NVIDIA's current Kepler chipset architecture will be followed by Maxwell sometime in 2014; however the company decided to ignore it and talk about its next successor, Pascal.
Pascal will have its DRAM memory stacked into dense modules and placed inside the GPU itself. According to NVIDIA, this should decrease die size and allow faster memory access. In the end, Maxwell should carry twice the memory capacity of current chipsets at much higher bandwidth and energy efficiency.
Pascal GPU is also designed to give the CPU access to its memory and vice versa through an 80GB per second channel. This is five times faster than the current limit of 16GB per second.
You'll have to hold your horses however as Pascal-based graphics cards won't hit the market before 2016.
