The true power of the QuadroPlex
Many of the announcements from NVidia at SIGGRAPH centered around the QuadroPlex. NVidia is really shifting alot of focus towards CUDA & GPGPU, and QuadroPlex and Tesla are the core of that push. Almost every display in their booth was powered by at least one QuadroPlex, with several of them powered by Two. Seen above is an 8-screen display driven via a pair of QuadroPlex units connected to a single PC. (shown right). Some interesting things to note about this:
- Each QuadroPlex can drive 4 monitors, and they appear to the host operating system (Windows in this case) as a single display. This means you don’t need special applications to spread across the monitors, it just appears as a single giant monitor.
- A pair of QuadroPlex’s can cooperate, as shown here, to offer the same functionality across more than 4 monitors. However, there is a limitation of 8k pixels on an axis, meaning a maximum desktop size of 8192×8192 for 3D applications.
- On a hardware level, the QuadroPlex supports overlap, blending, genlock, and frame-lock. While not really important for LCD’s, when used with projectors that means you get (for-free) blending and synchronization for multiple projectors. This is one of the reasons QuadroPlex and the Sony4K Projector go together so well.
One other interesting thing to note is in the adapter used to connect a QuadroPlex to a PC. The full QuadroPlex solution is an external one, requiring a small PCI-Express card installed in the host-computer that allows the umbilical to the outside box. Previously, if you wanted Two or Three QuadroPlex’s hooked to your machine, you would need a separate adapter card (and hence, a separate PCI-Express slot) to dedicate to each one.
Now, thanks to the advances of PCI-Express 2.0, they have combined two adapters into a single card. This is great not only for desktops, but particularly for rack-mount systems which typically only have a single PCI-Express slot (via riser-card). Now you can run two QuadroPlex’s on a single PCI-Express 2.0 x16 slot, with no performance penalties.
