Fiji Architecture and High Bandwidth Memory
Fiji is the newest architecture released by AMD in around 2 years. Based on the most recent GCN 1.2 and 28nm process, it aims to bring ‘revolutionary’ performance to the end-user. A close look at the package, we can see the GPU core surrounded by 4 HBM stacks. These stacks are extremely compact, measuring just 5mm x 7mm footprint. The advantages of moving the memory from the PCB to the Interposer is the decreased distance for communications, which increases bandwidth and decreases power and overall reduces the footprint of the PCB; enabling a high-end graphics option to be used in ITX size computer cases. With the integration of the memory chips, it means extra components such as VRMs no longer need to be included on the PCB, again reducing the overall size of the PCB.
This is great for the graphics card market, with HBM v2 introduced more widely in the next generation of cards, but we could also see designs like this enter the CPU market, mobile phones or any other device that currently has separate core and RAM modules.
Can the HBM on this card not be overclocked like it does on Fury X?
There is no need to its blistering ly fast
There is no effective HBM memory overclocking at the moment, mostly only GPU..