Why do all the new CPUs have heatspreaders?

[Concillian]

It seems like I distinctly remember FCPGA being touted as excellent for cooling when the Athlons, durons Pentium IIIs and Celerons all moved to the FCPGA packaging that had the heatsink mounted directly on the core. I mean, it makes sense right? What better way to cool the CPU than to directly cool the core?

So why do the K8s and P4s have heatspreaders? There has to be some efficiency loss in cooling through an extra layer of interface material and whatever the heatspreader is made from.

I know the early durons and Athlons saw some deaths do to heatsink installation, where the core got cracked or chipped from rotating the heatsink, but since they started putting on the little foam/rubber dots, I have heard nothing more of this.

So I'm just curious if anyone knows why a heatspreader is better than not having one? What technical advantage does a heatspreader bring? Or is it strictly a must to keep away risk of core breakage on heatsink installation? Or something completely different?

 

[CTho9305]

My guess: with a heatspreader, you have a larger area contacting the heatsink, and the junction between the core and heatspreader is completely controlled by the manufacturer. This means that even with a less-perfect application of thermal interface material by an inexperienced user, you still get decent cooling because of the larger surface area.

 

[Sahrin]

Originally posted by: Concillian
I know the early durons and Athlons saw some deaths do to heatsink installation, where the core got cracked or chipped from rotating the heatsink, but since they started putting on the little foam/rubber dots, I have heard nothing more of this.

So I'm just curious if anyone knows why a heatspreader is better than not having one? What technical advantage does a heatspreader bring? Or is it strictly a must to keep away risk of core breakage on heatsink installation? Or something completely different?

>>You touched upon one big reason right there, protection. Intel's move to the LGA only proves that the CPU Manufs are aware of and don't like lot's of RMA's from physical CPU damage. Additionally, like the name says, a heatspreader is designed to evenly distribute the heat of the processor across the core. I know this sounds kind of...counterintuitive, but parts of the core run much hotter than others (specifically execution units v. cache/registers). If only part of the core is hot and the rest is cold, in a direct contact-to-HSF situation you only get that tiny amount of surface area to conduct the heat into the Heatsink (only the hot part of the core transmits the heat, not the relatively cool parts). The heatspreader allows you to first take that heat away from the hot areas of the core immediately, and then gives you a much larger surface area over which to dissipate the heat to the Heatsink. Heatspreaders are generally made out of high-grade copper, just like the best heatsinks...so as long as the TIM is good on the outside (the stuff they use on the inside is *very* high quality) there is no real difference in heat conduction (as you might think there would be with stepping from core>IHS>HSF v. core>HSF) but there is a difference in the maximum amount of heat that can be produced (think of it like slowing your bus speed but increasing the bit-depth).

 

[borealiss]

in general, it's for protection of the core. silicon has basically the continuity of glass. you're not losing any thermal conductivity either by adding an extra layer with the heat spreader. it's also for the thermal protection of the die itself, as in most cases you can run an athlon 64/opteron without any thermal grease for quite a while, something i've done in the lab for quite some time, something not possible on earlier generation athlons. you can even boot into the bios of a motherboard for some time without a heatsink on the cpu before thermtrip shuts the system down.

 

[flashbacck]

Doesn't it also prevent you from modding the chip by cutting or filling bridges?

 

[thelordemperor]

If you're modding bridges it's not much more of a task to take the heat spreader off... if there are even any under there.

 

[pm]

It's for spreading heat - not for protection, although that's an added side-benefit. The surface of a C4 "flip-chip" CPU is not flat, due to the way the resin underfill flows under the bump array. When the die is fairly small, this effect is not very pronounced, but on larger dies it can be substantial. The IHS (Integrated Heat Spreader) on Intel CPU's is used to create a planar surface to the heatsink. It also allows for better themal conduction through the sides of the die.

 

[borealiss]

depends on the part. most cpus are fused now.

 

[CTho9305]

Originally posted by: pm
It's for spreading heat - not for protection, although that's an added side-benefit. The surface of a C4 "flip-chip" CPU is not flat, due to the way the resin underfill flows under the bump array. When the die is fairly small, this effect is not very pronounced, but on larger dies it can be substantial. The IHS (Integrated Heat Spreader) on Intel CPU's is used to create a planar surface to the heatsink. It also allows for better themal conduction through the sides of the die.

Any idea how much cost is added? If you figure something like 1% of chips are RMAed due to careless users cracking them, you'd have to have a pretty cheap heat spreader to make it a financially sound decision. I doubt even 1% of chips get cracked.

 

[Jeff7]

Originally posted by: Sahrin

Originally posted by: Concillian
I know the early durons and Athlons saw some deaths do to heatsink installation, where the core got cracked or chipped from rotating the heatsink, but since they started putting on the little foam/rubber dots, I have heard nothing more of this.

So I'm just curious if anyone knows why a heatspreader is better than not having one? What technical advantage does a heatspreader bring? Or is it strictly a must to keep away risk of core breakage on heatsink installation? Or something completely different?

>>You touched upon one big reason right there, protection. Intel's move to the LGA only proves that the CPU Manufs are aware of and don't like lot's of RMA's from physical CPU damage. Additionally, like the name says, a heatspreader is designed to evenly distribute the heat of the processor across the core. I know this sounds kind of...counterintuitive, but parts of the core run much hotter than others (specifically execution units v. cache/registers). If only part of the core is hot and the rest is cold, in a direct contact-to-HSF situation you only get that tiny amount of surface area to conduct the heat into the Heatsink (only the hot part of the core transmits the heat, not the relatively cool parts). The heatspreader allows you to first take that heat away from the hot areas of the core immediately, and then gives you a much larger surface area over which to dissipate the heat to the Heatsink. Heatspreaders are generally made out of high-grade copper, just like the best heatsinks...so as long as the TIM is good on the outside (the stuff they use on the inside is *very* high quality) there is no real difference in heat conduction (as you might think there would be with stepping from core>IHS>HSF v. core>HSF) but there is a difference in the maximum amount of heat that can be produced (think of it like slowing your bus speed but increasing the bit-depth).

But why transfer the heat first to a small piece of metal, rather than to a larger piece, with much more surface area? Think of the bottom part of the heatsink itself as a heatspreader - but it's just a heatspreader that has a lot of big fins attached to it, with a fan screwed into it.

I do think it has much to do though with protecting the core. Why not something like an integrated shim though? Something precision made from AMD/Intel that is the same height as the core, or maybe a very slight bit shorter? With the new organic packaging, the entire package could flex slightly to absorb the excess pressure, without damaging the core.

the stuff they use on the inside is *very* high quality

Maybe. I've take apart some K6-2 and 3 processors already. The rubbery interface material had tiny air bubbles all throughout it. Hopefully they use something better nowadays. Hopefully. It'd be good if they don't skimp on it the way motherboard manufacturers skimp on the compound beneath the chipset coolers.

The IHS (Integrated Heat Spreader) on Intel CPU's is used to create a planar surface to the heatsink. It also allows for better themal conduction through the sides of the die.

That would be a benefit. I didn't know they did that. Can't imagine you could have a heatsink like that - the slightest twisting motion would rip the core right off.

 

[NeoPTLD]

Perhaps better installation error and cooling error resistance offers superior total solution between Intel and prebuilt companies?
Chipped cores can result in not so obvious stability issues. Definitely not an issue Dell and the such want to deal with.


As long as the CPU works to the specification, Intel couldn't care less how theoretically good the thermal transfer is.

 

[Peter]

Why heatspreaders? What pm said. There's protection, and the fact that you present a machined, flat surface. On the other hand, it adds an extra two layers of material to the heat transfer path (the heatspreader and the attachment material on the CPU die). This has a negative effect.

Also, notebook CPUs are usually w/o heatspreader - simply because notebook makers like their stuff as flat as possible. The extra couple of millimeters from a heatspreader is unwelcome here.