Last night, during its keynote at Nvidia Geforce LAN event in Shanghai, China, Nvidia finally announced its dual-GPU GK104 Kepler based GTX 690 graphics card.

Squeezed on a single 10 layer 2oz copper PCB, the two fully enabled GK104 GPUs will share room with a total of 4GB of memory and PLX bridge chip. Since we are talking about fully enabled GK104 GPUs found on the GTX 680 card, this one feature a total of 3072 CUDA cores, 256 TMUs and 64 ROPs (1536, 128, 32 per GPU). There is a total of 4GB of GDDR5 memory or 2GB per GPU paried up with a 256-bit memory interface each.

Of course, squeezing two GPUs on a single PCB plate has its drawbacks, so Nvidia had to cut down the clocks. Unlike the GTX 680 which works at 1006MHz base and 1058MHz Boost clock, the each of the two GK104 GPUs on the GTX 690 ended up clocked at 915MHz base and 1019MHz boost clock. Luckily or impressively, the memory clock remained at 6Gbps which provides a stunning 375.5GB/s of bandwidth.

The TDP is still set at 300W and it needs two 8-pin PCI-Express power connectors. Due to 8+8-pin design, the card can pull up to 375W (150W per 8-pin PCI-E power connector + 75W from PCI-E slot) which leaves a whole lot room for GPU Boost. The new GTX 690 graphics card also supports quad-SLI, which means that you can pair two of these for some extra fun. 

Since we are looking at a highest-end product, Nvidia decided to go an extra mile and equip the new GTX 690 with nickel-plated fin stack heatsinks with dual vapor chamber that is cooled by center mounted axial fan. The entire cooler is covered by a cast aluminum and injection molded magnesium alloy shroud that should provide additional heat conduction as well as some noise insulation.

The new GTX 690 dual-GPU card is scheduled to launch on May 3rd with a rather hefty US $999 price tag.