Wow. Thx for putting together the numbers majord.
Yes, Integer is important, but also the Intel CPU inside the graph are all 22nm. And Zen is on 14nm.
Basically, this is not looking good.
Well, this is true (regarding all those HEDt parts being 22nm) , but with the exception of the 10 core flagship, the octocore and hex core broadwell (14nm) parts which Zen will compete against aren't much faster anyway are they:
5960x: 8 core @ 3.0/3.5t
6900K 8 core @ 3.2/3.7Ghz + a few % IPC (equiv to a 3.3/3.8Ghz)
So in other words a 6900K will be around 10% faster than the outgoing 22nm octocore's overall.
Now lets look at it another way completely:
Because we're talking about ultimately very similar architectures now (at a high level - i.e wide cores sharing similar resources for SMT) it's not
too hard to see what sort of clockspeeds zen would need to match, say a 6900K's performance. based on IPC claims and core count alone.
Referencing my own results (i'm glad to incorporate other tests if one can find some data ) which conclude
Skylake has a 60% higher IPC than excavator, Single threaded.
We can then use AMD's claim of
40% higher ipc over excavator to calculate that, if true Skylake would be left with a
14% IPC advantage over Zen
Now, we're talking about
broadwell-e here though, and unfortunately i don't have any of my own or anyone elses data to use, so we have to use the commonly accepted figure of
Skylake being 3-5% higher IPC than Broadwell. Feel free to correct that if you think it's wrong.
Which in turn gives a convienient round
10% figure for Broadwell IPC > Zen.
that's perfect for working out single/ <8thread performance , but to work out total throughput (that is - 8c/16threads utilized) it's all down to assumptions that Zen will fork out similar gains from SMT - and it
should be similar, but technically this 10% figure can only be used on a per core basis , HT 'scaling' may differ - likely in broadwell's favour a little to be honest.
but lets run with it:
So for a zen part to match a 6900K by using clockspeed to cancel out its IPC deficit:
6900K: 3.2Ghz base(all theads) / 3.7Ghz Turbo (single thead+)
+10% =
Zen: 3.5Ghz base / 4.0Ghz turbo.
Now to back a further generation to the 5960x you'd be looking at Haswell vs Zen IPC instead, which is getting tricky.. For arguments sake, ccould say it has a
7%? lead over zen's claimed IPC.
5960x: 3.0Ghz base / 3.5Ghz Turbo +~7%
Zen: 3.2Ghz Base / 3.7Ghz Turbo
Are these sort of frequencies possible? That's where everyones stuck on at the moment..
Looking back at the graph i posted though, actually makes the claim of double an FX8150 pretty plausible - you'd basically need less than 3Ghz to achieve the values for Zen in that graph based on the above