Related Articles

Ah, now we get to the crown jewel - the 780a. NVIDIA of course goes on... and on... about their new AMD flagship.

This diagram is fascinating.

Compare it with the 730a and 750a diagrams - carefully.

The only differences are:

• ESA certified
• addition of nForce 200

The ESA certification really needs no new silicon, so its not a hardware level difference between the 780a and the 730a/750a.

The nForce 200 is not new - its been around for a while on Socket 775 boards to add additional PCIe capabilities; and if you carefully noted the differences in the diagrams, you can easily see that the nForce 200 must use the 16 PCIe 2.0 lanes of the 780a to communicate with the main chip, so even though the nForce 200 "breaks it out" / "switches lanes", there are really only 16 native 2.0 PCIe lanes.

Don't worry though, I'd bet that NVIDIA "overclocked" the "native" lanes, so I imagine there may be some additional bandwidth available when using an nForce 200.

Funny thing is... it does not really matter. When using SLI, even triple SLI (or quad SLI), most of the inter-GPU traffic goes over the SLI bridges, and not over the PCIe lanes.

The "PW Shortcut" mode can be genuinely useful, and it shows that the nForce 200 is more than a silicon replacement for the old SLI "paddles" on motherboards (that switched between single 16x or dual 8x modes). By letting the GPU's talk to each other directly, there is a potential of getting useful transfers happening between the GPU's without involving the CPU.

Broadcast mode is also potentially very useful - imagine using SLI and sending the geometry and texture information to all GPU's at once instead of separately - this can virtually increase the bandwidth significantly.

NVIDIA has been doing its homework.