Re: Intel 10nm "Ice Lake" to Combine "Sunny Cove" CPU Cores with Gen11 iGPU
2018/12/13 20:42:27
(permalink)
Since it takes four years to make a 'brand new' architecture from the ground up, I suspected that IF they took AMD seriously when Zen debuted in May 2015, that by summer 2019 would be the first opportunity for Intel to bring something new. This looks like it, but I don't think this is all that's coming; they didn't anticipate Threadripper, so this new "Cove" architecture is just an attempt at parity, increase the IPC gap with AMD pre-Zen: Excavator was some 60% behind Skylake, Zen brought in 52%, Zen+ added 2% and Zen 2 could easily come right up to Skylake-- which is a threat to Intel as I'm sure they figured it was only a matter of time. Cove must allow for an initial boost of at least 40% IPC from the Lakes, with another 5-10% per Willow and Golden.
All Intel needs is parity; if their processors are just enough better, they lose high volume sales to cheaper competitors. If their processors are so much faster than competition, they keep their high volume sales, they keep their market share.
With Zen 2, Intel runs the risk of having AMD on par, they can't allow that.
But Jim Keller was only hired in April 2018, around then we first started hearing about the "Ocean Cove" architecture. So this "Sunny Cove" is a 2nd iteration posing as brand new. I'm guessing Ocean Cove was meant for the same 10nm architecture than Cannonlake was supposed ot be on, they were both canned and the 10nm refresh that was originally a Skylake based Icelake is now a Sunny Cover based Ice Lake.
It was clear by branding of the Core i9 9th gen that Intel was running out of sequence nomenclature, so something new ought to have been coming. But to reprupose the Icelake name with a different architecture is a smart move as I'm sure AMD was assuming (as I did) that Icelake was just another Lake, shrunk to 10nm would have allowed double the cores at same TDP but nothing more.