Confirmed Major Problems OC'ing Ivy Bridge - i7-3770K & i5-3570K

[jaquith]

I seen several Ivy Bridge articles by now, and I did notice some time ago odd and HIGH temperatures but the pattern is unmistakable and I'd say Confirmed - The IB has a major OC'ing problem. This is counter intuitive for the expected 'efficiency' of SB 32nm vs 22nm IB, I expected drops and higher obtainable OC's per both vCore and Temps?!

This is really bad, a 4.7GHz~4.8GHz @ 1.31v~1.36v vCore resulting in a near 100C on a Corsair H100 water loop! In comparison, my 6-core i7-3930K 4.8GHz @ 1.36v~1.40v is 70C on the hottest core using the same Corsair H100. That's at minimum 30C Hotter on the 4-Core/8-Thread Ivy Bridge vs 6-Core/12-Thread on Sandy Bridge Extreme.

The folks at EK, Koolance must be loving this and getting ready to count their money. Sure you can get a CrAzY high 7GHz+ on LN2 (Liquid Nitrogen), but I don't know too many if any Gamers playing BF3 running back and forth filling their LN2 pots??!!

ref - http://www.tweaktown.com/reviews/4663/asrock_z77_extreme6_intel_z77_with_ivy_bridge_motherboard_review/index11.html
ref - http://wccftech.com/intel-ivy-bridge-worse-overclocker-sandy-bridge/
ref - http://www.obr-hardware.com/2012/04/why-is-ivy-bridge-hot-power-hunger-and.html

 

[whatsthatnoise]

Physics, smaller die, less area to dissipate heat > things get hot. Even more noticable when you OC, since it consumes more power and therefore generates more heat.

From what I've read it seems like 4.5Ghz will still be fairly manageable most of the time, resulting in around 7x° load temps with a high end air cooler.

 

[noob2222]

looking like 22nm may be a bit more "leaky" than the 32nm. considering the size, it should have been a lot less power consumption.

Edit : link didn't go through.

only 23 watts less at load equates to less than 10% more efficient. and at stock speeds running 11 degrees hotter. Their 2600k at 5.2 ghz only hit 75c

This is the problem with die shrinks, once you get down to the theoretical limits, the parts get more and more leaky. Just because you got there first doesn't necessarily mean you win without any side effects (like sports players taking steroids)

 

[jaquith]

Physics, greater efficiency requires less current, problem the Ivy Bridge is requiring the same or more than the 'inefficiency' of the 32nm Sandy Bridge. Explanation, flaw in design.

The Chip size (cooling area) for the IB is the same as the SB, Chip size/cooling area 1.48" x 1.48" / 3.75cm x 3.75cm; the die size for the SB = 216 mm² vs IB = 160 mm² or 26% smaller. IB efficiency would further suggest (operative word 'suggest') a much lower vCore to be required (i.e. less voltage); problem the vCore isn't less.

"Current 22nm chips have high voltage, higher than they should have. The values are similar to Sandy Bridge chips. Properly should be the default voltage below 1V and it is not now. But Ivy Bridge needs a lower voltage, at the same voltage as Sandy Bridge consumption and temperature is significantly higher due to higher currents in the chip."

 

[josejones]

Ah geesh, I'm getting that Bulldozer feeling again ...

 

[brightsmith]

Physics, smaller die, less area to dissipate heat > things get hot. Even more noticable when you OC, since it consumes more power and therefore generates more heat.

From what I've read it seems like 4.5Ghz will still be fairly manageable most of the time, resulting in around 7x° load temps with a high end air cooler.

I think Ivy Bridge CPU should use bigger socket LGA 2011, not LGA 1155.

 

[brightsmith]

I just remembered, could this be the reason why the ASUS Maximus V Formula using watercooling design.

 

[popatim]

You're comparing oranges to tangerines. its not a valid comparision. IB is using new technologies like tri-gate; its not just a die shrink; so saying things like its a flawed design are not valid. Its a new design on a new nanufacturing process and you know Intel is going to keep on refining it to make it better and bettter.

I know where you're coming from and feel for those hoping for an awesome OC, but your conclusions are just wrong.

 

[bemused_fred]

Could still be work in progress, y'know. I doubt intel would release an inferior product as the 'next gen'.

 

[whatsthatnoise]

Who ever said it's inferior?!? Just because some overclockers won't see as high ghz-numbers? Poor guys... Most people don't overclock at all, anyway.

 

[Cazalan]

Not all of the design is using 22nm gates.

"UBM TechInsights has an Ivy Bridge processor marked as a 3.3-GHz Core i5-3550 chip packaged in Malaysia. It has a die size of 170 mm2, down from 208 mm2 for the current Sandy Bridge i7 2600K CPU.

In its initial tests, UBM TechInsights found gate pitches of 90-nm in the embedded SRAM array in the processor. It also found logic regions with gate lengths of 22 nm."

http://www.edn.com/article/521476-Analysts_start_Intel_Ivy_Bridge_CPU_teardown.php?cid=Newsletter+-+EDN+Fun+Friday

 

[jimmysmitty]

The title of this thread is a bit misleading sincce it acts as if every last review site and Intel has confirmed this problem and IB is released in a final silicon state. I am half tempted to change it to something more proper.

http://wccftech.com/intel-22nm-core-i7-3770k-ivy-bridge-overclocked-46ghz-11v-msi-z77agd65-motherboard/

Yet this guy with a MSI mobo hits 4.6GHz at 1.1v.

 

[jaquith]

For comparison to TweakTown, take a look vCore and Temps. Assuming you OC, I assume you have OC'ing experience, you can set the vCore Fixed, Offset and a variety of LLC & Current values depending on the BIOS.

Where's it "misleading"? I am simply re-posting information I've seen and there are tons of other threads in other forums stating the exact same thing - globally.

Duh, I prefer TH but TweakTown isn't full of Trolls, they're very reputable.

4.7GHz @ 1.36v / 98C ; assumption = fixed - http://www.tweaktown.com/reviews/4663/asrock_z77_extreme6_intel_z77_with_ivy_bridge_motherboard_review/index4.html
4.8GHz @ 1.176v / 97C ; assumption = offset - http://www.tweaktown.com/articles/4621/intel_ivy_bridge_overclocking_with_the_core_i7_3770k_and_core_i5_3570k_cpus/index2.html

 

[zakattak80]

seems like a legit concern, with so many ivy bridge cpus out being tested theirs no way intel can change and improve the chip when it's suppose to be released in a few months.

 

[absorber]

seems like a legit concern, with so many ivy bridge cpus out being tested theirs no way intel can change and improve the chip when it's suppose to be released in a few months.*

*A few weeks.

 

[FALC0N]

You're comparing oranges to tangerines. its not a valid comparision. IB is using new technologies like tri-gate; its not just a die shrink; so saying things like its a flawed design are not valid. Its a new design on a new nanufacturing process and you know Intel is going to keep on refining it to make it better and bettter.

What a load of fanboyism that is. Its a die shrink. Bulldozer was "new technology" too.

Sauce for the goose people. If you don't make excuses for AMD, don't make them for intel. If it turns out to be true at all.

 

[FALC0N]

The title of this thread is a bit misleading sincce it acts as if every last review site and Intel has confirmed this problem and IB is released in a final silicon state. I am half tempted to change it to something more proper.

http://wccftech.com/intel-22nm-core-i7-3770k-ivy-bridge-overclocked-46ghz-11v-msi-z77agd65-motherboard/

Yet this guy with a MSI mobo hits 4.6GHz at 1.1v.

I agree that everyone is jumping the gun a bit. There might be a lot of variability in these parts as well. Some that have good thermals and some that are ovens.

 

[noob2222]

The thing is if you shrink 30% but only reduce power by only 10%, internally you have a smaller area to dissipate the same amount of power (aka heat) before it gets to the heatsink.

One easy solution is to spread the transistors apart, but that just offsets the die shrink.

 

[jimmysmitty]

I agree that everyone is jumping the gun a bit. There might be a lot of variability in these parts as well. Some that have good thermals and some that are ovens.

The thing about early ES chips is that they do vary greatly compared to final ES or final consumer silicon. Thats why this could just be bunk. Some ES will overclock like crazy (normally the ones THG/Anandtech and the like get for reviews) and some wont overclock much at all (the earlier ES that go out to mobo manufactures for certain testing.

The thing is if you shrink 30% but only reduce power by only 10%, internally you have a smaller area to dissipate the same amount of power (aka heat) before it gets to the heatsink.

One easy solution is to spread the transistors apart, but that just offsets the die shrink.

Or an easier solution would be to add more transistors to the chip in a manner such as cores or L3 Cache.