Metabyte TNT SLI system works similarly but differs with two possible SLI combos: 1 AGP TNT + 1 PCI TNT 1 PCI TNT + 1 PCI TNT

The AGP + PCI version of the Metabyte SLI rig will offer slightly faster performance (about 5 to 10%) than the 2 TNT PCI cards when linked together. This is due to the way that Metabyte has chosen to implement SLI, to maximize their driver's capabilities. Instead of rendering each 3D image's odd or even lines in succession the way 3Dfx's SLI system works, Metabyte's process separates the entire 3D image into two halves: The top half of the on-screen image, and the bottom half of the on-screen image. This supposedly reduces the CPU overhead versus the odd/even 3Dfx approach while making the whole operation smoother and more seamless.