| Trumpstyle said: I suspect Turing will be more of a refresh then a generational leap. Unless they made some big architectural improvements I don't see 12nm being a big enough leap to give that 50-70% performance gain, I suspect we will get about 30%. Another issue is memory size, are they just gonna double it when memory is so expensive right now (example, 2080/2070 getting 16gb of ram)? I don't think so. Another interesting aspect is 7nm. Will they launch Turing this year then another architecture next year on 7nm? |
12 nm is basically a slightly improved 16 nm, which on turn is already a 20 nm process in reality, so just about all the performance gains will be due to architecture. That is, assuming it is derived from Volta instead of just a Pascal refresh. As for 7 nm, I guess that depends of when the smartphone SoCs will start using it, since they make it to larger chips about a year afterwards. I used to expect it for the next year, but it's not looking like there will be smartphones later this year or early on the next using it, so far, so maybe that's a 2020 launch to keep things on a nice even 2 year cadence, as it has been for a while now.
A wild card here is GlobalFoundries' 7 nm process which is very close to Samsung's 10 nm, and a 50% improvement over their 10 nm, which could be available earlier than the more commonly expected 7 nm from TSMC and Samsung. It would mean a nice boost to AMD if they could release GPUs on it maybe even a year before Nvidia gets their next process node, though sadly, AMD are still be so far behind Nvidia in architecture that it would end up making things even at best.
That's how I understand it at least.
