The 1600MHz came about from people calculating the required clocks to hit 12.5 TFLOPs based upon what was known at the time and the 4096 cores, compared to Fiji.
However this was applied to MI25, and the MI25 is not watercooled but passive, meaning cooling from the server-node-blade chassis structure.
It is a lot to ask especially when Nvidia is more conservative with their Boost clocks and the limits they are at with HPC Pascal on the larger dies (even GV102 that is smaller than Vega).
If AMD manage 1600MHz and raw performance actually scales to the theoretical peak limit and importantly sustained, then that is one great engineering feat from AMD IMO.
Personally I am expecting performance around 1080 level.
CHeers
However this was applied to MI25, and the MI25 is not watercooled but passive, meaning cooling from the server-node-blade chassis structure.
It is a lot to ask especially when Nvidia is more conservative with their Boost clocks and the limits they are at with HPC Pascal on the larger dies (even GV102 that is smaller than Vega).
If AMD manage 1600MHz and raw performance actually scales to the theoretical peak limit and importantly sustained, then that is one great engineering feat from AMD IMO.
Personally I am expecting performance around 1080 level.
CHeers