The temperatures are what will confuse and surprise people. It really does run this hot - Sandy Bridge and Ivy Bridge, even though they are similar architectures, are made on a different process and using different transistor designs. This has seemingly had a very large effect on how the processor reacts. We may see it improve over time as the process matures, but as it stands, 95C at 1.30V is a real value - for a liquid cooler outside a case. So in a case it would perhaps be worse.
Wouldn't this be expected from a pure physics perspective. Since a 22nm node has drastically shrunk the surface area in contact with the heat sink. Especially with the IGP off. The heat is now concentrated to a much smaller area, so it is all the more difficult to transfer that heat to the heat sink.
That's true, given the same heat; however, if you're using less current (lower voltage), the heat generated should also be less. If the current:surface area is a perfect ratio, the correlation of heat should also be close to 1. So as size decreases, so should heat.
The problem, though, is that the efficiency of the transistors are questionable. Sure there may be more, but because you're creating them at a smaller scale, it could be less efficient releasing more electricity/heat to the external system.
In addition, the materials used in the tri-gate setup is unknown, which could also impact the heat retained in the system.
As Ian stated, as the manufacturing process matures, they could find ways to improve the process. This could lead to reduction in heat, and thus an increase in performance.
In a move quite uncharacteristic for me I'd just like to praise you for the very worthy and interesting article well done and setting the skewed misperceptions record straight.
Very good. Much appreciated. Definitely unique and cutting edge. Worthwhile.
To answer your original question, for these tests I was running on discrete graphics without Virtu MVP, so the iGPU should have been power gated into a low state.
Would it be possible to adjust the graphs to have similar ranges for their axis? It'd the make the two first two graphs at stock and 4.4 Ghz much easier to compare. This would also apply to the constant 1.25V graph and the underclocking graph.
Also would it be able to label each point on the PovRay test with the clock speed at that particular voltage (or are these values obtained at a constant clock)?
Apologies, I should make it more clear. The graph with a single line of Temperature against voltage should be read in conjunction with the overclock graph. So at 0.90 V, we are at 3.9 GHz, 1.1V @ 4.4 GHz, 1.15V @ 4.6, 1.20V @ 4.7, 1.30V @ 4.8. Which means that perhaps 1.15 V at 4.6 GHz is a happy medium.
"Ivy Bridge is now released and behaves differently to Sandy Bridge."
That probably would better read with the "to" replaced with "than", a word that means a comparison. In your sentence above, Ian, to what does the phrase "to Sandy Bridge" refer? Do you mean in a comparison to Ivy Bridge, perhaps? Then, the word you used, "to", is wholly inappropriate and meaningless. Instead, that sentence would more properly read with the "to" replaced with "than", a word that means a comparison, i.e. I'd rather have a bottle in front of me THAN a frontal lobotomy.
So, it'd read, "Ivy Bridge is now released and behaves differently than Sandy Bridge." The sentence now makes the comparison you were trying to say but failed.
I could also write "Ivy Bridge is now released and it behaves differently to Sandy Bridge." or a variety of other ways. Along with this, I have to deal with avoiding British idioms for a mainly American based audience, as well as the issue of company plurality which seems very alien to me. However, if you do find other areas of English in one of my articles that perhaps need a touch up, please be tactful and email me (ian AT anandtech DOT com) rather than habitually pick out one phrase and demonstrate your horrific disgust about it on a public forum. Writers work their fingers to the bone to get some of these articles done in time, and we would appreciate it if you didn't beat down their hard work without at least attempting to contact them first as a polite indication.
The form you used is fine in British English, so he is wrong and crass. I don't think you should need to kowtow and (unreasonably) modify your language just for those who speak another form of English, given that it is perfectly understandable.
Some exposure to the outside world might be an educational and enlightening experience for them :-).
actually there's nothing wrong with using "to" instead of "than" when making comparisons. Google it to see its regular use, u can then proceed on your littl grammar nazi crusade and email the thousands of people on the internet that use "to" to make comparisons. I'm assuming ure not a native speaker as a native speaker correcting another native speaker on how to use the language correctly is laughable.
I'm an English teacher and would never mark such a sentence wrong. Language develops, especially one as international as English which is used in all corners of the world and will develop little differences.
Exactly. Even if there was a slight deviation (not that there was in this case), what matters ultimately is that the author still communicates what it is that (s)he is trying to convey.
You're taking a grammatical preference and asserting it as fact. It is not fact; different to /from / than is a free option - your chosen example uses a different sentence structure.
To do this in such an overtly critical and unfriendly fashion is repugnant; the fact that you are also incorrect merely serves to make you look very silly.
C'DaleRider doesn't mind looking stupid. He made a similarly ignorant and rude grammar comment a few weeks ago. He's just trolling, trying to disrupt the comment stream and generally mess up a good thing. That's the way some people are, sad to say.
If you shrink crystalline Si dimensions, heat conductivity breaks down as less phonon modes can fit into the volume and contribute to heat conductivity. For the fin-FET like tri-gate design this should be even more severe, as we're quenching the phonon modes in 2 dimensions. Might well be that the problem is to get the heat out of the transistors in Intels 22 nm process. If this is correct, things are only going to get worse at even smaller process nodes. The problem is then not about putting heat into the system, but rather to get it conducted even to the heat spreader or back side of the chip.
Thanks Ian, this is the first I've heard of these issues... I had been holding out for Ivy Bridge results before choosing between it and Sandy Bridge (which will hopefully now drop in price).
The difference on my i7 920 between HT-on and HT-off with respect to voltage/temperature under load is incredible. Do you have any evidence to base your statement up?
Maybe I'm just a dummy, but what is the stock voltage for Ivy Bridge? Does the article even say? The closest statement I found was 4.4GHz at 1.05V which is "close to stock."
I think it could have been a little bit more clear in the article.
Big case + strong cooling + don't mind high power usage (+more noise) = SB @ 5-5.1+GHz will be better (faster)
Small case + not-so-good cooling + power usage figures matter / low noise is important = IB @ 4.6-ish will be the better choice (uses less power / will be faster than SB @ 4.6-ish with the better IPC from IB)
So yea, it'll matter on how you want to run your system.
Ian hit ~4.5 GHz at a mere 1.05 V with IB. That's significantly less power consumption than anything SB could do in this performance range. Top clock may be a bit lower, but you're saving tons of electricity with IB. And if you're buying new: IB should be cheaper anyway. Exception: getting a good deal on a used chip, saving 100+$, would probably be worth it.
PLEASE for the love of god use Prime95 for stability/temps.... are we forgetting that prime95 IS A REAL WORLD APPLICATION??? Plenty of people are actively searching for primes!!
Instead, you are using a synthetic heat generater (OCCT) and another real-world that doesn't get as hot as prime...
i agree, prime95 is a "certified" heat generator... and a D@mn fine stablity checker ( but requires alot of time)... 6 hours for 80% assured stability, 24 hours for 95% or "good-enough" (of course for "guaranteed" stability run at stock, needed for checking primes)
if you need an absolute worst case power draw/heat, run prime95 on all but-one-core, and run furmark on the GPU...
The sweet spot for performance vs heat seems to be at 4.5 GHz @ 1.1v. I've been reading the IB threads in the CPU OC forums on Anandtech. The few enthusiast members that have been testing 3570K haven't come close to 4.5 GHz at 1.1v. In fact, the best I've seen is 4.5 GHz at 1.236v.
What settings did you use to get a stable 4.5 GHz OC at 1.1v?
Well done Ian, you covered a lot of ground and all the relevant points in a nice, graphical way.
If you ask me, 1.15 Volt is the 'Sweet Spot' for this processor, you can crank upto 4.6 GHz (depending on processor lottery, some variation will exist between processors and future steppings), while still keeping temperatures reasonable and not compromising long term reliability.
Power Consumption ^ 3770K I / I _______/___________ 2700K I _________ / / I____/ __/ I ________/ I_________/ I --------------------------------------------------> Avg. Gaming Performance
This article is lol-worthy as it seems the author believes there is a linear relationship between voltage and power. Your increase of .35 volts nearly DOUBLES the power consumption of the CPU.
Power = voltage ^2 / resistance
(1.25v)^2 / (.9v)^2 = 193%
Ivy bridge does not hate voltage, but it does hate people who don't understand ohms law.
Absolutely love the methodical nature of your investigations and your understanding of the subject matter Dr. Cutress. With it and the various ways you've illustrated things, it's givenme in a very short time what seems to be a good understanding of the basics of how Ivy and the current 22nm Tri-Gate processors behave, and how to best overclock them.
No mean feat considering the differences. Thanks very, very much :)
Everything I've read about high temperatures on IB suggests that there is a large amount of thermal resistance between the active parts of the IB chip and the outside surface of the integrated heat spreader (IHS) to which we attach our heatsinks. It is suggested that the IB heatspreader is not soldered onto the chip, but uses thermal paste of unknown quality. That would go a long way to explaining the temperature problems found when overclocking IB. Of course there are other factors such as the increased power per unit area of the chip, and heat may be trapped by the 3-dimensional nature of the new transistors.
If you can do so, please pop the lid off your press-release ES CPU and re-run some of the tests with your heatsink mounted directly on the CPU die with top-quality TIM paste. What will happen to the CPU otherwise? If it has to go back to Intel, so what? They gave it to you to test. Well, you'll have really tested it :)
How hard would it be to remove the IHS and just connect the heatsink directly to the chip (like in the old days)? Keep in mind the physical size of the die is smaller now so you'll have less surface area to work with. Add in the new tri-gate process and how well that will dissipate heat and you have a heat-gate (tm, hehe) trifecta.
The original article on which theTPU one linked above is based: http://www.overclockers.com/ivy-bridge-temperature... They do wonder whether the retail CPUs will have soldered IHSs instead of the TI paste used on the ESs.
And I forgot to say thanks to Ian Cutress for writing this article. The charts setting out the volts/temps/max clocks envelope of the CPU are excellent.
The AMD HD 7970 is the first video card to use PCIe 3.0 bandwidth, and that is a single card. So, PCIe 3.0 on new motherboards is good to see.
However, games that use the CPU heavily and high bandwidth GPUs that tax CPUs beyond their limit [if not overclocked] is NOT a good thing.
Intel needs to introduce a better heat disipation method for the latest generation of CPUs that allow core clock changes [K series, for example], to allow higher overclocks at reduced temperatures and voltage.
Why have a motherboard that is PCIe 3.0 capable if the CPU is over-taxed and bottlenecks [game] video?
Intel needs to fix this problem quickly or they will be selling fewer of the latest K series CPUs.
i think after amd, now intel crashed headlong into the 'wall' of Moore. the amount of ipc increase we might still see over the next 10 years might be much smaller than most people realize. the amount of clock increase we might see over the next 10 years might be much smaller than most people realize. ok, they can still add more cores, but for 9 of 10 applications, 10 % ipc/clock increase will do more than doubling the number of cores.
implications of moore's law already failing this bad in 2011-2012, which is much earlier than anticipated could be 'dramatic' .
"Ivy Bridge behaves differently *than* Sandy Bridge [behaves]" I don't know why you keep putting "to" in place of "than".
You may also write "Ivy Bridge over clocking is different from Sandy Bridge".
In any event, "different to" has no meaning in standard English. You are not a commenter on a blog, you are professional writer. Some errors I can tollerate, but this is just awful to my ear.
Would be better for new builders to buy a 2500k instead? It seems like there's a lot more risk involved with Ivy Bridge, esp. for people new to overclocking (like myself). Not really sure what to do from this article.
Complete noob here on a first build, hoping not to screw up. ^_^
"Increased density and new transistor technology can certainly give rise to more heat. But we've seen these things before, and each time, cooling and packaging technology manage to cope with the resulting output. What could be holding Ivy Bridge back? It could be the fact that Intel decided to use thermal paste instead of the usual fluxless solder (you'll need to have a solid grasp of Japanese to understand the original article, or use Google Translate) between the CPU die and heat spreader.
Using a box knife (we do not recommend doing this at home), the author of the linked story pried the heat spreader away from the chip and was able to replace the cheap paste Intel used, trying out both OCZ Freeze Extreme and Coollaboratory Liquid Pro thermal pastes. The OCZ offering allowed for 1.55 V at 4.9 GHz, while the Coollaboratory material ramped up to 5.0 GHz, operating stably. This was accomplished even with air cooling, although the author did not use a stock heat sink, opting for a Thermalright Silver Arrow SB-E instead (Ed.: this information was pulled from the original Impress PC Watch site with Google's rather shaky translation of Japanese). If there is a smoking gun in this equation, we think this is it, especially considering that the researchers at Impress PC Watch managed 20% more efficient cooling."
Hi, Thank you for this article. Using the Max. Overclock vs. CPU Voltage graph, I figured out that my CPU was less overclockable than yours, but I reached easily 4.6 GHz @ 1.25 V. At the same frequency, le processor was unstable @ 1.20 V. System : i7 3770k, GA-Z77-D3H (BIOS F18), 2x2 GB Kingston 1600 DDR3. Idle : 33.5 °C, 80 W Load : 72 °C, 175 W These settings are used on a 24/7 basis, mainly for images and video processing (Win 8.1)
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
84 Comments
Back to Article
Shadow_k - Monday, April 23, 2012 - link
Did you disable the igp?95c you mad at 1.3V wow I wonder if Haswell face the same heat problem when overclocked
IanCutress - Monday, April 23, 2012 - link
The temperatures are what will confuse and surprise people. It really does run this hot - Sandy Bridge and Ivy Bridge, even though they are similar architectures, are made on a different process and using different transistor designs. This has seemingly had a very large effect on how the processor reacts. We may see it improve over time as the process matures, but as it stands, 95C at 1.30V is a real value - for a liquid cooler outside a case. So in a case it would perhaps be worse.Ian
bah12 - Monday, April 23, 2012 - link
Wouldn't this be expected from a pure physics perspective. Since a 22nm node has drastically shrunk the surface area in contact with the heat sink. Especially with the IGP off. The heat is now concentrated to a much smaller area, so it is all the more difficult to transfer that heat to the heat sink.MrSpadge - Monday, April 23, 2012 - link
Yes, dissipating heat from a smaller die is harder, i. e. will leave your chip hotter, given the same cooling device.vol7ron - Tuesday, April 24, 2012 - link
That's true, given the same heat; however, if you're using less current (lower voltage), the heat generated should also be less. If the current:surface area is a perfect ratio, the correlation of heat should also be close to 1. So as size decreases, so should heat.The problem, though, is that the efficiency of the transistors are questionable. Sure there may be more, but because you're creating them at a smaller scale, it could be less efficient releasing more electricity/heat to the external system.
In addition, the materials used in the tri-gate setup is unknown, which could also impact the heat retained in the system.
As Ian stated, as the manufacturing process matures, they could find ways to improve the process. This could lead to reduction in heat, and thus an increase in performance.
CeriseCogburn - Friday, May 25, 2012 - link
In a move quite uncharacteristic for me I'd just like to praise you for the very worthy and interesting article well done and setting the skewed misperceptions record straight.Very good. Much appreciated. Definitely unique and cutting edge. Worthwhile.
MrSpadge - Monday, April 23, 2012 - link
I don't thik they're equal as IB should consume considerably less power for similar clock speeds (if you stay at sane voltage levels).IanCutress - Monday, April 23, 2012 - link
To answer your original question, for these tests I was running on discrete graphics without Virtu MVP, so the iGPU should have been power gated into a low state.Ian
Kevin G - Monday, April 23, 2012 - link
Would it be possible to adjust the graphs to have similar ranges for their axis? It'd the make the two first two graphs at stock and 4.4 Ghz much easier to compare. This would also apply to the constant 1.25V graph and the underclocking graph.Also would it be able to label each point on the PovRay test with the clock speed at that particular voltage (or are these values obtained at a constant clock)?
IanCutress - Monday, April 23, 2012 - link
Apologies, I should make it more clear. The graph with a single line of Temperature against voltage should be read in conjunction with the overclock graph. So at 0.90 V, we are at 3.9 GHz, 1.1V @ 4.4 GHz, 1.15V @ 4.6, 1.20V @ 4.7, 1.30V @ 4.8. Which means that perhaps 1.15 V at 4.6 GHz is a happy medium.Ian
IanCutress - Monday, April 23, 2012 - link
*1.1V @ 4.5 GHzC'DaleRider - Monday, April 23, 2012 - link
"Ivy Bridge is now released and behaves differently to Sandy Bridge."That probably would better read with the "to" replaced with "than", a word that means a comparison. In your sentence above, Ian, to what does the phrase "to Sandy Bridge" refer? Do you mean in a comparison to Ivy Bridge, perhaps? Then, the word you used, "to", is wholly inappropriate and meaningless. Instead, that sentence would more properly read with the "to" replaced with "than", a word that means a comparison, i.e. I'd rather have a bottle in front of me THAN a frontal lobotomy.
So, it'd read, "Ivy Bridge is now released and behaves differently than Sandy Bridge." The sentence now makes the comparison you were trying to say but failed.
IanCutress - Monday, April 23, 2012 - link
I could also write "Ivy Bridge is now released and it behaves differently to Sandy Bridge." or a variety of other ways. Along with this, I have to deal with avoiding British idioms for a mainly American based audience, as well as the issue of company plurality which seems very alien to me. However, if you do find other areas of English in one of my articles that perhaps need a touch up, please be tactful and email me (ian AT anandtech DOT com) rather than habitually pick out one phrase and demonstrate your horrific disgust about it on a public forum. Writers work their fingers to the bone to get some of these articles done in time, and we would appreciate it if you didn't beat down their hard work without at least attempting to contact them first as a polite indication.Ian
anirudhs - Monday, April 23, 2012 - link
Perhaps he comes to AnandTech to find mistakes in your articles.dustofnations - Monday, April 23, 2012 - link
The form you used is fine in British English, so he is wrong and crass. I don't think you should need to kowtow and (unreasonably) modify your language just for those who speak another form of English, given that it is perfectly understandable.Some exposure to the outside world might be an educational and enlightening experience for them :-).
Spunjji - Thursday, April 26, 2012 - link
Indeed. Apparently British English is some sort of mutant, awful version of his native American.philosofool - Wednesday, May 9, 2012 - link
I had no idea that "different to" was relatively common in British English.It sounds awful to my American ear. "He behaves differently to her" would certainly not imply to me that his and her acts are dissimilar.
poohbear - Tuesday, April 24, 2012 - link
actually there's nothing wrong with using "to" instead of "than" when making comparisons. Google it to see its regular use, u can then proceed on your littl grammar nazi crusade and email the thousands of people on the internet that use "to" to make comparisons. I'm assuming ure not a native speaker as a native speaker correcting another native speaker on how to use the language correctly is laughable.I'm an English teacher and would never mark such a sentence wrong. Language develops, especially one as international as English which is used in all corners of the world and will develop little differences.
silverblue - Wednesday, April 25, 2012 - link
Exactly. Even if there was a slight deviation (not that there was in this case), what matters ultimately is that the author still communicates what it is that (s)he is trying to convey.haar - Thursday, April 26, 2012 - link
how, a could get an 'A' in your class, all for just speaking the way i want too...not the way I should... LOL
anirudhs - Monday, April 23, 2012 - link
I hope you have a bottle in front of you. Much less painless than a frontal lobotomy.Spunjji - Monday, April 23, 2012 - link
You're taking a grammatical preference and asserting it as fact. It is not fact; different to /from / than is a free option - your chosen example uses a different sentence structure.To do this in such an overtly critical and unfriendly fashion is repugnant; the fact that you are also incorrect merely serves to make you look very silly.
Arbie - Tuesday, April 24, 2012 - link
C'DaleRider doesn't mind looking stupid. He made a similarly ignorant and rude grammar comment a few weeks ago. He's just trolling, trying to disrupt the comment stream and generally mess up a good thing. That's the way some people are, sad to say.Spunjji - Thursday, April 26, 2012 - link
'Tis a shame, to be sure.LoneWolf15 - Tuesday, April 24, 2012 - link
You aren't worthy of owning a Cannondale. Please turn it in to the nearest worthy person.Spunjji - Thursday, April 26, 2012 - link
You're assuming he actually has one. I'm not sure what being a basement-dwelling troll pays these days.GaMEChld - Thursday, April 26, 2012 - link
Is that guy for real? What a douche bag. (<--- Was that gramatically correct?)Spell check this ..|..
Breach1337 - Monday, April 23, 2012 - link
Looks like a completely pointless upgrade for 'normal' overclockers who already have a 2500/2600/2700K...1ceTr0n - Monday, April 23, 2012 - link
As a i5 2500k user at 4.6ghz on stock voltage, I concurSpunjji - Thursday, April 26, 2012 - link
Yup! Nothing to see here.MrSpadge - Monday, April 23, 2012 - link
If you shrink crystalline Si dimensions, heat conductivity breaks down as less phonon modes can fit into the volume and contribute to heat conductivity. For the fin-FET like tri-gate design this should be even more severe, as we're quenching the phonon modes in 2 dimensions. Might well be that the problem is to get the heat out of the transistors in Intels 22 nm process. If this is correct, things are only going to get worse at even smaller process nodes. The problem is then not about putting heat into the system, but rather to get it conducted even to the heat spreader or back side of the chip.zipz0p - Tuesday, April 24, 2012 - link
This is an interesting point. I hadn't thought of the fin-FET in this way before. You may have just hit the nail on the head, though.silverblue - Wednesday, April 25, 2012 - link
Apologies if this is a little out of left field, but isn't this something that AMD is hoping to tackle with resonant clock mesh technology?Spunjji - Thursday, April 26, 2012 - link
Lovely interpretation of the facts. Kudos to you.ceravis - Monday, April 23, 2012 - link
Thanks Ian, this is the first I've heard of these issues... I had been holding out for Ivy Bridge results before choosing between it and Sandy Bridge (which will hopefully now drop in price).This helps much.
Arnulf - Monday, April 23, 2012 - link
"Then it will be a case of deciding whether the small IPC gains that Ivy brings will be worth 200 MHz less on your CPU compared to Ivy Bridge."Did you mean to say "... compared to Sandy Bridge." ?
klmccaughey - Monday, April 23, 2012 - link
Ian, could you confirm please whether or not the GPU was active in these tests? Someone asked the question but you maybe missed it? Thanks, KevinCeriseCogburn - Thursday, May 3, 2012 - link
It was not active and yes you missed Ian's reply noting that, as did the article, with 2x 79xx.DParadoxx - Monday, April 23, 2012 - link
I'm interested in seeing how things go if you disable HT and graphics.MrSpadge - Monday, April 23, 2012 - link
Not much change, just a little less cpu performance, power consumption and temperature.DParadoxx - Monday, April 23, 2012 - link
The difference on my i7 920 between HT-on and HT-off with respect to voltage/temperature under load is incredible. Do you have any evidence to base your statement up?PolarisOrbit - Monday, April 23, 2012 - link
Maybe I'm just a dummy, but what is the stock voltage for Ivy Bridge? Does the article even say? The closest statement I found was 4.4GHz at 1.05V which is "close to stock."I think it could have been a little bit more clear in the article.
Dracusis - Monday, April 23, 2012 - link
So most recommendations seem to be, if you have SB, don't bother with IB.But what if you're still rocking a Core2Duo, you're looking to upgrade and planning to OC?
Would it be better to pick up an older Sandy Bridge CPU chip in light of all this?
Serion - Monday, April 23, 2012 - link
Big case + strong cooling + don't mind high power usage (+more noise)= SB @ 5-5.1+GHz will be better (faster)
Small case + not-so-good cooling + power usage figures matter / low noise is important
= IB @ 4.6-ish will be the better choice
(uses less power / will be faster than SB @ 4.6-ish with the better IPC from IB)
So yea, it'll matter on how you want to run your system.
For general usage, I recomend the later.
MrSpadge - Monday, April 23, 2012 - link
Ian hit ~4.5 GHz at a mere 1.05 V with IB. That's significantly less power consumption than anything SB could do in this performance range. Top clock may be a bit lower, but you're saving tons of electricity with IB. And if you're buying new: IB should be cheaper anyway. Exception: getting a good deal on a used chip, saving 100+$, would probably be worth it.Dracusis - Monday, April 23, 2012 - link
Thanks Serion & MrSpadge.Think I'll go with IB unless local prices for early shipments down under are unnaturally high (sadly, they often are).
Zink - Monday, April 23, 2012 - link
Besides the GPU performance increase, all IB brings is reduced power. If you have a system that doesn't need a low power chip get a last gen CPU.MrSpadge - Monday, April 23, 2012 - link
Not if you're buying new (details above).nitrousoxide - Monday, April 23, 2012 - link
I can't overclock them but that's fine because all I care is a cooler and quieter system. Can I undervolt an i7 3770 too?MrSpadge - Monday, April 23, 2012 - link
That's only a matter of the motherboard for SB and IB. Ian even showed you underclocking in the article.MrSpadge - Monday, April 23, 2012 - link
Sorry, overclocking in conjunction with undervolting. Yes, the chip is that good!Iketh - Monday, April 23, 2012 - link
so it appears I'll be able to passively cool a 4.5ghz IB... silent users and sound studios should be rejoicing... can't wait to swap out my 2600kIketh - Monday, April 23, 2012 - link
well, 50% chance for 4.5 passive, and maybe 80% for 4.4Iketh - Monday, April 23, 2012 - link
Ian, was the heat @ 1.25+ volts dissipating into the heatsink the same as SnB? Did you try touching the heatsink to see if it was too hot to touch?Iketh - Monday, April 23, 2012 - link
PLEASE for the love of god use Prime95 for stability/temps.... are we forgetting that prime95 IS A REAL WORLD APPLICATION??? Plenty of people are actively searching for primes!!Instead, you are using a synthetic heat generater (OCCT) and another real-world that doesn't get as hot as prime...
haar - Thursday, April 26, 2012 - link
i agree, prime95 is a "certified" heat generator... and a D@mn fine stablity checker ( but requires alot of time)... 6 hours for 80% assured stability, 24 hours for 95% or "good-enough" (of course for "guaranteed" stability run at stock, needed for checking primes)if you need an absolute worst case power draw/heat, run prime95 on all but-one-core, and run furmark on the GPU...
Minot - Monday, April 23, 2012 - link
The sweet spot for performance vs heat seems to be at 4.5 GHz @ 1.1v. I've been reading the IB threads in the CPU OC forums on Anandtech. The few enthusiast members that have been testing 3570K haven't come close to 4.5 GHz at 1.1v. In fact, the best I've seen is 4.5 GHz at 1.236v.What settings did you use to get a stable 4.5 GHz OC at 1.1v?
Iketh - Tuesday, April 24, 2012 - link
What part of my post explaining why I can't wait don't you understand?meloz - Tuesday, April 24, 2012 - link
Well done Ian, you covered a lot of ground and all the relevant points in a nice, graphical way.If you ask me, 1.15 Volt is the 'Sweet Spot' for this processor, you can crank upto 4.6 GHz (depending on processor lottery, some variation will exist between processors and future steppings), while still keeping temperatures reasonable and not compromising long term reliability.
MFL - Tuesday, April 24, 2012 - link
Power Consumption^ 3770K
I /
I _______/___________ 2700K
I _________ / /
I____/ __/
I ________/
I_________/
I
--------------------------------------------------> Avg. Gaming Performance
MFL - Tuesday, April 24, 2012 - link
Sorry, the spaces were removed. I hope you know what i mean.Also maybe you could make the same chart with Temperature instead of Power Consumption.
Or the same with Hyper-Threading disabled.
lyeoh - Tuesday, April 24, 2012 - link
You can buy what he's smoking from AMD ;).owdi - Tuesday, April 24, 2012 - link
This article is lol-worthy as it seems the author believes there is a linear relationship between voltage and power. Your increase of .35 volts nearly DOUBLES the power consumption of the CPU.Power = voltage ^2 / resistance
(1.25v)^2 / (.9v)^2 = 193%
Ivy bridge does not hate voltage, but it does hate people who don't understand ohms law.
luke929 - Tuesday, April 24, 2012 - link
P = IV is Joule's law, V = IR is Ohm's law.so....you know Ohm's law, right?
Death666Angel - Thursday, April 26, 2012 - link
Not sure where you read that the author states a linear relationship between voltage and power consumption.b3nzint - Wednesday, April 25, 2012 - link
wow, now intel knows that a decent performing IGP is so heat sensistif. I'll keep my 2600K.zodiacfml - Wednesday, April 25, 2012 - link
...it is because the base clock speed is pretty high to start with for this 22nm process.philosofa - Wednesday, April 25, 2012 - link
Absolutely love the methodical nature of your investigations and your understanding of the subject matter Dr. Cutress. With it and the various ways you've illustrated things, it's givenme in a very short time what seems to be a good understanding of the basics of how Ivy and the current 22nm Tri-Gate processors behave, and how to best overclock them.No mean feat considering the differences. Thanks very, very much :)
Marburg U - Wednesday, April 25, 2012 - link
maybe they will need a different kind of package in the future.haar - Thursday, April 26, 2012 - link
same thing happems with bulldozer (8150). increase the voltage and the temps get out of hand quickly...haar - Thursday, April 26, 2012 - link
first pick a clockrate... then see what is the minimum voltage required for the clock rate...BlindFreddie - Friday, April 27, 2012 - link
http://www.techpowerup.com/164858/Ivy-Bridge-Tempe...Everything I've read about high temperatures on IB suggests that there is a large amount of thermal resistance between the active parts of the IB chip and the outside surface of the integrated heat spreader (IHS) to which we attach our heatsinks.
It is suggested that the IB heatspreader is not soldered onto the chip, but uses thermal paste of unknown quality. That would go a long way to explaining the temperature problems found when overclocking IB.
Of course there are other factors such as the increased power per unit area of the chip, and heat may be trapped by the 3-dimensional nature of the new transistors.
If you can do so, please pop the lid off your press-release ES CPU and re-run some of the tests with your heatsink mounted directly on the CPU die with top-quality TIM paste. What will happen to the CPU otherwise? If it has to go back to Intel, so what? They gave it to you to test. Well, you'll have really tested it :)
EJ257 - Tuesday, May 1, 2012 - link
How hard would it be to remove the IHS and just connect the heatsink directly to the chip (like in the old days)? Keep in mind the physical size of the die is smaller now so you'll have less surface area to work with. Add in the new tri-gate process and how well that will dissipate heat and you have a heat-gate (tm, hehe) trifecta.BlindFreddie - Friday, April 27, 2012 - link
The original article on which theTPU one linked above is based:http://www.overclockers.com/ivy-bridge-temperature...
They do wonder whether the retail CPUs will have soldered IHSs instead of the TI paste used on the ESs.
Thread at xtremesystems.org that discusses this article:
http://www.xtremesystems.org/forums/showthread.php...
And I forgot to say thanks to Ian Cutress for writing this article. The charts setting out the volts/temps/max clocks envelope of the CPU are excellent.
swing848 - Monday, April 30, 2012 - link
The AMD HD 7970 is the first video card to use PCIe 3.0 bandwidth, and that is a single card. So, PCIe 3.0 on new motherboards is good to see.However, games that use the CPU heavily and high bandwidth GPUs that tax CPUs beyond their limit [if not overclocked] is NOT a good thing.
Intel needs to introduce a better heat disipation method for the latest generation of CPUs that allow core clock changes [K series, for example], to allow higher overclocks at reduced temperatures and voltage.
Why have a motherboard that is PCIe 3.0 capable if the CPU is over-taxed and bottlenecks [game] video?
Intel needs to fix this problem quickly or they will be selling fewer of the latest K series CPUs.
muy - Tuesday, May 1, 2012 - link
i think after amd, now intel crashed headlong into the 'wall' of Moore. the amount of ipc increase we might still see over the next 10 years might be much smaller than most people realize. the amount of clock increase we might see over the next 10 years might be much smaller than most people realize. ok, they can still add more cores, but for 9 of 10 applications, 10 % ipc/clock increase will do more than doubling the number of cores.implications of moore's law already failing this bad in 2011-2012, which is much earlier than anticipated could be 'dramatic' .
Snorren - Saturday, May 5, 2012 - link
My 3770K needs 1,015 V @ 3,9Ghz on a MAXIMUS V GENE BIOS 0701Is HT disabled in this test or is this a golden sample? Or am I missing something :)
philosofool - Tuesday, May 8, 2012 - link
"Ivy Bridge behaves differently *than* Sandy Bridge [behaves]" I don't know why you keep putting "to" in place of "than".You may also write "Ivy Bridge over clocking is different from Sandy Bridge".
In any event, "different to" has no meaning in standard English. You are not a commenter on a blog, you are professional writer. Some errors I can tollerate, but this is just awful to my ear.
Kepe - Tuesday, May 8, 2012 - link
This was already discussed in length in the comments. "Different to" is British English and grammatically perfectly fine.lucky9 - Wednesday, May 9, 2012 - link
Just a matter of time. Cool those little pins right down. SB is looking pretty good ATM.On the other popular subject, you can compare something TO something else in the English language.
Glorfin - Monday, May 21, 2012 - link
Would be better for new builders to buy a 2500k instead? It seems like there's a lot more risk involved with Ivy Bridge, esp. for people new to overclocking (like myself). Not really sure what to do from this article.Complete noob here on a first build, hoping not to screw up. ^_^
Youssef 2010 - Monday, June 4, 2012 - link
Quoting Tom's Hardware Sandy Bridge overclocking article:"Increased density and new transistor technology can certainly give rise to more heat. But we've seen these things before, and each time, cooling and packaging technology manage to cope with the resulting output. What could be holding Ivy Bridge back? It could be the fact that Intel decided to use thermal paste instead of the usual fluxless solder (you'll need to have a solid grasp of Japanese to understand the original article, or use Google Translate) between the CPU die and heat spreader.
Using a box knife (we do not recommend doing this at home), the author of the linked story pried the heat spreader away from the chip and was able to replace the cheap paste Intel used, trying out both OCZ Freeze Extreme and Coollaboratory Liquid Pro thermal pastes. The OCZ offering allowed for 1.55 V at 4.9 GHz, while the Coollaboratory material ramped up to 5.0 GHz, operating stably. This was accomplished even with air cooling, although the author did not use a stock heat sink, opting for a Thermalright Silver Arrow SB-E instead (Ed.: this information was pulled from the original Impress PC Watch site with Google's rather shaky translation of Japanese). If there is a smoking gun in this equation, we think this is it, especially considering that the researchers at Impress PC Watch managed 20% more efficient cooling."
For the full article: http://www.tomshardware.com/reviews/ivy-bridge-ove...
sunzt - Tuesday, August 28, 2012 - link
What did you use to undervolt the CPU?PHerrmann - Friday, March 13, 2015 - link
Hi,Thank you for this article.
Using the Max. Overclock vs. CPU Voltage graph, I figured out that my CPU was less overclockable than yours, but I reached easily 4.6 GHz @ 1.25 V. At the same frequency, le processor was unstable @ 1.20 V.
System : i7 3770k, GA-Z77-D3H (BIOS F18), 2x2 GB Kingston 1600 DDR3.
Idle : 33.5 °C, 80 W Load : 72 °C, 175 W
These settings are used on a 24/7 basis, mainly for images and video processing (Win 8.1)