Why isn't Intel fixing the 'delidding' issue?

[GWestphal]

Sandy, Ivy and Haswell have all seen huge thermal improvements from delidding. Why is Intel not fixing this at the manufacturing level? By tightening up the tolerance on the the IHS and putting on a higher quality thermal compound in an appropriate quantity, they can get 10-20C drops in temps. So why aren't they correcting this? It seems like it would be money saving all around, less material in the IHS, less compound used, better thermal performance...so why not do it?

Edit: I'm sure they could figure out an easy way to 'spray" an appropriately thin layer of compound on if the mass production/ automation part is the issue?

 

[JimmiG]

It's simple. The TIM is good enough at stock speeds. That's really all that matters for Intel. As long as people don't start returning non-overclocked CPUs because they overheat, they're not broken and don't need fixing.

 

[Mushkins]

Because it's not an issue. The standard TIM is absolutely fine for stock, and even moderate to high OC. You can easily push an IB 3550K to 4.3-4.5 with a decent cooler and the standard TIM depending on the chip lottery.

A more efficient TIM solution would cost Intel a significant amount of money while providing negligible to nonexistent benefits to all but the 1% of enthusiasts. For example, with a Corsair H80 and a 3550k, I bumped it up to 4.0 and can run prime95 all day without going past 64C. I could easily push it further. TJMax is what, around 105C on these chips? Technically I still have 40C of headroom without delidding.

 

[GWestphal]

for the minimal amount of engineering/legwork it would take to drop 10-20C it seems short sighted on there part not to fix it. Lowering your thermal envelope by 20-30% for fractions of a penny, not doing it seems silly. We're talking 10-20C at stock clock here. That is good for everyone.

 

[Atreidin]

It would be nice to see some large system integrators lean on Intel to fix it, but I doubt that will happen. What are they going to do, threaten to stop buying from them?

 

[PliotronX]

Because it's not an issue. The standard TIM is absolutely fine for stock, and even moderate to high OC. You can easily push an IB 3550K to 4.3-4.5 with a decent cooler and the standard TIM depending on the chip lottery.

A more efficient TIM solution would cost Intel a significant amount of money while providing negligible to nonexistent benefits to all but the 1% of enthusiasts. For example, with a Corsair H80 and a 3550k, I bumped it up to 4.0 and can run prime95 all day without going past 64C. I could easily push it further. TJMax is what, around 105C on these chips? Technically I still have 40C of headroom without delidding.

Although every 10 degree celsius drop doubles the life of an IC.

 

[sm625]

What is this 3550K?

 

[Keysplayr]

for the minimal amount of engineering/legwork it would take to drop 10-20C it seems short sighted on there part not to fix it. Lowering your thermal envelope by 20-30% for fractions of a penny, not doing it seems silly. We're talking 10-20C at stock clock here. That is good for everyone.

I don't understand. Are you speaking as an engineer, or a consumer that overclocks?

It makes no difference if Intel offers it's own solution. If they make it run cooler by default, overclockers will only go higher with voltage until they reach the very same thermal barrier that before delidding, only with more of a chance of failure and accelerated electro-migration.

For example, if I buy an Ivy (whichever one), put an aftermarket cooler on it and overclock it to 4.2GHz and that is my comfortable limit due to temps, what is my next move? That is easy. De-lid. Now I have lowered my temps considerably. So now what do I do? I increase voltage and clocks until I reach my same thermal limit as I did before or until I get artifacts.

Bottom line is, there is no issue. Only according to an overclocker there is an issue. Not for the average overclocker or regular end user.

 

[SunnyD]

Why? $

 

[GreenChile]

The IHS issue started with Ivy. Sandy's IHS was soldered.

But to answer your question. Nobody here knows for sure but it could be engineered that way to address a thermal expansion problem Intel identified during development. Or maybe it is cost. Whatever the case, Intel has its reasons and we will only be speculating.

Also, the thermal improvements are really only seen when overclocking. At stock there isn't much benefit.

 

[Rakehellion]

The TIM is good enough at stock speeds.

Intel dictates stock speeds. With better cooling, they could make the stock speeds higher.

 

[kimmel]

Intel dictates stock speeds. With better cooling, they could make the stock speeds higher.

They could, if they thought people would pay more money for that few extra megahertz than the price differential in solder vs TIM.

There is no 'delidding' issue. Delidding is the solution for people unhappy with the amount of TIM under their heat spreader.

 

[TerryMathews]

Intel dictates stock speeds. With better cooling, they could make the stock speeds higher.

You're assuming that is an issue. It is likely a feature.

 

[Mushkins]

Although every 10 degree celsius drop doubles the life of an IC.

And with the cheap TIM and the gap it's still rated to operate much longer than the time it will be a feasibly useful and relevant piece of equipment.

Really, it's a non issue. Pretty sure Intel's engineering team knows how to design a processor better than anyone on this forum. They did it for a reason.

 

[ghost03]

It's simple.

1.) They don't need slightly faster chips to be competitive.
2.) Retooling their semiconductor packaging operations would cost FAR more than savings from using extra material.
3.) Instead of spending that money on retooling, they can spend it on, e.g., developing a faster GPU, something they DO need to be competitive.

In big corporations, "enhancing shareholder value" is number 1. Unfortunately, they far too big to care too much about reducing an inefficiency that only enthusiasts are aware of.

 

[calc2]

It's simple. The TIM is good enough at stock speeds. That's really all that matters for Intel. As long as people don't start returning non-overclocked CPUs because they overheat, they're not broken and don't need fixing.

Actually its not good enough at stock speeds. Not even with a good aftermarket cooler. When running Linpack 11 (AVX2) with a Noctua NH-U12P SE2 my 4770K throttles at stock settings, due to the automatic overvolting the chips do.

Yes, Linpack is an artificial benchmark but once apps start using AVX2 more this will be a legitimate issue.

Perhaps they are intending to address some of this issue with the haswell refresh?

 

[Phynaz]

Your cooling is borked if you are throttling at stock clocks.

 

[Durp]

It's done on purpose to get more enthusiasts off of the mainstream platform and onto their much more expensive high end platform where all processors use solder.

They have the ability to do this because competition is lacking.

 

[JimmiG]

for the minimal amount of engineering/legwork it would take to drop 10-20C it seems short sighted on there part not to fix it. Lowering your thermal envelope by 20-30% for fractions of a penny, not doing it seems silly. We're talking 10-20C at stock clock here. That is good for everyone.

By having loose tolerances in the manufacturing process, costs are lowered. When I delidded my 4770K, I could see how the stock TIM was applied. They put a gob on the inside of the heat spreader and then plonk it down. The TIM didn't "stick" to the die, which was almost clean when I lifted the IHS off. So it's not even being properly applied.

This is sloppy, but very quick and easy, so the child laborers in third-world countries can assemble as many CPUs as possible in the shortest time possible. This is what saves money for Intel. They are simply taking advantage of the fact that their chips can take up to 105C, reducing the need to carefully apply TIM or use more expensive solder.

 

[Kenmitch]

Your cooling is borked if you are throttling at stock clocks.

Try Linpack 11 with stock cooler and stock settings. It's temps are mind blowing.

 

[calc2]

Try Linpack 11 with stock cooler and stock settings. It's temps are mind blowing.

This also only affects Haswell but that should be obvious from my previous post since SB/IB don't support AVX2 and I mentioned it was on the 4770K. Linpack 10.3 temps are around 20C cooler vs Linpack 11 on the same Haswell CPU. I'm also using custom problem sizes to maximize memory usage, 60000/60000. It doesn't throttle at the small problem sizes some people use, eg 5000/5000.

Another point is that between Linpack 10.3 and Linpack 11 GFlops nearly double on Haswell (in the neighborhood of 120 vs 230)

I also doubt my cooler is having any real problems as the wall power usage, with no other cards in the system using IGP, is well over 200W when it starts throttling.

 

[dmens]

so the child laborers in third-world countries can assemble as many CPUs as possible in the shortest time possible.

Come on, that's low.

Also these chips are guaranteed for 7 years at 110C. You want more speed? Feel free to do the legwork yourself.

 

[fixbsod]

This is very reminiscient of the Northwood --> Prescott fiasco back in the Netburst era. Prescott was a crap chip, IB/Haswell isn't, it's just a manufacturing issue. Get the feeling intel is PURPOSELY making a hot chip as they would rather push enthusiasts to the -E line or make them delid themselves.

 

[PliotronX]

This is very reminiscient of the Northwood --> Prescott fiasco back in the Netburst era. Prescott was a crap chip, IB/Haswell isn't, it's just a manufacturing issue. Get the feeling intel is PURPOSELY making a hot chip as they would rather push enthusiasts to the -E line or make them delid themselves.

Best option right there, these noobshields even with solder interface just inhibit heat dissipation. I miss the days of Coppermines and AthlonXP chips.

 

[kimmel]

Your cooling is borked if you are throttling at stock clocks.

That's what ARM does as standard practice. :whiste:

They have the ability to do this because competition is lacking.

No, they do it because no one is going to pay them in commensurately higher selling prices if they use a more expensive packaging method.