NVIDIA enables the linking of two cards via NVLink2, this gives you 10240 CUDA cores and 64GB HBM2, for a total of 236 teraflops of Tensor Cores. No Volta (or Ampere/Turing) consumer cards were teased at the conference, so we'll have to wait a bit longer. Full specifications of a single Quadro GV100 card can be viewed below.
On a related note, NVIDIA also replaced its Tesla V100 series with 32GB HBM2 editions. The previous model, which featured 16GB HBM2, is no longer in production. Better availability of HBM2 and the adoption of 8GB HBM2 stacks enabled NVIDIA to launch this upgrade. No details about pricing, but it seems there's no price hike involved with this drop-in replacement.