Overclocker checks Intel Core i9-10900K delidding benefits

[watercooled]

Neither were Intel's heatspreaders until recently and that only happened (IMO) because they're trying to squeeze extra life out of their 14nm node.

Intel CPUs only used paste between Ivy Bridge and Coffee lake 1 on desktop parts and only started using paste with Skylake on HEDT. All of AMD's desktop CPUs back to at least Phenom (and some of the APU parts) have been soldered - at least in the time I've been around PCs paste is the exception, not the rule. And it's besides the point anyway since Intel are now back to using solder again.

Your point about it 'not doing any more than a heatsink would' when using paste - I literally already said that in my last post before yours.

I believe you're talking about bonded dies in the second paragraph of your previous post? The thermal concern wasn't a non-issue as, besides being widely ridiculed in the press and by enthusiasts, it meant that it wasn't uncommon for stock CPUs to get excessively hot and throttle as a result. Adding more of the same cores on the same node, at higher clocks speeds - yeah, there probably wasn't much of an alternative but to start bonding the heatspreader again. I have a stock 7700 non-k which will get to the mid-80's under load when adhering to the 65W limit (my motherboard/CPU combination seems to stick to that limit and pulls clocks back to maintain it under heavy multi-threaded loads). That's not far off the throttle point despite the cooler barely being warm, and the CPU having a considerably lower TDP than the k variant. Those generations of CPUs had a horribly inefficient thermal path between die and IHS, and it's quite ridiculous IMO when you had people buying >£100 AIO coolers as a new normal to maintain the clocks of a stock quad core CPU...

[Corky34]

Intel CPUs only used paste between Ivy Bridge and Coffee lake 1 on desktop parts and only started using paste with Skylake on HEDT. All of AMD's desktop CPUs back to at least Phenom (and some of the APU parts) have been soldered - at least in the time I've been around PCs paste is the exception, not the rule. And it's besides the point anyway since Intel are now back to using solder again.

Not from what i understand, solder seems to be something reserved for CPUs higher up the product stack, and IIRC for things like CL STIM was reserved for the K series'.

Your point about it 'not doing any more than a heatsink would' when using paste - I literally already said that in my last post before yours.

I didn't exactly say it's not doing any more than a heatsink would' when using paste, did I? From a thermal POV i said...

You're also right in what you say later that direct die contact doesn't really buy you much more thermal headroom (maybe 5-10°c) so it doesn't really effect maximum clock speeds a great deal so i guess because they were seeing slowly increasing RMA rates and customer dissatisfaction as dies got smaller they wanted to address that issue more than what at the time was pretty much a non-issue.

And i said it's more a structural thing when i said it should really be called a loadspreader, didn't I?

[watercooled]

I didn't exactly say it's not doing any more than a heatsink would' when using paste, did I? From a thermal POV i said...

Well achtually...

It's a bit of a misnomer IMO as while it technically spreads the heat it doesn't do it any more than a heatsink would if it was in direct contact with the die, a more apt name would be loadspreader as that's the primary purpose of them, as dies got smaller the chances of chipping or cracking the die increased. Balancing a big heavy thing on a small bit of what's essentially glass and worse yet apply pressure to it with clips or screws can cause all sort of problems, especially if you've got little experience in building computers.

Can we please not do this? I'm so sick of ridiculous back-and forth arguments on this site lately. You incorrectly called out a straightforward factual statement that I stood by. I never said spreading heat was the only purpose, I said a heatspreader spreads heat. And guess what, by jove it does spread heat! If you read it wrong or inferred something else from that statement, argue with yourself instead, eh?

https://www.anandtech.com/show/13400...9600k-review/2

You're splitting hairs in defence of calling me out about heat spreaders spreading heat. Simple fact is, they do. I also mentioned they offer mechanical protection if you actually read what I post.

Can we move it back to a worthwhile discussion - it seems we do actually agree about the function of a heatspreader, in that it spreads heat and that its presence also offers mechanical protection like I originally said.

[Corky34]

Erm, i wasn't calling you out, splitting hairs, or having an argument.

I thought we were having a conversation, if i gave you another impression then I'm sorry as it wasn't my intention, i just thought it would be nice to have a talk about why heatspreaders became a thing and that sort of stuff, i guess i was wrong though.

[watercooled]

Fair enough, sorry I over-reacted. I did see direct back-and-forth contradiction of statements as a bit of a red flag though, but maybe that's just me after having that lead to actual arguments in other threads lately.

Like I say though I think we are fully in agreement, and either way IMO it's good to see a return to solder on Intel's desktop CPUs. I've heard various explanations of why they moved to paste in the first place - the simplest being that it's cheaper. A more complex reason I heard being something to do with smaller dies being more likely to crack but can't remember if this was attributed to manufacturing or thermal cycling. Applying Occam's Razor it does seem a tad convenient though?

[Corky34]

Just to clarify when i said it doesn't do it any more than a heatsink would i sort of meant in a spreading of the heat type thing, in a sort of spreading jam on toast sort of way.

I didn't do good job of explaining myself there did i and I'm still probably not but never mind as it's not that important.

Convenient or not i suspect they had to invest more (time+money) into addressing the smallish downsides of using a heatspreader as they've had to eek a few more years from a node they intended to retire, was it, a couple of years ago?

It would be nice to see them keeping those improvements when they do switch to a new node as even though they maybe small in the grand scheme of things it would be a shame to throw them in the bin, so to speak.

[watercooled]

In an ideal setting, which I don't consider a paste-attached heatspreader to be, a heatspreader's thermal job is to take heat from a small, power-dense area and effectively transfer that to a much larger surface area, making the transfer to a heatsink less critical. As a metal, solder does that very well as can be seen where the likes of der8auer has tried direct die cooling of manufacturer-soldered dies:
https://www.youtube.com/watch?v=c1wyUMZVv4k
https://www.youtube.com/watch?v=wz_-Q5QzRqg
From memory and without watching the whole videos again, I believe the results were similar or slightly worse than stock.

But again I completely agree, a paste-attached IHS really doesn't do that job very well and in many cases potentially worse than direct-die cooling because of the material itself and the fairly large gaps between die and IHS that were reported in previous generations.

Smaller nodes means we're likely to see even higher power densities moving forward, so I don't see the necessity of bonded dies going away any time soon, at least for the high-clocked desktop parts.

[Corky34]

I've not been able to stop thinking about heat spreading, yes I'm that sad, and I've got a question not really for any reason other than wanting to know.

Does the thickness of something, so the verticality of it from the heat source, effect the transfer of heat in the horizontal plane, again from the heat source? For instance if you heated a single point on a 5mm thick sheet of copper would you get burnt quicker holding it at a fixed distance from the heat source than if it was 25mm?

(Would that be called the thermal transfer rate?)

Sorry if it's not exactly related to the benefits of delidding an i9-10900K.

[watercooled]

Thermal conductivity works similarly to electrical conductivity in that it depends on cross sectional area of a material. However, you also have to factor in the thermal mass itself.

I'll use a copper wire as an example as it's easier to visualise - a thicker wire will have a lower thermal resistance much like it has a lower electrical resistance, and will therefore be able to conduct more thermal energy (for a given temperature gradient) than a thinner wire. However, in terms of which would burn more quickly, it depends. If the wire is so thin it can't conduct heat quickly enough, your fingers could sink the heat away quickly enough it would never get hot enough. On the other hand, a really thick piece of wire would take longer to reach temperature itself for a given thermal power.

Thermal mass can play an important role in processor cooling - the heatsink can store a certain amount of heat even if it could not continuously dissipate that power, and this is the principle behind some boost algorithms.

[Corky34]

So my instinct tells me how thick the material is shouldn't make a difference to how quickly the heat travels in the horizontal axis but then when i think about the logical of it all the stuff you mention comes into play and I'm no longer sure if my instinct is right.

I have a picture in my head of heat travelling out from the heat source like one of those slow-mo shots of a compression wave moving out from an explosion and i guess that's right if we're talking about a big mass like the air is, but when it's a thin mass in the vertical sense i keep getting the feeling it would change but I've got no sense of why, so confusing.

I get what you're saying about a wire BTW but doesn't that change more than two dimensions, IDK and sorry if I'm not doing a very good job of explaining my thoughts.

[watercooled]

No I do get what you're saying. The thickness would make a difference, but the significance of that may be small if you're away from the extremes. Ignoring the practical realities of such a heatspreader, a microscopically thin layer of copper isn't going to help a great deal because there's not enough 'meat' to carry that much thermal energy away.

Make that ridiculously thick, say a 1cm thick copper slug and in theory it makes a great heatspreader, but then you're at the point where you're relying on a potentially less conductive material than a decent heatsink's heat pipes, so it's no efficient as an overall system. You need the heatspreader to be thick enough to be useful, but not so thick it starts detracting from the performance of the heatsink. I wouldn't like to guess where the crossover point would be! It would vary depending on power output though.

In theory, having the heatpipes bonded (soldered) directly to the die would be better than going through a heatspreader but that's obviously impractical, and the reason the heatspreader is useful is because of the interface between it and the heatsink. And in reality, a thin copper shim isn't likely going to hurt much given heatpipes have thin copper envelopes anyway.

Thermal interface material isn't very conductive vs metal, even decent stuff (liquid metal stuff is a weird one I'm ignoring for now) - it's just better than the air gaps that inevitably exist between two surfaces.

And this thermal interface is really the crux of the matter - this is what you're having to send heat through. Sending say 100W of heat through 100mm2 (roughly the die area) of this low-conductivity substance is obviously going to be more of a challenge than efficiently moving that heat to the e.g. ~1400mm2 area of an AMD heatspreader, and through that surface area to the heatsink (assuming it's all in contact of course).

There are a lot of variables in play.

[Corky34]

There are a lot of variables in play.

Indeed, the more i thought about it the more i started to realise that and then thanked goodness that computers handle all the variables now, can you imagine trying to do computational flow dynamics (if that's even what it's called when talking about the movement of heat through things) with nothing but a pen and paper.

I've taken this post way OT haven't I so i best leave it hear, thanks for taking the time to enlighten me.

[Xlucine]

1d flow is not too hard (just like V=IR), 2d flow (e.g. point source in middle of thin heatspreader, or a line source in the middle of a idealised resistor) starts involving ln, and I'd rather pretend 3d flow doesn't exist

There's probably a lot of crossover between approaches for calculating the flow of charge around an IC and for predicting the heat flowing out of the IC, except in heat transfer calculations you normally get to assume steady state. I prefer to leave all that for the electronic lot & ignore time-variant stuff unless it's acoustics, because acoustics can get smushed into RMS values

[watercooled]

Yeah hence using a wire as an example