While a lot of attention has been paid to both generations of High Bandwidth Memory since AMD launched its R9 Fury GPUs with HBM1, and then RX Vega with HBM2, the way HBM is used can make it more expensive to work with compared to the aging GDDR5 of most video cards today. While HBM will still offer some important advantages, GDDR6 is coming, especially now that Samsung has begun mass production of the next generation memory.
The chips Samsung are working now are 16 Gb GDDR6 chips, made using Samsung's 10 nm class process. The company's GDDR5 chips are based on a 20 nm process and are only 8 Gb, so this offers double the density while also increasing performance. Up to 18 Gbps pin speed and transfers of 72 GB/s is possible with these GDDR6 chips, compared to the 8 Gb GDDR5 chips' 8 Gbps pin speed. With its low-power circuit design, these chips will achieves this speed at just 1.35 V, which is appreciably lower than the 1.55 V used by most implementations of GDDR5.
Last I saw, there is no word on when AMD or NVIDIA will adopt GDDR6.
Back to original news post