sandybridge vs ivybridge vs haswell

[Hellhammer]

Pentium was an entirely different thing that thing potentially supported enormous clock speeds. Compare the AMD CPU which stayed architectureally mostly the same. They had a quite steady increase in clock speed over all process nodes. Check out the actual released CPUs of the last years and what overclocking freaks got from them. I think saying we have been stuck just below 4Ghz is a bit of a misinterpretation. Mainstream 65 was about 2.4 Ghz, 45 was around 3Ghz, 32 3.4. If you check out the CPU list on notebookcheck sorted by TDP or process your theory isn't well supported by data. Pentium doesn't count it is the oddball.
Maximum Overclocks went from 4 to 5 then 7 Ghz and now we are at around 8 something Ghz. A magical 4 Ghz ceiling doesn't exist I think.
Even if there was, I'd expect some TDP downgrades that maybe move the lower end quads to 35. Kill of the 55W TDP bucket and make the 45W the new highend. There is just no sign of the supposed 2 node jump till now. What became of all the promises of 22nm TriGate. You have to admit what is known till now is underwhelming. I expected more than a Penryn like shrink but that is what it is now.

I never said 4GHz can't be achieved, I simply said it's the point where power consumption goes up exponentially. Below 4GHz, the increase is linear like the graph I linked shows. Heck, for decades the performance increase was solely based on frequency increases, so yes, we have seen an increase in clock speeds. However, what we haven't seen is +4GHz in mainstream parts.

The only logical explanation is that silicon doesn't operate at its ideal frequency at over 4GHz. Yes, you can make it vibrate at higher frequency but your performance/power ratio won't be as good, and that's what consumer chips are all about. We've been stuck at 3.7-3.9GHz for years, and we should have gone past this point years ago if the frequency/power graph was totally linear and every die shrink helped.

 

[dusk007]

It could be but I did understand what you meant. It is jsut I think that the point where the exponential curve really turn upward was different on all the last generations and just because Penitum 4 on 90nm didn't do well past 4Ghz doesn't mean it is the same for <22nm process nodes.
I mean this ceiling after which the power consumption grows much more than what it is worth was different over all the last gen chips and effecttively determined what mainstream parts are clocked at. Used to be 2.5 Ghz, was 3 Ghz is now around 3.5 Ghz may be 4 or more Ghz.
Can be that it is a Si problem and they need new Ge or whatever they are working on but could also be that 90nm Si had the that ceiling and 22nm is something else.
Afaik the problem of Pentium 4 was the power consumption and the low performance. The architecture was good for 10+ Ghz. They are running Pentium 3 test chips now from a tiny solar cell. Who says that a Pentium 4 at 22nm couldn't run 10 Ghz at 22nm with just 25W or so. Would still be slower than a current CPU but still it could be.

Again just because they stopped with Pentium 4 and 90nm at 4Ghz back than does not imply that this is the magical Si Transistor limit. It is a hypthesis but no more.
Actually just talking about this I remebered a news once about IBM building a Silicon transistor laced with some Germanium and it ran at 300 Ghz at room temperature and 500 cooled to almost 0 K. According to the IBM researcher we are no where near the limits of Si Transistors. I really doubt that this hypothesis is true and it definitely isn't directly implied by the information we two have.
Looking at some google scholar articles it seems like there are limits but we are no where near them and there there are so many proposed new transistors that can fix this or that.
Intel brought TriGate to fix Leakage much the same with High K metal gate. If there was a frequency problem they could fix that too but afiak it is just a power consumption problem and that is pushed every shrink a bit further. Otherwise they'd use ever more cores instead of more clockspeed. Like with ARM cores which are in the same situation just a different power budget. They moved from 500 -> 600 -> 1 Ghz -> 1.5 Ghz and are expected to hit 1.8-2 Ghz soon. Nobody could say their is a 1 Ghz barrier.

 

[Hellhammer]

It could be but I did understand what you meant. It is jsut I think that the point where the exponential curve really turn upward was different on all the last generations and just because Penitum 4 on 90nm didn't do well past 4Ghz doesn't mean it is the same for <22nm process nodes.

It's not just Pentium 4. The graph I linked you earlier was based on Clarkdale. Here are two other graphs showing the exact same thing, at two different process nodes

http://www.xbitlabs.com/images/cpu/power-consumption-overclocking/Charts-1/core-i7-950-1.png
http://benchmarkreviews.com/images/...dge_OC_Analysis_CPU_Power_vs_Voltage_Idle.png

Maybe Tri-Gate will allow the 4GHz barrier to be broken, who knows. However, right now when you go past 4GHz, the power consumption goes up exponentially. Judging by IVB's clock speeds, Tri-Gate won't help much as we are still stuck at below 4GHz.

 

[macintoshi]

what come after hashwellbridge? Because I will wait until that one!

We all understood, its haswell, not hashwellbridge or hashwell, so please stop beeing idiotic. Doesn't shows you are a little better or smarter, but dumber on this matter.Thz

----------

Not to be picky here, though it is just simply Haswell. Not Hashwell or Hashwell Bridge or Hashwellbridge.

One word pronounced: Has well.

Thz for the clarification, didn't researched well on that matter, thats why i misspelled it or missunderstood it, but got it now, thz

----------

Broadwell(bridge )2014
Skylake 2015
Skymont 2016
---
Skynet 2017
2018: Skynet starts a nuclear war
2019: T-800 (Arnie)
2020: T-1000 (the Cop)
2021: T-X (the hot chick)

http://en.wikipedia.org/wiki/Intel_Tick-Tock

Well as i said, i don't wait till haswell! but can wait some months, for ivy-bridge

----------

Buy now and enjoy while you can, since the world is going to end in 2012.

Hahah, well hopefully not