The "Nvidia SLI" technology follows a similar announcement by Alienware, which designed a custom "X2" motherboard to accept a pair of graphics components, either from ATI Technologies or Nvidia. Nvidia, however, will work with its system-builder and add-on card partners to allow three of its latest PCI Express card designs to be sold in pairs, improving rendering performance.
At this point, however, the SLI (Scan-Line Interface) technology is just that -- a technology. Add-on card partners said they received their first briefings on Friday. More importantly, the technology will require motherboard vendors to design boards with two PCI Express slots, soon after their first single-slot PCI Express boards began shipping earlier this month.
All PCI Express based GeForce 6800 Ultra, GeForce 6800 GT, or Quadro FX 3400 boards support SLI technology, Nvidia said. The technology will not be available for AGP cards based on the chips, the company said.
Scan-Line Interleave technology was first developed in conjunction with 3Dfx Interactive Inc. by a company called Metabyte Technology in 1999, which was later acquired by Alienware. Nvidia, in turn, snapped up 3Dfx. Both Alienware and Nvidia's solution divvy up the rendering tasks between the two graphics cards according to the rendering load. Theoretically, the graphics performance of the system should double, although Alienware's solution will offer improvements of between 50 and 90 percent, according to the company.
"We are working very closely with Alienware and are extremely excited that they are pushing the idea of multi-GPU systems," an Nvidia spokesman said in a statement emailed to hardware review sites. "That being said, the current performance numbers that Alienware has announced do not reflect any of the dedicated scalability logic, inter-GPU digital interface or dynamic load balancing software of SLI."
The SLI launch will be designed around system builders, probably using Nvidia's new PCI Express-based Quadro cards. Nvidia announced threw new Quadro cards Monday morning: the Quadro FX 3400, with 256 Mbytes of G-DDR3 memory, 3-pin stereo display support, and dual-DVI connectors; the FX 3100, with 128-Mbytes of memory; the FX 330, which matches a 64-bit memory interface to 64-Mbytes of DDR-1 memory; and the Quadro NVS280, designed for rendering 2D applications.
Instead of being connected on the motherboard, each SLI card will be connected via a dedicated connector, which will connect to special slots on the card's PCB. Add-on card vendors described the card as a free-floating piece of logic connected by two connectors, one to each card.
So far, Nvidia has not informed its partners how the SLI connector will be sold, or its specifications. A concept demonstration of the technology on Nvidia's web site indicated that the graphics data would be fed out of a single vide connector, implying that the cards will have to talk amongst themselves via the connector. According to add-on card vendors, the second card is limited to an effective PCI Express rate of 8X, not the 16X bandwidth available to the primary card, meaning load balancing will have to be done. Load-balancing capability will be built into Nvidia's Detonator driver model, Nvidia said.
While the technology is an exciting way to sell an extra graphics card, one partner said the infrastructure will need to be developed. "They're way ahead of the curve right now, especially on motherboards," a source at one card maker said.
Meanwhile, Nvidia GeForce 6800 cards are still difficult to acquire, and are selling for upwards of $600 apiece.
Channel News and Analysis 
