There are workload with more operations per each data. Example: there are sub bench of SPEC FP 2006 that have up to 2.4 of IPC (one thread). So 2 threads need 4.8 instructions per clock, mean. Even not counting conflicts, INTEL pipelines are not enough for the peak, just enough for the mean. Zen pipelines probabily yes.
10 uops because we have 4 alu, 2 agu and 4 fp. I was talking of the execute stage. I know that the dispatch can be up to 6 uops of which max 4 FP. Also 10 can't be sustained because retire rate is 8 uops/cycle. But 6 uops cycle can be sustained. For INTEL it depends, because due to int/fp pipe conflicts, there could be slowdown that on Zen can't happen.
Yes this is exactly the point, Peak numbers aren't meaningful, sustained numbers are, while there will always be corner cases where these different architectures will have advantages (BD had some great int performance if you used the int ALU + the int SIMD at the same time) in a general normalized sense they wont be and on average the usable resources for both a Zen and a Skylake core are comparable.
Like you agreed, it can go either ways
So? aMD have demoed a 8c16t zen having less lag while streaming than an i7 6900k and attributed it to better multithreading gains due its infinity fabric. I.e. gains from multicore and SMT.
Now your just jamming words together, the infinty fabric is the collection of IP that connects different IP blocks together coherently, it plays no part within a CCX. it also wont have lower latency across the infinity fabric between two CCX's then broadwell-E ringbus. Now better streaming performance could be because of better predictive selection of thread priority but that also doesn't mean its throughput and thus performance is higher.
Now you are cherry picking results. I said on average. Also amd said average.
I didn't do any such thing, i broke them down into two broad categories and then referenced the actual Zen design.
They are better than BD, but we don't know how they stack up against Intel.
We do have limited data which on a early ES form CPC on game benchmarks which have a far higher weighting to core clock then thread count. if we linear scale ( i know thats unlikely) the 3.15 Ghz Zen to 3.6 then its beating the 6900k quite comfortably.
Those are just theory note written down by the architect. On paper BD was also a monster which led to it hype before. I think you should not try to fly so high on arbitrary details with no benchmark to back you.
Actually is wasn't a monster, It was narrow, had high latency caches and bad miss penalties. It never clocked high enough to offset those issues. As we learnt more about the Core/SOC we found out about its glass jaws. If you are aware of these glass jaws you will find amd have specifically mentioned the improvements in those area's.
Trying to Compare the two architectures and say its the same situation is just being disingenuous.
Lol. Yes they did. Compitative. To a server socket with 32 core. Yes indeed. but:
1. servers usually have better SMT.
2. At 32 core its clocked lower, so amd may be referring to its better TDP for lower clocked parts. That will allow them to clock better than Intel at lower TDP. Otherwise they could have picked any CPU. Most importantly, it has nothing about IPC.
What do you mean by they have "better SMT" they are the exact same core.
its also looking like both AMD and intel will have 180watt TDP 32core servers, we have no data to say Zen core is lower perf watt then skylake EP core.
Dont be so arrogant all the time sir, there are far smarter people in this thread. don't make arbitrary connections to make your case like juan does it for Intel.
What arbitrary connections have i made, things dont just happen auto magically because of "insert buzz word" , From all we know there is nothing to say Zen doesn't have higher single thread perf per clock then broadwell. But there are architectural limits that are far more likely to be reached in SMT that limit both broadwell/skylake and Zen to around the same maximum throughput.
Facebook
Twitter