Both AMD and NVIDIA will be making use of HBM2 in their next-generation graphics cards that will be launching later this year. The JEDEC standards for HM2 were recently released and with that Samsung has just started mass production of their HBM2 memory chips.
Samsung’s HMB2 will make use of the 20nm manufacturing process. Each HBM2 stack will be 4GB. Each stack is made up of four 8Gb core dies on top of a buffer die at the bottom. Each one of these packages feature 256 GB/s of bandwidth, that is 4x the speed of HBM1. A typical HBM2 setup on a graphics card would make use of four of these stacks giving the user a total of 16GB and 1TB/s of bandwidth! HBM2 will bring with it a space savings of up to 95% compared to GDDR5.
In addition to the typical HBM2 Samsung is also working on HBM2 with ECC. This will likely be used for GPGPU and Enterprise applications. This means we could see enterprise graphics cards outfitted with HBM2 in the future. Samsung is expected to continue to ramp up production of HBM2 as the demand for it increases.
