Ivy Bridge OC Results Came in (Engineering Sample)

[Absolute0]

So one of the good people at XS got their hands on an ES i7-3770K

Cooling is Thermalright True Spirit 140

4.8 Ghz 1.30v fully stable, and the chip running hotter than blazes (broke 100C, far hotter than his previous 2600K, and yes he remounted heatsink to no effect)

Speculation is that either 1) ES sample is weird and thus hot or 2) the 3d "low leakage" gate technology @ 22nm gets HOT when Vcore is increased

IDK if this is representative of all IB chips or not. But for now it's the only OC results we got other than a couple screens that didn't show much. Anyway a thanks to Meteor|War of XS for sharing

http://www.xtremesystems.org/forums...neering-sample-landed-on-my-desk-today./page4

 

[Idontcare]

Temperature is proportional to the die area and the power usage.

A 100W chip with 10mm^2 die is going to have much much higher temperature than a 100W chip with 100mm^2 die.

So IB, being much smaller than SB, is going to have higher temperatures at every point on the power-consumption curve regardless the underlying process technology.

 

[tynopik]

Temperature is proportional to the die area and the power usage.

A 100W chip with 10mm^2 die is going to have much much higher temperature than a 100W chip with 100mm^2 die.

So IB, being much smaller than SB, is going to have higher temperatures at every point on the power-consumption curve regardless the underlying process technology.

but the hope is that a new node would require less power in the first place

 

[Idontcare]

but the hope is that a new node would require less power in the first place

Sure, typically you get 20% lower power at the same clockspeed. But the die shrinks by more than 20%.

You lose on the temperature end of that equation, which is why CPU HSF's get larger and larger.

If you don't keep the temps down then the leakage will go up. It is all quite self-defeating.

 

[piasabird]

Looks like a design issue. Need to design with 6 or 8 cores, to spread out the landscaped. Maybe a 9core processor could spread it out to 3 core cubed. Then alternate between cores to distribute both the processing and the heat dissipation. Similar to how memory works. Interleving processing. Might also have to spread out the memory registers.

There may be a point at which smaller is not better. The more you cram electrical current into a smaller space the more heat is produced. One way this was handled in the past is to design transistors to use less power. I think this new design is using more tri-state switches which is Intels attempt to go smaller. Maybe this is a sign that necessity will require a differnt kind of design which revolutionizes what we currently are using.

It is still too eary to tell. Since this is a new core, possible a slight redesign will help it a bit.

 

[Magic Carpet]

Temperature is proportional to the die area and the power usage.

Agreed. Explains why Thuban is one of the coolest running modern processors.

 

[Joseph F]

Would a solid copper base heatsink help with the temperatures?
I would think that if the copper covers 100% of the die area, it would cool better than if it didn't, especially with the smaller die of IVB.
:hmm: The cooler used in this OC has a solid-base. I wonder how bad the temps would be with direct-contact heatpipes. D:

 

[frostedflakes]

Yeah those temps are pretty high. Even with the smaller die and having less area to spread heat over, they still seem higher than they should be, wonder if that coupled with high leakage is the problem.

Looks like these might respond really well to exotic cooling, but will be kind of disappointing if they can only hit 4.8 or so on high end air because of temps holding you back. I wouldn't even feel comfortable running one at 4.8 if they load at 90-100C, to keep them at a more reasonable 70-80C under load you might have to back it off to like 4.5-4.6. Not much better than SB can do. :'(

Hopefully the retail chips run cooler.

 

[Grooveriding]

So IVB-E may be the overclocker's choice for 22nm?

 

[Absolute0]

Sure, typically you get 20% lower power at the same clockspeed. But the die shrinks by more than 20%.

You lose on the temperature end of that equation, which is why CPU HSF's get larger and larger.

If you don't keep the temps down then the leakage will go up. It is all quite self-defeating.

Thanks for the link, this has greatly improved my understanding.

I had kept wondering why IB was running hotter, and I understand thermal leakage better now.

Still I would hope that retail chips aren't as bad as this ES... But it was theorized on XS that this is why the IB mobile line was delayed! I can just imagine those poor laptops burning up.

 

[piasabird]

I remember this from the past. Intel comes out with a new processor design and it is barely better than the previous one and then it gets a little faster and a little faster. I think behind the scenes they are making tweaks to the design to make it a little faster or run cooler or whatever it is that they do to get more performance or run faster.

I am still using two computer with two processors. One is a E7200 C2D and one is a i-5 2500K. The quad seems faster, but a quad should be faster. The i-5 is not really that much better than the C2D. It is really the Intel HD3000 video that makes the difference. I usually use my computer for a little light gaming and mostly watching video, So I dont really stress out the computer that much. I also never overclock.

 

[VirtualLarry]

4.8 Ghz 1.30v fully stable, and the chip running hotter than blazes (broke 100C, far hotter than his previous 2600K, and yes he remounted heatsink to no effect)

Speculation is that either 1) ES sample is weird and thus hot or 2) the 3d "low leakage" gate technology @ 22nm gets HOT when Vcore is increased

Doesn't a lot of heat indicate high leakage? IDC?

 

[nyker96]

Agreed. Explains why Thuban is one of the coolest running modern processors.

cause that thing is huge, got a ton of die area. But this isn't a good thing either, smaller die is cheaper to make. I guess everything from temp, power draw, die area, process tech, cost of manufacturing etc all a balancing act. The cpu with the right combo wins. Still, I'd get a IB over Thuban any day, just the fact that you can do close to 5Ghz with less power draw is awesome, not to mention great IPC. Only wish AMD can put out something just remotely competitive.

 

[grkM3]

I didnt read the whole thread but did the op get the latest bios and latest software to monitor temps?

I can see up to 10-12c difference in idle and load temps from running a few different bios on my asus board.

Lets hope there is a bios bug and If a sandy shuts the computer down at 100c shouldnt the temp be lower on a smaller chip and 22nm? When I saw 104c and the computer not even throttling the clocks I figured it was just not reporting the temps right.

 

[Magic Carpet]

nyker96,

It's best to wait for retail samples. IB should make AMD/Intel's current offerings cheaper.

grkM3,

Temps could be right though. Remember, this is ES, fail-safe could be off.

 

[T_Yamamoto]

nyker96,

It's best to wait for retail samples. IB should make AMD/Intel's current offerings cheaper.

grkM3,

Temps could be right though. Remember, this is ES, fail-safe could be off.

Hopefully
but i see Core 2 Quad Extreme still at 1k+

 

[Don Karnage]

I'll wait till i get my hands on a 3570K and test for myself

 

[Tempered81]

So IVB-E may be the overclocker's choice for 22nm?

IMO the oc potential of your 3930k, a 3960X, or 2700k / 2800k + higher bins will not be beat by 22nm tri-gate asics because of the heat. They def wont fix the heat with 6 or 8 cores.

If you want to beat 4.8 or even your own amazing 5.0 on ivy, you're gonna need water. Tri-gate heat is the culprit. In some ways maybe the physics of dropping to 22nm too, since intel has been talking for about 6 years now about the physical limitations of such a short distance between paths. That was why the 3d transistor was invented in the first place to let them continue to shrink beyond gulftown. 1.3micron down to 32nm always brought more mhz within the same Tjunction. ie oc an e6400 to 3ghz, a i7 920 to 4ghz and a 3930k to 5ghz (all with the same thermalright ultra 120 air cooler). Different story for the 3770k, you're gonna need water to continue this trend. It's the 3d transisitor design mainly, and secondarily the physical limitations of the chip size and lithography.


Sad but true: A 2700K is going to have higher overclocking potential than a 3770k. 5.0 versus 4.7 (or relative numbers). You get the +5-10% IPC with Ivy, but you get the +15% speed potential with Sandy...

Unless you have cold

 

[rgallant]

and who said you don't need big water for a 77w chip. lol

-should prove to be interesting given the tick-tock cycle , if the high oc w\voltage causes high temps , E1 might not see a new stepping with tock coming in 12 months.

 

[T_Yamamoto]

and who said you don't need big water for a 77w chip. lol

-should prove to be interesting given the tick-tock cycle , if the high oc w\voltage causes high temps , E1 might not see a new stepping with tock coming in 12 months.

you make a valid point

 

[Grooveriding]

IMO the oc potential of your 3930k, a 3960X, or 2700k / 2800k + higher bins will not be beat by 22nm tri-gate asics because of the heat. They def wont fix the heat with 6 or 8 cores.

If you want to beat 4.8 or even your own amazing 5.0 on ivy, you're gonna need water. Tri-gate heat is the culprit. In some ways maybe the physics of dropping to 22nm too, since intel has been talking for about 6 years now about the physical limitations of such a short distance between paths. That was why the 3d transistor was invented in the first place to let them continue to shrink beyond gulftown. 1.3micron down to 32nm always brought more mhz within the same Tjunction. ie oc an e6400 to 3ghz, a i7 920 to 4ghz and a 3930k to 5ghz (all with the same thermalright ultra 120 air cooler). Different story for the 3770k, you're gonna need water to continue this trend. It's the 3d transisitor design mainly, and secondarily the physical limitations of the chip size and lithography.


Sad but true: A 2700K is going to have higher overclocking potential than a 3770k. 5.0 versus 4.7 (or relative numbers). You get the +5-10% IPC with Ivy, but you get the +15% speed potential with Sandy...

Unless you have cold

Is this going to wind up translating over to GPUs as well ? They already run at low clocks compared to CPUs, is that going to go down further as they move down to these smaller processes ?

 

[Ferzerp]

GPUs have the advantage of having the quintessential embarassingly parallel workload, so they can always just go wider if need be.

 

[Tempered81]

Is this going to wind up translating over to GPUs as well ? They already run at low clocks compared to CPUs, is that going to go down further as they move down to these smaller processes ?

Dunno, likely. Tahiti is clocking higher than i thought it was going to.... 1300mhz... woah. 40nm was getting like 960-1050, now we see 1250-1300. I bet the gk104 clocks high compared to gf114 and gf110. No idea about further down the road.

 

[ehume]

Dunno, likely. Tahiti is clocking higher than i thought it was going to.... 1300mhz... woah. 40nm was getting like 960-1050, now we see 1250-1300. I bet the gk104 clocks high compared to gf114 and gf110. No idea about further down the road.

That's a 28nm process, right?

So a 32nm process SB runs cooler than the 45nm process. And the 28nm process is running cooler than the 40nm process. But the 22nm process is running hotter than the 32nm process (maybe).

So is the 28-32nm process zone some sort of sweet spot? Where reduced resistance crosses with increased leakage, and below that zone the leakage predominates?

Does that sound like a possible explanation?

I was worried that the progress of ever cooler cpu's was going to render Big Air and WC obsolete. Perhaps not.

 

[Paul98]

After reading that thread, I am not sure those temp numbers are actually true. Or if they are true the heat sink may not be getting full contact. I will be waiting to hear some more info and testing.