While we are just a few months away from the launch of AMD's Fiji GPU, which is the new Rx 300 series GPU, we now have some of the first details regarding the successor, AMD Greenland GPU.

While the Fiji GPU is made on a 28nm manufacturing process at TSMC, same as Nvidia's flagship GM200 Maxwell GPU, it appears that both Nvidia and AMD will definitely skip the 20nm manufacturing process, due to bad yields, and that the Fiji successor will be based on 14nm manufacturing process.

According to a report from Fudzilla.com, the future AMD Greenland GPU, which should be based on Arctic Islands architecture and be based on a 14nm manufacturing process, will not be a "radically new core" and will be mostly focused on performance per watt. This worked well for Nvidia and its Maxwell architecture and AMD wants to do the same next year.

One thing that will make the Greenland GPU better than Fiji, in addition to advanced 14nm manufacturing process, is the presence of HMB2 memory, something that SK Hynix already talked about earlier and which brings more memory per DRAM die and doubles the bandwidth per stack.




While the Greenland GPU will be built at Globalfoundries, Nvidia's Pascal chip could be based on either TSMC's 16nm FinFET+ manufacturing process or go for Samsung's/Globalfoundries 14nm process. Earlier reports suggest that Nvidia tapped Samsung for some of its chips but it is not clear if Samsung will be making low power Tegra chips or GPUs as well.

In any case, HBM2 is definitely going to replace GDDR5 as standard memory in both AMD and Nvidia high-end graphics cards next year and 2016 will certainly be an interesting year for GPUs.



Source: Fudzilla.com.