Originally Posted by Waitng4realGPU
The graphics score is also only 500 points off a 1500mhz GTX 980.
So expect a 1400mhz+ 480 to match a reference 980.
Overclocking doesn't seem to scale as well on Pascal either judging by reviews, and judging by that video. So a 23.5% OC'd (which of course is not guaranteed as we see with the 1070/1080 only hitting 2000+ sometimes) 1060 could gain less than a 15% OC'd 480.
He reduced the tessellation work load which boasts scores by 10-15% if you watch the video.
Originally Posted by BiG StroOnZ
My only problem with these conservative clocks is that it doesn't line up with the clockspeed gains that are to be received from going to 14nm from 28nm. If it only clocks to 1400-1450MHz there is a serious problem with AMD's 14nm card here. As 14nm explicitly states 40-50% improvement in clockspeed compared to 28nm:
We know AMD is able to bin up to 1050MHz with their last gen cards. So you use that as a baseline number to figure out where a 14nm card would end up with 40-50% improvement in clockspeed.
So that would put us at 1470MHz to 1575MHz.
Now we know these 14nm and 16nm processes are capable of offering their clockspeed promises because here is what 16nm has to say about clockspeed improvements:
Also 40% improvement in clockspeed. And as we see with Pascal 1080, 1070, and now 1060 NVIDIA indefinitely has increased their clockspeeds by at least 40% compared to previous gen. As a matter a fact they were able to go over 40% in the end as we can see with 2200MHz clockspeeds provided by Pascal.
So basically if the 480 or any future AMD products cannot clock high, it is a problem with the 14nm process that they are using and perhaps it is because it is not as mature as 16nm is. In that 16nm was able to immediately provide the clockspeed advantage as promised and shown by NVIDIA with Pascal. Whereas we are still waiting around to see where 14nm will end up, because the 480 reference was so power limited.
It might not necessarily be a nodal one.
I posted this well over a month ago and it's another on of my predictions coming true.
LN2 is useless as far as showing max usable overclocks will be on a paticularly node, but they do gives some indication of how well an architecture will overclock on a more advance node. Look at the gtx 280ln(980mhz) vs overclocked gtx 480/580(980mhz h20) for example and now the gtx 980ln2 clocks vs what the 1080 does. The gtx 980/980 ti could get to 2300 on a very good ln2 bench run. 2200 are a more normal result. This could very well mirror what we get for the overclocks of the gtx 1080.
What I suspect is what 14nm finfet will show is AMD needs a big redesign to use get bigger clocks because with every new iteration of GCN where they add more functionality, the worse the clocks get. GCN 1.0 could get to 1800mhz on ln2. GCN 1.3 or fiji, even those it is made on a better process only gets to 1450mhz on ln2. GCN is just not an architecture designed to go as fast as maxwell. The cores are smaller and the pipelines are shorter but as a result, they can put more cores.
Basically GCN as an architecture just doesn't clock high. Even under ln2, fiji which is the variant most similar to the one in Polaris only overclocks to 1.45ghz under LN2. What does Maxwell overclock to with ln2? 2.2 ghz. Notice a pattern here? This is with the same node, but look at the clock advantage maxwell has over GCN 1.3.
GCN was was never going to overclock that high, even with the aid of tsmc. Fanboys are just looking for a scapegoat so they can put their hopes on the next GCN derivative.
The problem with GCN and why it will never catch Polaris or maxwell in performance per watt or mm2 is deep down it is just GCN.
What AMD has been doing is they have been ricing their GCN architecture out. Adding mods, turbo's, air intakes and exhausts, but what they really need is a new car to win the race. GCN is an architecture that has aged well, but we won't get big changes until their truly big next gen architecture.Edited by tajoh111 - 7/5/16 at 12:38am