Why do most desktop x64 CPUs have lids these days?

[mikeymikec]

I'll start with what may be my misunderstanding as to why CPU lids became a thing in the first place: As CPUs started to draw and waste more power, heftier coolers were required, and heftier coolers generally meant heavier coolers that have to stay attached to the CPU with increasingly elaborate mounting structures, and exerting as much pressure as safely possible to ensure maximum conductivity. The greater the pressure that's exerted, the more likely it is that mistakes are going to be made when attaching or detaching the HSF, and the lid is designed to protect the die.

The lids are also called heat spreaders, and I'm coming to the crux of my question now: To save money at the expense of conductivity, the lid is no longer soldered to the heat spreader; is there still a lot of point in having a heat spreader in most cases? When I say "most cases", I mean the vast majority of users out there who aren't having the highest end and/or overclocked CPUs. Also, given that most modern desktop CPUs do not waste anywhere as much energy as their predecessors (mainly evidenced IMO by the ever-shrinking stock HSFs), so CPU lids/heat spreaders still serve much of a purpose these days?

 

[William Gaatjes]

For as far as i know, AMD lids are almost all soldered except the raven ridge apu's. Intel lids are not soldered but use TIM between the die and lid. And it is as you write, the lid is to protect the die. To prevent the corners to chip off when a heatsink is applied.

 

[Kenmitch]

The shift to heat spreader was to protect the silicon from the technically challenged end users. Intels shift to TIM instead of solder was most likely not to save on manufacturing cost. They probably did some internal testing and figured out if they thermally limited overclocking by doing so it would save them a lot of money on the rma front. Delid for better thermals and overclock also gets a warranty is void which in the end makes Intel happy.

 

[Hail The Brain Slug]

For as far as i know, AMD lids are all soldered. Intel lids are not soldered but use TIM between the die and lid. And it is as you write, the lid is to protect the die. To prevent the corners to chip off when a heatsink is applied.

The Raven Ridge APUs that came out recently have TIM, so we might start seeing it more and more in future releases from AMD.

The shift to heat spreader was to protect the silicon from the technically challenged end users. Intels shift to TIM instead of solder was most likely not to save on manufacturing cost. They probably did some internal testing and figured out if they thermally limited overclocking by doing so it would save them a lot of money on the rma front. Delid for better thermals and overclock also gets a warranty is void which in the end makes Intel happy.

Theres a whitepaper that can be found about microcracking in solder between a die and an IHS. It shows that with decreasing die size, microcracking becomes more and more prevalent. This is probably why Intel switched to TIM on the ever shrinking consumer dies, but doesnt explain the switch on Skylake-X/SP. That switch was probably motivated by cost savings since they are now able to avoid soldering in the production process of all their products.

 

[chrisjames61]

The Raven Ridge APUs that came out recently have TIM, so we might start seeing it more and more in future releases from AMD.

The APUs are considered "budget" parts. They used TIM to keep the price point down.

 

[William Gaatjes]

It makes sense that it is a cost reducing measure.
The process of soldering a lid to a die and taking into account the prevention of microcracking, is much more intensive manufacturing wise than applying a TIM patch and glueing the lid on.
The latter option allows for much more tolerance which means more speed and less accurate placement needed. Less time spend per cpu is more profit.

 

[Hail The Brain Slug]

It makes sense that it is a cost reducing measure.
The process of soldering a lid to a die and taking into account the prevention of microcracking, is much more intensive manufacturing wise than applying a TIM patch and glueing the lid on.
The latter option allows for much more tolerance which means more speed and less accurate placement needed.

For how much some of these parts cost, it would be reasonable to expect they didn't spare the expense of a proven better method of attaching the IHS.

Intel has had a rocky past with soldering quality and it seems like instead of improve or fix their process, they ditched it entirely instead. I shouldn't be able to get a 12C drop delidding a soldered CPU, but der8auer showed on multiple samples you can expect that on Haswell-E. Broadwell-E was better with 6C drops, but still too much.

IIRC someone delidded ryzen and only saw 3-4C drops.

 

[William Gaatjes]

For how much some of these posts cost, it would be reasonable to expect they didn't spare the expense of a proven better method of attaching the IHS.

Intel has had a rocky past with soldering quality and it seems like instead of improve or fix their process, they ditched it entirely instead. I shouldn't be able to get a 12C drop delidding a soldered CPU, but der8auer showed that on multiple samples you can expect that on Haswell-E. Broadwell-E was better with 6C drops, but still too much.

IIRC someone delidded ryzen and only saw 3-4C drops.

Interesting.
That raises the following questions :
What is the temperature range of an AMD cpu and what is the temprature range of an Intel CPU ?
What kind of solder alloy does AMD use ? And what kind of solder alloy does/ did Intel use ?
Is the lid of the same thickness and material. Is the surface of the die prepared in the same way ?
Is the silicon being doped differently ?


I can imagine it is a serious technical challenge with all the constant temperature cycles.
All the materials which have different thermal expansion coefficients.
To match that all to get a product that has a life time of at least 10 years... Serious research.

 

[Ichinisan]

Heat spreader is meant to make heatsink installation idiot-proof. It has a slightly negative impact on thermal conductivity.

 

[Excessi0n]

What I want to know is why they don't have a protective... frame or spacer or whatever you want to call it, like on GPUs.

 

[William Gaatjes]

What I want to know is why they don't have a protective... frame or spacer or whatever you want to call it, like on GPUs.

Normally Graphic cards comes pre-assembled with heatsink. The graphics card manufacturer is fully adhering to the assembly instructions of the gpu designer. One cannot expect that from end user customers. Any fail on the graphic card is the graphic card manufacturers responsibility to find out the cause.

 

[PG]

Someone forgot the AMD socket A days and crushed cpus.

 

[mikeymikec]

Someone forgot the AMD socket A days and crushed cpus.

More likely someone forgot to read the OP before posting.

 

[Kenmitch]

Someone forgot the AMD socket A days and crushed cpus.

The good old days of the #2 pencil overclock. I never had any issues damaging my cpus.

 

[maddie]

I'll start with what may be my misunderstanding as to why CPU lids became a thing in the first place: As CPUs started to draw and waste more power, heftier coolers were required, and heftier coolers generally meant heavier coolers that have to stay attached to the CPU with increasingly elaborate mounting structures, and exerting as much pressure as safely possible to ensure maximum conductivity. The greater the pressure that's exerted, the more likely it is that mistakes are going to be made when attaching or detaching the HSF, and the lid is designed to protect the die.

The lids are also called heat spreaders, and I'm coming to the crux of my question now: To save money at the expense of conductivity, the lid is no longer soldered to the heat spreader; is there still a lot of point in having a heat spreader in most cases? When I say "most cases", I mean the vast majority of users out there who aren't having the highest end and/or overclocked CPUs. Also, given that most modern desktop CPUs do not waste anywhere as much energy as their predecessors (mainly evidenced IMO by the ever-shrinking stock HSFs), so CPU lids/heat spreaders still serve much of a purpose these days?

For the record, all of the power consumed by a CPU, within 1% accuracy is transformed to heat and carried off to the environment by the cooling system. There is often a false belief that the heat generated by a CPU is the wasted portion of the energy consumed and the useful part goes somewhere else. Where? Try asking those with such beliefs.

The fact that laptop CPUs and graphic card GPUs do not have a IHS should explain the use of one on consumer installed CPUs.

 

[DrMrLordX]

The APUs are considered "budget" parts. They used TIM to keep the price point down.

Interestingly enough, AMD soldered their Godavari/Kaveri refresh parts, like the 7870k. Not sure if they soldered Bristol Ridge.

 

[IntelUser2000]

It makes sense that it is a cost reducing measure.
The process of soldering a lid to a die and taking into account the prevention of microcracking, is much more intensive manufacturing wise than applying a TIM patch and glueing the lid on.
The latter option allows for much more tolerance which means more speed and less accurate placement needed. Less time spend per cpu is more profit.

This is probably a more accurate statement than the simple "Oh Intel wanted to save 2 cents by using TIM". They use TIM on Knights Landing Xeon Phi chips. They have die sizes in the range of 650mm2. It would arguably be more reliable too. There's no better cushion than semi-liquid state TIMs are in.

There is often a false belief that the heat generated by a CPU is the wasted portion of the energy consumed and the useful part goes somewhere else

Yep. It's not like its a light bulb, or a mechanical device. It's especially true of modern CPUs that have a dedicated power control unit and a Turbo mode that tries to maximize TDP use to improve performance. TDP specs for Intel for example, says long average Turbo has to be within TDP specs. Otherwise, it would be a catastrophe for heatsink and system designers.

Earlier CPUs without Turbo and less advanced power management schemes had much harder time reaching TDP, because power usage would be application dependent and maximum frequency would be identical between applications.

As late as early 2000s, CPUs were rated at different power ratings per SKU. A 1GHz chip would have 30W rating, and an 800MHz one may have a 20W rating for example. They decided to move to a more generic rating system called TDP at some point. TDP is a family rating, where multiple SKUs are rated same. Quality can vary depending on the chip, so having a family TDP which is managed by Turbo and power management schemes are more accurate.

 

[Mr Evil]

For the record, all of the power consumed by a CPU, within 1% accuracy is transformed to heat and carried off to the environment by the cooling system. There is often a false belief that the heat generated by a CPU is the wasted portion of the energy consumed and the useful part goes somewhere else. Where? Try asking those with such beliefs...

That's the wrong question. The energy used by current CPUs is all wasted, so there is no need to justify where the useful part goes (although that is only true in the long-term over multiple computations. In the short term, the "useful" part goes into moving electrical charge around to store information).

 

[psolord]

AMD and Intel should sell both lidded and de-lidded cpus, for the users that want them. They could just use different warranties for the different models, if it was necessary.

 

[dullard]

Theres a whitepaper that can be found about microcracking in solder between a die and an IHS. It shows that with decreasing die size, microcracking becomes more and more prevalent.

Here is one of several papers on the issue:
https://pdfs.semanticscholar.org/40fe/11ccc30fcdac3d2c38cb65b42f112f040558.pdf
Figure 10 and 14 show these cracks. Here is a relevant quote from that paper:

The interaction between increasing gold thickness and increasing thermal cycling has been seen in other experiments conducted at Intel Corporation (details on the experiment will not be given here, see Figure E for results) and is attributed to thermal-cycling-driven fatigue cracking between the indium and the indium-gold intermetallic. Stresses are created in the solder bondline in the assembled package as during thermal cycling due to the differing coefficients of thermal expansion of the package components (see Figure F). When thermally cycled to extremely long readouts, the stresses in the package can lead to growth of fatigue cracks in the solder TIM.

Basically, chips that are soldered form cracks in the corners and microcracks throughout. XabanakFanatik is correct, the types of cracking is a function of die size. These cracks slows the chip down over time (damaged chip or simply worse heat transfer). We get regular posts here about overclocks degrading over time (or voltages needing to be changed). That is the best case scenario, the worst case is a chip failure.

With highly adjustable turbos (faster temperature changes in thermal cycles) and often the maximum temperature allowed to be in the 90°C range instead of what used to be about 70°C (larger temperature changes in thermal cycles), the cracking due to thermal stresses will just get worse. Liquid TIM is a simple way to avoid any of these concerns. A chip with liquid TIM on day one will perform the same as on day 3000.

Solder is the enemy of reliability and end-of-life performance. Intel choose to have slightly worse performance with TIM in order to have a more reliable chip. Yes, this is a financial reason too. Soldering does cost money. But soldering also costs in warranty, service, customer satisfaction, and potentially lost customers.

 

[maddie]

Here is one of several papers on the issue:
https://pdfs.semanticscholar.org/40fe/11ccc30fcdac3d2c38cb65b42f112f040558.pdf
Figure 10 and 14 show these cracks. Here is a relevant quote from that paper:

Basically, chips that are soldered form cracks in the corners and microcracks throughout. XabanakFanatik is correct, the types of cracking is a function of die size. These cracks slows the chip down over time (damaged chip or simply worse heat transfer). We get regular posts here about overclocks degrading over time (or voltages needing to be changed). That is the best case scenario, the worst case is a chip failure.

With highly adjustable turbos (faster temperature changes in thermal cycles) and often the maximum temperature allowed to be in the 90°C range instead of what used to be about 70°C (larger temperature changes in thermal cycles), the cracking due to thermal stresses will just get worse. Liquid TIM is a simple way to avoid any of these concerns. A chip with liquid TIM on day one will perform the same as on day 3000.

Solder is the enemy of reliability and end-of-life performance. Intel choose to have slightly worse performance with TIM in order to have a more reliable chip. Yes, this is a financial reason too. Soldering does cost money. But soldering also costs in warranty, service, customer satisfaction, and potentially lost customers.

Electromigration is totally a non-issue it seems.

If soldering is the main issue then it will express itself in all the extremely dirty, poorly maintained desktops worldwide which run at quite high temps daily. Soldering is your hammer and degraded overclocks appears to be the nail.

 

[DrMrLordX]

AMD and Intel should sell both lidded and de-lidded cpus, for the users that want them. They could just use different warranties for the different models, if it was necessary.

I know, right? I've been calling for that for some time now, but the likelihood of that happening seems very low. Everyone's still allergic to the thought of end users cracking dice or what have you.

Meanwhile, people are selling delid tools to get around the crappy TIM underneath the lids of Intel (and some AMD) chips.

Basically, chips that are soldered form cracks in the corners and microcracks throughout. XabanakFanatik is correct, the types of cracking is a function of die size. These cracks slows the chip down over time (damaged chip or simply worse heat transfer). We get regular posts here about overclocks degrading over time (or voltages needing to be changed). That is the best case scenario, the worst case is a chip failure.

I'm gonna call foul on that leap of logic. Just because someone observed die damage due to long-term thermal cycling with solder in use, does not mean that enthusiast overclocks degrading over time has anything to do with that phenomenon. Electromigration is real, and many of those enthusiast overclockers have kept temps quite low via watercooling etc. while pumping obnoxious volts through their chips. See below about temp ranges. Also consider that you are seeing this phenomenon on chips with TIM under the lid as well! Someone's Haswell that starts to slow down had/has TIM under the lid. No solder-induced cracking there.

With highly adjustable turbos (faster temperature changes in thermal cycles) and often the maximum temperature allowed to be in the 90°C range instead of what used to be about 70°C (larger temperature changes in thermal cycles), the cracking due to thermal stresses will just get worse. Liquid TIM is a simple way to avoid any of these concerns. A chip with liquid TIM on day one will perform the same as on day 3000.

Well guess what, not every CPU company allows their chips to hit 90C or higher during any kind of operation. Most of AMD's CPUs still shut down somewhere in the 70s. Ryzen can get a little bit toasty if you are not careful, but it's been awhile since I let mine hit temps above 75C, and the package temp never got that high.

Solder is the enemy of reliability and end-of-life performance.

Bollocks, I've got a soldered Phenom downstairs and a soldered Athlon II x2 upstairs, both in working condition. Years from now, you will probably have soldered Ryzen chips plugging away without complication, and I would suspect the same of Haswell-E.

 

[William Gaatjes]

Here is one of several papers on the issue:
https://pdfs.semanticscholar.org/40fe/11ccc30fcdac3d2c38cb65b42f112f040558.pdf
Figure 10 and 14 show these cracks. Here is a relevant quote from that paper:

Basically, chips that are soldered form cracks in the corners and microcracks throughout. XabanakFanatik is correct, the types of cracking is a function of die size. These cracks slows the chip down over time (damaged chip or simply worse heat transfer). We get regular posts here about overclocks degrading over time (or voltages needing to be changed). That is the best case scenario, the worst case is a chip failure.

With highly adjustable turbos (faster temperature changes in thermal cycles) and often the maximum temperature allowed to be in the 90°C range instead of what used to be about 70°C (larger temperature changes in thermal cycles), the cracking due to thermal stresses will just get worse. Liquid TIM is a simple way to avoid any of these concerns. A chip with liquid TIM on day one will perform the same as on day 3000.

Solder is the enemy of reliability and end-of-life performance. Intel choose to have slightly worse performance with TIM in order to have a more reliable chip. Yes, this is a financial reason too. Soldering does cost money. But soldering also costs in warranty, service, customer satisfaction, and potentially lost customers.

Thank you, that is very confirming.
Intel seems to have used indium bonded to a goldplated area on the inner side of the IHS.
What they mention is that the Indium gold TIM must not only carry the heat away from the die but also absorb the mechanical stress from the thermal cycling.
I am going to read it all tomorrow morning. Interesting.
I am really interesting what causes the cracking. After quickly glancing through, it reads as the solder TIM cannot handle the mechanical stress caused by all the different materials combined together with extensive thermal cycling.
IHS-gold-plated-Indium-Die. And the die, i am sure also has had some surface treatment before the indium is applied.

That still makes me wonder about the following questions :

What is the temperature range of an AMD cpu and what is the temprature range of an Intel CPU ?

What kind of solder alloy does AMD use ? And what kind of solder alloy does/ did Intel use ?
According to the popular websites following der8auer AMD also uses Indium and a goldplated IHS for the current range of ryzen chips.
Is the Indium used the same or is it of different purity ?
Do they use the same materials for the IHS ?
Is it some alloy, percentage differences can really change the mechanical characteristics of an alloy.
So much to wonder about.

 

[dullard]

Bollocks, I've got a soldered Phenom downstairs and a soldered Athlon II x2 upstairs, both in working condition. Years from now, you will probably have soldered Ryzen chips plugging away without complication, and I would suspect the same of Haswell-E.

Yes, you will have soldered chips plugging away just fine. And you will have overclocked chips plugging away just fine. And yes, your computers may be just fine. Why? Because you are skipping over the important detail: thermal cycling. An overclocked computer crunching non-stop isn't thermal cycling (it is probably at a fairly steady elevated temperature). A well-cooled (enthusiast) computer often isn't thermal cycling much since it isn't allowed to get to a hot temperature. A computer in your basement doing next to nothing isn't thermal cycling as it probably isn't doing much work.

The problem is people who start and stop high powered turbo use over years of time. That thermal cycling is the key usage detail you seem to have left out of all of your your examples. Companies care about typical usage and worst-case-scenario usage -- not people who take care of their computers with proper water cooling. A computer just plugging away is probably pretty close to thermal equilibrium which is the best-case-scenario for soldered chips.

 

[William Gaatjes]

He Dullard, do you think that there is a difference between pga packages and lga packages when it comes to warping of the substrate under influence of temperature ?
Is that the reason why AMD is still using PGA for the "smaller" cpu's ? Because PGA packages have less warping because of being more stiff because of having all the pins and different substrate build up ?
And that in that scenario there is less mechanical strain on the indium layer because of the obvious different thermal expansion coefficients of the iHS and the substrate ?

The bigger CPU's like EPYC and Threadripper (both LGA packages) have such a large substrate that the IHS is also large. Because the IHS is large, it is more relatively pliable.
The reason why AMD uses 2 functional dies and 2 dummy dies for threadripper.
Perhaps because of sheer size, the thermal expansion effects of the IHS are more absorbed by the IHS itself instead of purely by the indium solder used as TIM.