why Core i7 is not overclockable as Phenom II and Core 2 ?

[bruceyg]

I am a little suprised to see this.
http://valid.canardpc.com/records.php

 

[RallyMaster]

Massive amount of heat is...massive?

 

[soccerballtux]

Larger die size? More heat.

Hyperthreading is making use of the CPU where it previously wasn't being used (in Penryn). IE, while a core may be stable at 80% load, at 100%, it's not-- Penryn at 100% load would actually be like an i7 (with the hyperthreading working) at 80% load.

 

[Nemesis 1]

Originally posted by: soccerballtux
Larger die size? More heat.

Hyperthreading is making use of the CPU where it previously wasn't being used (in Penryn). IE, while a core may be stable at 80% load, at 100%, it's not-- Penryn at 100% load would actually be like an i7 (with the hyperthreading working) at 80% load.

Why was P4c a better overclocker than AMD 64. First of all These overclocks are 1 shot wonders. Not real world usage. I think you will find that in real world usage Intel I7 overclocks better than PHII. As for the heat. Intels reading from hottest parts of die. Were as AMD isn't. Fact is AMDs heat sensor leaves muchto be desired. Take a readings from test equipment and see how it reads. Better yet in water cooled setups . Get readings from water out of Cpu for temp. Tells a differant story all together.

Another thing about those WR overclocks/How many were done with all cores enabled.

If ya want to only run 1 core. in a true 4 core cpu. Your really not testing that cpu. But only 1 core. Not impressive. 4ghz on air 24/7 that would be impresssive.

 

[thilanliyan]

Originally posted by: Nemesis 1
Better yet in water cooled setups . Get readings from water out of Cpu for temp. Tells a differant story all together.

How does that give accurate temps?

 

[DrMrLordX]

Correct me if I'm wrong, but don't Core i7 procs have the cold bug pretty bad? Core 2s do not and Phenom IIs do not. Many Athlon 64s and X2s did.

 

[BTRY B 529th FA BN]

LOL Intel's first attempt at a true monolithic 4 core design. O yeah, I have my i7 920 3841 A 437

 

[Nemesis 1]

Originally posted by: thilan29

Originally posted by: Nemesis 1
Better yet in water cooled setups . Get readings from water out of Cpu for temp. Tells a differant story all together.

How does that give accurate temps?

It doesn't . But if you use the same water block and loop on I7 and PHII in the same loop rad and tubing =. You will be surprised at the water differantial temps. Run both processors @ 3.9ghz. and tell me all about it.

 

[Nemesis 1]

I don't know if i7 has a cold bug nor do I care.Using LN. isn't something worth doing at all. Those are not 24/7 clocks and the cpu isn't capable of running those clocks long.

The guys that are really really good are pulling off high clocks on water 24/7. Which represents the true nature of the cpu running.


Ya see something about the P4P and AMD 64 x2 that I will always remember . Clock for clock is the only metric that matters. P4P o/c to 8ghz. but a conroe would make it look silly at lower clocks.

Just like the O/C records set by PHII . These are AMD O/C records. But intel has all the benchmark records at lower clocks.Thats what matters. I like O/Cing but LNC is for little boys with even smaller E-penis. They can't take world records unless they have a Clock advantage. To me talent is same clocks . Who has PC configured the best. Thats true talent.

 

[Idontcare]

Originally posted by: bruceyg
I am a little suprised to see this.

Why are you surprised?

What is it that you know about architecture and process node attributes that when taken into consideration leads you to conclude that the i7's over-clockability is surprising in relation to that of PhII and Core 2?

 

[JackyP]

Idontcare, actually I have been led to believe that i7 can overclock better than Core 2 and Ph II quadcores on high-end air and water, but on sub-zero cooling it seems to top out earlier.

I can understand why people are surprised, increased L2 latency and more stages, normally allow higher clocks (at least that's what the layman thinks).

What could cause such issues? Speed path or other architectural issues? Sub-zero cooling is certainly not limited by heat, now is it?
I'd venture to say that a new stepping will solve those problems and slightly decrease temps (together with even more mature process tech).

 

[Tempered81]

i7's do about 3.6-4.2ghz and phenomII's do about 3.8-4.05ghz on aftermarket air/water

It's just LN2 where the struggles of managing QPI, CPU, HT, vcore and all on the i7 give the phenomII the advantage

penryn oc's higher than both, but it's relatively simple, vcore & high fsb

 

[exar333]

This title is very misleading...

 

[bruceyg]

--"Intel has ensured less expensive Core i7 versions,the 920 and 940,will begin throttling back their clock speeds once they reach this threshold. Sadly, it looks like the days in which a $150 CPU is able to reach the performance of a $1,000 processor are over,at least for now. While small tweaks are still possible, major performance increases through overclocking are no longer an option. Unlike the Core 2 or Pentium D processors, the Core i7 CPU monitor the current it draws as well as its power consumption?if it exceeds a certain level, the processor automatically reduces its clock speed."

this might be the reason why a core i7 doesn't overclock as well as a c2d or a phenom2.

 

[Idontcare]

Originally posted by: JackyP
Idontcare, actually I have been led to believe that i7 can overclock better than Core 2 and Ph II quadcores on high-end air and water, but on sub-zero cooling it seems to top out earlier.

I can understand why people are surprised, increased L2 latency and more stages, normally allow higher clocks (at least that's what the layman thinks).

What could cause such issues? Speed path or other architectural issues? Sub-zero cooling is certainly not limited by heat, now is it?
I'd venture to say that a new stepping will solve those problems and slightly decrease temps (together with even more mature process tech).

The entire layout and design of the chip is optimized to intentionally target a particular clockspeed/power-consumption/IPC envelope.

It is no more surprising to me that i7 doesn't clock to 6GHz under LN2 than it is surprising to me that Phenom II does clock to 6GHz under LN2.

Yes you have what some would consider to be artificial speed-limits (as in consequences of architecture choices) placed on the peak circuit timing that an IC is going to achieve just as you equally have clockspeed limiting aspects due to your underlying process technology. Speedpaths in circuit layout, systematic parametric weakness stemming from process variation in xtor formation steps, etc.

For example the gate thickness determines Idrive as well as a whole host of other relevant xtor speed metrics. Well you can't drive your xtors at 6GHz regardless of temperature if the gate oxide thickness is greater than a minimum threshold thickness for all the transistors in the entire circuit loop of interest.

But having the gate oxide thin enough as to allow 6GHz operation would also mean it is needlessly thin for creating chips that operate at 3GHz as the gate oxide could be thicker and still hit 3GHz operation but with less leakage and power-consumption. (this is just an example, by no means am I implying this is the sole reason AMD hits 6GHz and Intel does not, there are literally hundreds of viable reasons that can each explain the outcome on their own rights, ergo my question to OP above)

But the point of asking why chip XYZ can hit 6GHz while chip ABC can hit 5GHz is to ignore the rest of the variables that were optimized during design/layout as well as process tech development...to name a few - power consumption, IPC, yields, production cycle-time, production costs, development cycle-time, development costs, etc, etc. Until you know more of the backstory you can't really expect yourself to be capable of correctly knowing why a given chip is only able to hit a certain frequency in peak overclocks.

 

[IntelUser2000]

I heard the D0 stepping coming with Core i7 975 will have much improved overclocking headroom.

 

[Diogenes2]

....if it exceeds a certain level, the processor automatically reduces its clock speed."

A feature that can be disabled..

 

[dmens]

Originally posted by: IntelUser2000
I heard the D0 stepping coming with Core i7 975 will have much improved overclocking headroom.

yes one of the things worked on was speed at higher core voltages.

 

[FAHgamer]

Originally posted by: JackyP
Idontcare, actually I have been led to believe that i7 can overclock better than Core 2 and Ph II quadcores on high-end air and water, but on sub-zero cooling it seems to top out earlier.

I can understand why people are surprised, increased L2 latency and more stages, normally allow higher clocks (at least that's what the layman thinks).

What could cause such issues? Speed path or other architectural issues? Sub-zero cooling is certainly not limited by heat, now is it?
I'd venture to say that a new stepping will solve those problems and slightly decrease temps (together with even more mature process tech).

What do you mean "increased L2 latency"? Compared to what? (Nehalem's L2 latency is 11 cycles, whereas in Penryn it's 14-16 - I don't remember how much it is exactly in the latter, but it is much more than Nehalem.)

Edit: Oops, my bad, you meant Core2 and PhenomII 's higher L2 latency; I got it now.
But if you factor in the fact that Nehalem's L2 cache is smaller in total size and larger in cell size, it should be able to reach much lower latency at the same voltage.

How many pipeline stages does the Nehalem architecture have?

 

[JackyP]

Originally posted by: IntelUser2000
I heard the D0 stepping coming with Core i7 975 will have much improved overclocking headroom.

Cherry picked i7 975 did 5ghz on air IIRC (over at xtremesystems).

Some golden i7 do 4,5-4,6 ghz on air/water, how many PhII and penryn QC can do it these days? I am not sure.

Originally posted by: FAHgamer
What do you mean "increased L2 latency"? Compared to what? (Nehalem's L2 latency is 11 cycles, whereas in Penryn it's 14-16 - I don't remember how much it is exactly in the latter, but it is much more than Nehalem.)

Edit: Oops, my bad, you meant Core2 and PhenomII 's higher L2 latency.
But if you factor in the fact that Nehalem's L2 cache is smaller in total size and larger in cell size, it should reach much lower latencies at the same voltage.

No, actually I didn't mean what I said in this case. I was thinking about L1 latency (3->4 cycles). It is generally believed that increased L1 latency (may) allow higher clocks, all else being equal. Well, obviously Nehalem kept nothing else equal and SMT may have made such an increase necessary anyway or anything else for that matter.

 

[Jabbernyx]

Originally posted by: bruceyg
I am a little suprised to see this.
http://valid.canardpc.com/records.php

Why would you be surprised? Timing closure is a tough part of the design. Couple that with power, IPC and other requirements and its actually a damn miracle to successfully pull off a successful IC design. Unless you designed it, you wouldn't know attributes such as the length of the critical path(s), individual transistor characteristics, etc. and would therefore be a little ill-informed to make a judgement call on the i7 vs. other CPUs.

 

[aigomorla]

ive had both the high end, and low end.

On both cpu's i have had no problems pushing it to 4.0ghz with HT on.

http://i125.photobucket.com/al.../aigomorla/Final-1.jpg

http://i125.photobucket.com/al.../aigomorla/Linpack.jpg


Infact its quite common to see an i7 @ 3.8-4ghz on the highest tier sink (lapped TRUE) or water.

So i dont get the statement on i7 not being overclockable then PHII.

In a straight overclocking competition using retail chips, My 965 is as close to retail as you can get while being an ES, the i7 would wipe the floor off PHII in overclocking.

Of course the setup required cost 2.5x more. But were looking at OC not price.

 

[BTRY B 529th FA BN]

Originally posted by: dmens

Originally posted by: IntelUser2000
I heard the D0 stepping coming with Core i7 975 will have much improved overclocking headroom.

yes one of the things worked on was speed at higher core voltages.

? did you mean higher speeds at lower core voltages?

 

[dmens]

Originally posted by: BTRY B 529th FA BN
? did you mean higher speeds at lower core voltages?

well that was worked on too, but that does not interest overclockers right?

 

[IntelUser2000]

Originally posted by: FAHgamer

Originally posted by: JackyP
Idontcare, actually I have been led to believe that i7 can overclock better than Core 2 and Ph II quadcores on high-end air and water, but on sub-zero cooling it seems to top out earlier.

I can understand why people are surprised, increased L2 latency and more stages, normally allow higher clocks (at least that's what the layman thinks).

What could cause such issues? Speed path or other architectural issues? Sub-zero cooling is certainly not limited by heat, now is it?
I'd venture to say that a new stepping will solve those problems and slightly decrease temps (together with even more mature process tech).

What do you mean "increased L2 latency"? Compared to what? (Nehalem's L2 latency is 11 cycles, whereas in Penryn it's 14-16 - I don't remember how much it is exactly in the latter, but it is much more than Nehalem.)

Edit: Oops, my bad, you meant Core2 and PhenomII 's higher L2 latency; I got it now.
But if you factor in the fact that Nehalem's L2 cache is smaller in total size and larger in cell size, it should be able to reach much lower latency at the same voltage.

How many pipeline stages does the Nehalem architecture have?

Nehalem has 16 stages.