We first heard about AMD’s Greenland GPU when we details AMD’s upcoming 16-core APU. Greenland will be the successor to Fiji (which will be powering the upcoming R9 390X). It has been reported that this new core will not be a radically new core. AMD wants the main goal of Greenland to bring more performance per watt to AMD GPUs in 2016.
AMD’s upcoming Fiji has a 28nm design, which is the same as NVIDIA’s GM200-based GPUs. It seems with the really bad yields of 20nm both NVIDIA and AMD decided to not to go with this node. Based on early information it looks like Greenland with be based off the 14nm node and be produced by GlobalFoundries / Samsung. AMD has been using TSMC for the past two decades, but it looks like they have not been happy with their yields.
Greenland’s main goal is to focus on low power. AMD hopes this will get them back in the good graces of notebook manufactures and bring down the overall power consumption of desktop GPUs as well. This worked quite well for NVIDIA when they brought out the Maxwell architecture.
Greenland will be using the second generation of High Bandwidth Memory (HBM2), which doubles the bandwidth and doubles the maximum memory size. NVIDIA’s upcoming Pascal will also use the same HBM2 memory as well.
We are still not sure who AMD or NVIDIA will go with for their 16nm or 14nm manufacturing process, but one thing is clear, after many years of 28nm GPUs the industry will finally move to smaller manufacturing node. This means that you can fit four times the transistors on 14nm on the same die size.