Ivy Bridge Versus Sandy Bridge: Temperature Analysis

[Concillian]

Less heat / less power consumed than SB at any given temperature.

Nobody knows where the temp sensor is in relation to where it is in SB, but throttling appears to happen at the same temperature as it did with SB.

Anyone going for less than ultimate / extreme overclocks will see less power usage with IVB than with SB for a given performance level. This given performance level will happen at a slightly lower clock due to IVB having ~5% extra performance at a given clock speed. 5% is about 200 MHz difference... so a 4.3 IVB is about equal to a 4.5 SB.

So basically for 4.3-4.5 GHz normal man type OCs, IVB will be superior.
For power consumption be damned 5+GHz OCs SB seems better.

In my case, I OC conservatively, and if I owned an SB, I probably wouldn't have run it more than 4.3 - 4.4. In this case an IVB is a clear winner. For those who have or would have pushed an SB to 4.8GHz+, IVB is clearly not a winner.

I primarily game, and have found very few cases where my 4GHz i3-530 (2 core + Hyperthreading) was inadequate. Even BF3 seems to do okay (not the best, but okay) with a dual + HT at those clock speeds. I don't expect IVB at 4.3-4.5GHz to limit me in ANY way. It will have about +40% single threaded performance over my 4 GHz clarkdale, which is a pretty significant improvement. I'm hoping around the same power consumption as my 2C / 4T for a 4C / 4T CPU. In things that I do other than gaming, I'm generally fine with my current performance level, and IVB will offer significant performance improvement just from the IPC improvement. The extra cores is like super bonus. I really think I'd be happy with an OCed 2C / 4T IVB, but Intel doesn't want me to have one.

 

[Tempered81]

In my case, I OC conservatively, and if I owned an SB, I probably wouldn't have run it more than 4.3 - 4.4. In this case an IVB is a clear winner. For those who have or would have pushed an SB to 4.8GHz+, IVB is clearly not a winner.

This is the best summary of what is going on with Ivy Overclocking. But it was not supposed to be this way. If anything, Ivy was expected to run COOLER with MORE headroom, not hotter with less. The fact that there is a 200-300mhz difference in performance in those ranges, and a 200-300mhz difference in maximum overclocking capacity makes their maximum overclocking potential equal. Whats the point of having two high-clocking "K" series top-end CPUs that have equal potential (excluding pcie 3.0, usb3, light peak, blah blah).

It's no faildozer, but still a fail.

 

[prtuc2]

Is it due to the higher performance on chip graphic that escalate the temperature vs the sandy bridge counterpart?

 

[Don Karnage]

Is it due to the higher performance on chip graphic that escalate the temperature vs the sandy bridge counterpart?

Has Intel stated that the temp sensor is in the exact same place that Sandy Bridge's was?

Also

 

[Tempered81]

^ Yes Don is right, that is the #1 reason. #2 is the die surface area : TDP.

 

[BD231]

1 Minute into Prime. Water/EK High-Flow/XTX 240 Rad

http://www.overclock.net/t/1242313/...sandybridge-comparison-3770k-on-the-way-d/440

Yikes!

 

[Tempered81]

ehh looks like tjmax is 105 instead of 110

 

[MPiland]

If anything, Ivy was expected to run COOLER with MORE headroom, not hotter with less.

Where did people get this idea from? Did Intel say this? Because honestly, if they would've used simple physics and some common knowledge, they would've realized this was never the case. Think about it: Yes, less power is being consumed so theoretically less heat would be produced, but now that heat is in a smaller area with less room to escape. And on top of that, the transistors now move electricity differently. Unless Intel said that this was going to run cooler and have more headroom, then this is the people's fault for running with this speculation.

Whats the point of having two high-clocking "K" series top-end CPUs that have equal potential (excluding pcie 3.0, usb3, light peak, blah blah).

That last part says it...plus the new transistors (new technology to move forward with) in a smaller die.

 

[Tempered81]

Where do people get that idea from? People got that idea from the last 9 architectures

from anandtech's ivy bridge overclocking article:

http://www.anandtech.com/show/5763/undervolting-and-overclocking-on-ivy-bridge

 

[dinker99]

I`d be more impressed if someone put one in a calorimeter and measured the actual heat output rather than relying on Core temp.

 

[cytg111]

This is the best summary of what is going on with Ivy Overclocking. But it was not supposed to be this way. If anything, Ivy was expected to run COOLER with MORE headroom, not hotter with less. The fact that there is a 200-300mhz difference in performance in those ranges, and a 200-300mhz difference in maximum overclocking capacity makes their maximum overclocking potential equal. Whats the point of having two high-clocking "K" series top-end CPUs that have equal potential (excluding pcie 3.0, usb3, light peak, blah blah).

It's no faildozer, but still a fail.

This has problary been mentioned a 1000 times, but heat<>temp.
Your chip can dissapate half the watts just fine but if the die size is 1mm^2, that chip is going into fusion temperature range fast.

 

[MPiland]

Where do people get that idea from? People got that idea from the last 9 architectures

from anandtech's ivy bridge overclocking article:

http://www.anandtech.com/show/5763/undervolting-and-overclocking-on-ivy-bridge

I'm still not getting where this shows less heat and more headroom? This just shows that it puts out more heat...So again, where did people get the idea that it was supposed to put out less heat? I think it was all speculation and self-hype. I get that the last 9 architectures did this, but that doesn't make it a general rule. If those last 9 were the rules, this was the exception.

 

[Smoblikat]

I'm still not getting where this shows less heat and more headroom? This just shows that it puts out more heat...So again, where did people get the idea that it was supposed to put out less heat? I think it was all speculation and self-hype. I get that the last 9 architectures did this, but that doesn't make it a general rule. If those last 9 were the rules, this was the exception.

I understand what youre trying to say, but again...........HEAT ISNT TEMPERATURE. It puts out less heat, but it has a higher temperature.

 

[Tempered81]

I'm still not getting where this shows less heat and more headroom?

You're saying it backwards. The correct way is 'more' heat and 'less' headroom. The comparison is in overvolting & overclocking against its predecessor Sandy Bridge (2500k/2600k/2700k). The Sandy Bridges clock higher (more headroom) and run cooler (less heat). The Sandies are slower at instructions per clock (4-8%). The result equals 'not a meaningful upgrade' but more like a 'side-grade'.

This is definitely the exception to the rule. We were spoiled by Sandy Bridge evidently.

 

[MPiland]

I understand what youre trying to say, but again...........HEAT ISNT TEMPERATURE. It puts out less heat, but it has a higher temperature.

Ture, I should've said runs hotter rather than puts out more heat. But, the point remains nonetheless. It is a smaller area for the heat to dissipate so it builds faster leading to higher temps even though the heat output is less. But, people ran wild with assumptions and became disappointed when those assumptions turned out to be wrong. Personally, I see it as a good chip and a step in the right direction. Lower production costs, new technology, higher IPC albeit only marginally.

 

[AtenRa]

I'm still not getting where this shows less heat and more headroom? This just shows that it puts out more heat...So again, where did people get the idea that it was supposed to put out less heat? I think it was all speculation and self-hype. I get that the last 9 architectures did this, but that doesn't make it a general rule. If those last 9 were the rules, this was the exception.

The IvyBridge IC put out less heat because it uses less power. The die itself operates at higher temperatures (vs SB) because of the inability of the cooling solution.

 

[MPiland]

You're saying it backwards. The correct way is 'more' heat and 'less' headroom. The comparison is in overvolting & overclocking against its predecessor Sandy Bridge (2500k/2600k/2700k). The Sandy Bridges clock higher (more headroom) and run cooler (less heat). The Sandies are slower at instructions per clock (4-8%). The result equals 'not a meaningful upgrade' but more like a 'side-grade'

No, I said it right. You said it was expected to run cooler and give more headroom. I asked where you got that idea and you posted a graph showing the opposite.

The IvyBridge IC put out less heat because it uses less power. The die itself operates at higher temperatures (vs SB) because of the inability of the cooling solution.

I know, I said it wrong. My bad. I meant runs hotter (even though it puts out less heat).

 

[blckgrffn]

I know, I said it wrong. My bad. I meant runs hotter (even though it puts out less heat).

Hah, yeah. With terms like that, it gets easy to mess up a sentence.

It runs at a higher temperature while having lower thermal energy output? Less ambiguous, I suppose... but still not an ideal wait to put it, I am sure.

Like BFG said in that other thread, if that higher temp is going to be spinning fans up more aggressively (and what will that really help? It's likely your big tower HS is going to be pretty cool, its just getting that energy transferred that is the issue... I suppose you keep the delta higher to better facilitate the transfer? Anyway...) that is going to be annoying.

 

[MPiland]

Hah, yeah. With terms like that, it gets easy to mess up a sentence.

It runs at a higher temperature while having lower thermal energy output? Less ambiguous, I suppose... but still not an ideal wait to put it, I am sure.

Like BFG said in that other thread, if that higher temp is going to be spinning fans up more aggressively (and what will that really help? It's likely your big tower HS is going to be pretty cool, its just getting that energy transferred that is the issue... I suppose you keep the delta higher to better facilitate the transfer? Anyway...) that is going to be annoying.

LOL yup. This is why I'm glad I went with a H100 instead of the Noctua D-14 or similar. I knew I wouldn't want some big fan on my CPU, but for a completely different reason - aesthetics lol I like the H100 look better, but my case is designed to use it (Corsair 500r).

 

[Tempered81]

Where did that idea come from? Well, I myself, like many others, were hoping that the new technology would grant us faster speeds. Node transitions historically result in a higher clock. Did you know that cpus from the past ran not in GHZ, but in MHZ? It's true. Dies can be larger or smaller, and as with gpu dies the smaller ones usually clock better, but the clock ceilings are really relative in the cpu world and dependent on other stuff. Not sure how long you've been following cpu overclocking, but if you look back on our history, each time the lithography process shrinks to a new node, or intel goes with a revision or new stepping, temperatures go down along with voltage requirements while efficiency metrics and overclocking ceilings rise. CPU makers try to meet the same profile (TDP), the end result is: the cpu product in your category of interest usually ships with a higher default frequency. 1ghz, new node, 2ghz, new node, 3ghz, new node, rinse, repeat. Etc, etc. We're now at ~4ghz out of the box on these mainstream SKUs. Remember the G0 Kentsfield? Well, there are a few E0 and E1 Ivys out there, and E1 looks worse (debatable) which is flowing against the grain. Did you know that Ivy Bridge showcases a revolutionary change in semiconductor fabrication? Yes this is true also. Fabs quit measuring trace width in microns and now measure density in nanometers. Someone will be able to explain it much clearer to you than I can. Now instead of using the flat, tradition two dimensional planar transistor gate, Intel's new 22nm process uses a 'never before seen' fin-fet 3D gating transistor developed because of difficulties faced shrinking beyond 32 and 22nm nodes; future transitions will also use the new technology. Should they iron out the kinks, hopefully a new revision will give better results.

3570K/3770K run fine at stock, and run great at up to 4500-4600mhz, and usually within a Vcc range of 1100-1250mv again depending on the cooling. There is a strange curve that you can see in the Anandtech graph around the 4600mhz point, that coincides with a 'skyrocketing' temperature increase. Different sources show the increase happening around the 1350mv range. You're looking at temperature based limitations occurring at either the 4600-4700mhz range or the 1300-1350mv range (or both) that affect overclocking potential with mainstream water and high-end air cooling. Ivy may have a 64x multi, and with no coldbug on Ln2, the sky's the limit, however the average Joe like myself will have a tough time with 1.4v/4.8ghz looking for stability without exotic cooling.

If you have no desire for a beastly 5ghz overclock it's a fine chip. If you've got a helluva chilled liquid loop, you can likely still get five Gs. The moral of the story is they aren't clocking as high as Sandy, and they run hotter. New semiconductor technology does not usually perform worse, rather it does the opposite and runs cooler and switches faster. Regarding die sizes, look at it this way: A computer chip has been relatively the same size as a postage stamp for 4 decades, and all the while silicon operating temperature thresholds have not changed, while the speed and performance has increase a million fold. I'm not a fan of E1 Ivy Bridges, and I'm not a fan of Bulldozer. The folks with 4.8-5.1ghz 3960X's and 2700K's are sitting like fat cats right now. They know a 4.7ghz 3770K will give them a run for their money, but that contest will be almost equal within a ~1% margin of error. The expectation may have been 10% faster IPC combined with 10-15% higher overclocking ceilings; at least for those of us not reading up for the last 3 years.

 

[MPiland]

Where did that idea come from? Well, I myself, like many others, were hoping that the new technology would grant us faster speeds. Node transitions historically result in a higher clock. Did you know that cpus from the past ran not in GHZ, but in MHZ? It's true. Dies can be larger or smaller, and as with gpu dies the smaller ones usually clock better, but the clock ceilings are really relative in the cpu world and dependent on other stuff. Not sure how long you've been following cpu overclocking, but if you look back on our history, each time the lithography process shrinks to a new node, or intel goes with a revision or new stepping, temperatures go down along with voltage requirements while efficiency metrics and overclocking ceilings rise. CPU makers try to meet the same profile (TDP), the end result is: the cpu product in your category of interest usually ships with a higher default frequency. 1ghz, new node, 2ghz, new node, 3ghz, new node, rinse, repeat. Etc, etc. We're now at ~4ghz out of the box on these mainstream SKUs. Remember the G0 Kentsfield? Well, there are a few E0 and E1 Ivys out there, and E1 looks worse (debatable) which is flowing against the grain. Did you know that Ivy Bridge showcases a revolutionary change in semiconductor fabrication? Yes this is true also. Fabs quit measuring trace width in microns and now measure density in nanometers. Someone will be able to explain it much clearer to you than I can. Now instead of using the flat, tradition two dimensional planar transistor gate, Intel's new 22nm process uses a 'never before seen' fin-fet 3D gating transistor developed because of difficulties faced shrinking beyond 32 and 22nm nodes; future transitions will also use the new technology. Should they iron out the kinks, hopefully a new revision will give better results.

3570K/3770K run fine at stock, and run great at up to 4500-4600mhz, and usually within a Vcc range of 1100-1250mv again depending on the cooling. There is a strange curve that you can see in the Anandtech graph around the 4600mhz point, that coincides with a 'skyrocketing' temperature increase. Different sources show the increase happening around the 1350mv range. You're looking at temperature based limitations occurring at either the 4600-4700mhz range or the 1300-1350mv range (or both) that affect overclocking potential with mainstream water and high-end air cooling. Ivy may have a 64x multi, and with no coldbug on Ln2, the sky's the limit, however the average Joe like myself will have a tough time with 1.4v/4.8ghz looking for stability without exotic cooling.

If you have no desire for a beastly 5ghz overclock it's a fine chip. If you've got a helluva chilled liquid loop, you can likely still get five Gs. The moral of the story is they aren't clocking as high as Sandy, and they run hotter. New semiconductor technology does not usually perform worse, rather it does the opposite and runs cooler and switches faster. Regarding die sizes, look at it this way: A computer chip has been relatively the same size as a postage stamp for 4 decades, and all the while silicon operating temperature thresholds have not changed, while the speed and performance has increase a million fold. I'm not a fan of E1 Ivy Bridges, and I'm not a fan of Bulldozer. The folks with 4.8-5.1ghz 3960X's and 2700K's are sitting like fat cats right now. They know a 4.7ghz 3770K will give them a run for their money, but that contest will be almost equal within a ~1% margin of error. The expectation may have been 10% faster IPC combined with 10-15% higher overclocking ceilings; at least for those of us not reading up for the last 3 years.

LOL yes I know the history of PCs. I've played with them for the last 20 years. But, I've come to learn to not expect anything and to actually wait for reviews on the new tech to see how it ACTUALLY performs.

As others have stated, the 4.5-4.7 range for IB is equal to the 4.8-5.0 range of Sandy. So yes, the numbers themselves aren't equal, but the power relatively is. People get too caught up in the numbers and not how the chip actually performs.

The semiconductors aren't performing worse this time around. They're running more efficiently using less power which was the whole point. They didn't say it would disperse heat in the same manner. What they said was, a reduction in power usage while having an increase in performance and that's exactly what IB does. Put a 4.5 SB against a 4.5 IB and the IB holds better numbers, yes only slightly, but regardless they are still better while using less power. Intel delivered on what they said they would. Everyone else just assumed the rest.

 

[Tempered81]

LOL yes I know the history of PCs. I've played with them for the last 20 years. But, I've come to learn to not expect anything and to actually wait for reviews on the new tech to see how it ACTUALLY performs.

As others have stated, the 4.5-4.7 range for IB is equal to the 4.8-5.0 range of Sandy. So yes, the numbers themselves aren't equal, but the power relatively is. People get too caught up in the numbers and not how the chip actually performs.

The semiconductors aren't performing worse this time around. They're running more efficiently using less power which was the whole point. They didn't say it would disperse heat in the same manner. What they said was, a reduction in power usage while having an increase in performance and that's exactly what IB does. Put a 4.5 SB against a 4.5 IB and the IB holds better numbers, yes only slightly, but regardless they are still better while using less power. Intel delivered on what they said they would. Everyone else just assumed the rest.

Excellent points my friend, and I agree. Now I wonder what are the benefits of a 3770K versus a 2700K. Since we're discussing the new chips in the "CPUs and Overclocking" forum in the "Ivy versus Sandy Temp" thread, we must dutifully compare the two architectures in an overclocking and temperature contest. It will be a very close race, and tough to call a winner between the two chips! You make good points: Costs are down, computations are more efficient, power usage is down, there is an increase in performance.

 

[MPiland]

Excellent points my friend, and I agree. Now I wonder what are the benefits of a 3770K versus a 2700K. Since we're discussing the new chips in the "CPUs and Overclocking" forum in the "Ivy versus Sandy Temp" thread, we must dutifully compare the two architectures in an overclocking and temperature contest. It will be a very close race, and tough to call a winner between the two chips! You make good points: Costs are down, computations are more efficient, power usage is down, there is an increase in performance.

Yeah, I wouldn't say this is a great improvement over SB, but it's a step in the right direction by creating new tech. Now, if they could find a way to keep the temps down (maybe a new material for dispersing heat?) then we'd be great. Let's hope they figure it out before Haswell comes around!

Honestly, I don't think there's a "winner" between either chip. I think IB was more of a "let's create new tech and see how it performs" type of situation. And, I don't think Intel designs their chips with major OCing involved so the temps they reach aren't nearly what we'll reach. Maybe this will give them an eye opener and they'll keep these issues in mind with Haswell.

 

[Tempered81]

http://www.overclockers.com/ivy-bridge-temperatures

 

[Concillian]

Very interesting. If this is true, then de-lidding should see considerable improvement in temps. I recall de-lidding S939 Opterons back in the day yielding pretty significant temperature reductions (10C or so).

However I think the way Intel does their sockets, the hinge / lever might be in the way if you de-lid. Your HSF block and / or retention might interfere with the reduced height.