During the keynote speech at Nvidia's annual GPU Technology Conference in San Jose, California, CEO Jen-Hsun Huang updated the company's public GPU roadmap with the announcement of Pascal, the next-generation GPU family that will follow Maxwell GPUs, which are incoming this year.
Pascal, which is named after the 17th century French mathematician Blaise Pascal, features three key new features:
- Stacked DRAM, or 3D memory: This is where DRAM chips are stacked into dense modules with wide interfaces, which allows the DRAM to be inside the same package as the GPU. This allows the GPUs fast access to the data held in RAM, boosting throughput and efficiency. It also allows for more compact GPUs which can be fitted into smaller devices. This offers a several-fold increase in bandwidth, more than twice the memory capacity and quadrupled energy efficiency
- Unified memory: This allows applications that take advantage of both GPUs and CPUs and allows the CPU to access the GPU’s memory, and the GPU to access the CPU’s memory. This means that developers don’t have to allocate resources between the two chips.
- NVLink: This increases the bandwidth between the CPU and GPU from the current 16GB/s to more than 80GB/s.
Because Pascal GPUs come on a board one-third the size used today, the chips can be used to bring more power to smaller devices.
GPUs making use of the Pascal architecture are due to land in 2016.