Originally Posted by gamervivek
And a significantly bigger chip than Polaris 10 which makes those high clocks even more doubtful.
It looks more likely that a 64CUs chip would be repeat of Fiji on 14nm with a die size compared to Polaris 10 like Tahiti to Pitcairn rather than Hawaii to Pitcairn.
ROPs aren't going to be the problem if it has enough memory bandwidth, it's more likely that Fiji was bottlenecked by the front end and if AMD are going for a <400mm2 die then they wouldn't be doubling it.
Yeah, making the chip bigger and likely increasing the transistor density is not going to effect clocks positively. Rather negatively is the likely case.
GP102 takes a small hit vs the gp104. And maxwell and pascal are exception for clock scaling, when scaled up.
Until they change the cores themselves, we are not going to get big frequency jumps, particularly when you note, GCN frequency has been trending downward since tahiti as far as max overclock potential. Polaris helped a bit from the note switch, but if polaris was made on 28nm, you would likely see 1100-1150mhz being the max overclock like fiji or tonga. 1500mhz server parts means 1600mhz gaming parts. This has been ln2 clocks on polaris. The max someone reached so far is 1700mhz on ln2 on polaris and there was no cold bug. So all of a sudden we expect 1600mhz gaming which is close to the ln2 clocks of polaris? It's silly.
Until AMD goes to navi, i don't expect any big frequency potential changes. Nvidia got there big frequency jump which they really changed the cores vs kepler. Pascals clock are basically unleashing the potential which maxwell already had. Compare what Fiji did on ln2 and what maxwell did on ln2 and the current clocks of both chips should not surprise anyone.