Why SLI chipset?

[PeteRoy]

I want someone to clear this up for me.

In the Voodoo 2 SLI days you didn't need a special SLI chipset on the motherboard, all you needed was Pentium 2 440BX chipset or a Pentium 1 chipset and 2 PCI slots.

Why today Intel needs Nvidia to make SLI chipsets for Intel motherboards? Why Intel can't just put 2 PCI Express slots?

Thanks.

 

[sparkyclarky]

Simple answer - the V2 did very primitive SLI, where each card rendered either the even or odd lines of a frame. This didn't require any load balancing or anything of that sort and IIRC the cards were synched via an external passthrough cable (or maybe that was just used to connect the 3D to the 2D cards).

I'm not incredibly knowledgeable on this, but overall I'd have to say that it is an issue of complexity. The V2 is stone age tech compared to the latest gen NV and ATI parts. The faster/more complex a chip gets, the faster and more complex the interface to the computer will get. The SLI chipset is just one link in this chain.

Note that the above is just general speculation though.

 

[Matthias99]

Why today Intel needs Nvidia to make SLI chipsets for Intel motherboards? Why Intel can't just put 2 PCI Express slots?

AFAIK, there's nothing magical about NVIDIA's SLI motherboards other than that they have dual PCIe x16 slots (albeit running at x8/x8, not that this makes any real difference in performance). Hence how you can mod the DFI Lanparty boards to enable SLI.

Didn't NVIDIA demo SLI originally on an Intel Tumwater chipset (a server board with dual PCIe x16 slots), since NForce4 wasn't ready yet?

 

[Venomous]

It comes down to bandwidth. No way can AGP handle the fillrate of today's GPU's. It should be interesting to see if ATI ends up using Tile Based Rendering for their SLI solution however. I actually had a dual V2 and also the single dual canopus card 🙂

 

[Pariah]

It comes down to available PCI-E lanes. There is nothing SLI specific on NVidia's boards, they simply allocate the available PCI-E lans differently. Intel's 915 and 925X chipsets support a maximum of 20 PCI-E lanes. Assign 16 to one video card and you are left with only 4 lanes which need to be split into 4 1x PCI-E slots for other peripherals or there won't be any other expansion slots on the board. NVidia gets around this problem by placing two 16x expansion slots on the board which are really only 8x electrically, leaving 4 available lanes for other slots. Since PCI-E video cards will not fit physically fit into slower slot, the SLI slots have to be physically identical to 16x slots despite not being real 16x slots.

So basically, the difference between Intel PCI-E chipsets and NVidia's SLI chipsets are that Intel has one true 16x PCI-E slot with 4 lanes left over, while NVidia has two 8x slots which are physically identical to 16x slots with 4 lanes left over.

AGP 8x is in no way limiting the fillrate of today's GPU's. Not in the slightest.