How it works (cont’d)
Each GPU on NVIDIA’s Quad SLI card is outfitted with a single-slot heatsink, with the cooler used somewhat similar to the standard cooling NVIDIA employs for the GeForce 7800 GTX 256MB. The heatsink itself though is longer, with the card’s fan located offset of the GPU in order to increase its effectiveness.
The PCB used for each GPU is longer as well. This is necessary in order to house the larger cooler as well as provide room for the additional circuitry NVIDIA adds to Quad SLI cards. This added circuitry is comprised of an additional chip that is used to provide both GPUs on the Quad SLI card with 16 PCI Express lanes despite the fact that only one physical x16 PCI Express graphics slot is used. How does it do this you ask?
It turns out that the chip acts as a PCI Express “splitter” of sorts. From one x16 PCI Express graphics slot, the chip splits it into two x16 connections, enough for one full x16 connection per GPU on the card. The chip is also responsible for merging the data back from both GPUs into one x16 connection.
We’ve been told that NVIDIA’s Quad SLI solution is only compatible with their nForce4 SLI X16 chipsets. Older nForce4 SLI chipsets aren’t compatible with Quad SLI.
In addition, just to clarify some of the confusion, while the Quad SLI platform at CES was demonstrated on an nForce4 SLI X16 Intel Edition motherboard, NVIDIA’s Quad SLI card solution won’t be unique to Intel – both X16 nForce4 SLI chipsets are supported. In fact, considering AMD’s high adoption rate among gamers, NVIDIA expects Quad SLI to be just as popular on the AMD platform as Intel.
It’s important to note however that pseudo-Quad SLI can’t be rigged to run on motherboards with four PCI Express graphics slots by using four different graphics cards.
If you recall, late last year Gigabyte announced an nForce4 SLI X16 Intel Edition motherboard with four PCI Express graphics slots, the GA-8N-SLI Quad Royal. While the board uses NVIDIA’s nForce4 SLI X16 Intel Edition chipset, only two of the board’s four PCI Express graphics slots provide full 16-lane PCI Express capability. Gigabyte hasn’t been specific on how many lanes the secondary graphics slots are capable of supporting, but since the South Bridge that the PCI-E slots are tied to is capped to 20 lanes total, our guess is that you’re limited to just one lane per slot, with the remaining two lanes going to the secondary x1 PCI Express slots on the motherboard (16 lanes for one graphics slot + 2 lanes for the third and fourth graphics slot + 2 lanes for the x1 expansion slots for a total of twenty).
In other words, while two NVIDIA Quad SLI cards should work on the Gigabyte GA-8N-SLI Quad Royal (provided the two Quad SLI cards are running in the board’s two x16 PCI-E slots), running four GeForce 7800 GTX 512MB boards on the Quad Royal for pseudo-Quad-SLI won’t work.
Likewise, you also can’t combine two GeForce cards with dual GPUs on one card together for Quad SLI. As you probably know by now, both ASUS, Gigabyte, and MSI have all developed GeForce 6800 GT and GeForce 7800 GT cards that combine dual GPUs onto one PCB for single-card SLI. NVIDA’s Quad SLI announcement won’t allow you to combine two of these boards together for pseudo-Quad SLI.