It looks like NVIDIA is going to be launching their flagship graphics card based on the “Pascal” architecture, the GeForce GTX Titan P at Gamescom 2016, which is held between August 17-21 in Cologne, Germany. The card is expected to be based off the GP100 silicon and is likely to come in two variants, 16GB and 12GB.
Not only will the variants differ by memory size, but also memory bus. The 16GB version could feature four HBM2 stacks over a 4096-bit memory bus, while the 12GB variant could have three HBM2 stacks over a 3072-bit bus. This is the exact same approach NVIDIA took with the Tesla P100-based PCI-Express accelerators. The TDP of the card could be rated between 300-375W, drawing power from two 8-pin PCI-Express power connectors.
Now the GP100 and GTX Titan P are not the only high-end graphics solutions NVIDIA is working on for enthusiasts, they also have the GP102 silicon, which is positioned between the GP104 and GP100. This chip will likely lack the FP64 CUDA cores that are on the GP100 silicon and feature up to 3840 of the same CUDA cores found on the GP104 silicon. The GP102 is also rumored to feature GDDR5X memory over a 384-bit bus. NVIDIA could base the GeForce GTX 1080 Ti off this chip.