This is a good question as to why we don't see better heatsinks.

[AmbientFiction]

Ok I was just thinking.Why are Heatsinks flat on the bottom we are wanting to take heat away from the CPU right so why not make a heatsink that has a hole in the center for the core of the chip. If you think about it. If it was recessed in the center to say a perfect core cut depth and width then it would in turn take away more heat because you would have more of the core in touch with the heatsink.This way the thermal paste would form a better bond around the core , you would stand less chance of cracking the core and over all it would carry more heat away from the cpu.

[William]

no, you would raise your temps. You can see the effect you would have by looking at what happens when you have a copper shim on a cpu, the temp go higher because you have heat on the hsf radiating into the CPU ceramic.

[AmbientFiction]

William (Jul 17, 2001 11:37 p.m.):
no, you would raise your temps. You can see the effect you would have by looking at what happens when you have a copper shim on a cpu, the temp go higher because you have heat on the hsf radiating into the CPU ceramic.
So your saying a copper heatsink with a notch to fit over the core of a P/// would raise the heat and not lower it?
I am not saying sit the Heatsink flat on the chip just a small recess into it as so it wouldn't crack the core

[merlins_wraith]

I had the same question several months ago. Never really got a good answer on it though. If you have the tools to do it I think it would be worthwild. I tried it with a dremel ... HA! Decided that water was the way to go instead. Best of luck!

[Colin]

Interesting thought. Take the four vertical sides of a CPU die and add their surface area to the HS. I believe the next problem would be CPU manufacturers controlling the amount of goop between the CPU die and the ceramic plate it's mounted one. There seems to be quite a variance in the chips I have sampled.

[Hoot]

I don't recall the exact value, but I believe the core only sits about .036 inches above the ceramic substrate. That is not a lot of additional surface area. All the same, I also toyed with the prospect of doing some precision milling to not only inset the core, but also all the surface mount components, so that the HS baseplate mated with the core at the same it met the ceramic.

People poo-poo this because it causes more of the core heat to take a different path (through ceramic) to the HS. I believe that any parallel heat conducting path, no matter how resistive, is a plus in reducing the core temp.

Consider you car engine. The design is for the heat in it to get removed by the water cooling system. Some still gets wicked away in the motor oil, transmission, etc. Those secondary paths help lower the temp a little bit and in the case of the CPU, every little bit helps. It certainly would screw with your in-socket thermistor readings, but who's worse off? The person who's thermistor reads too low, or reads too high?

This point is mostly moot and for the same reason, I did not pursue it. Namely, how many OC'ers have access to a decent mill? Documenting and writing it up for publishing as an article would not benefit many people while constituting a lot of work.

Now don't crowd each other, stepping up to take a swing at my logic. ;D

Hoot

[William]

well if you kept it off the ceramic it would probably be better, but the second it touchesit you will get the "shim effect" that is present with even the nonconductive shims. Still, if the temp difference was minimal(2-3C) but the core was safer, it would be worth it.

[AmbientFiction]

I wasn't so worried about the temp as I was the core in my thought process think about the size of the new copermines the core is flippin tiny.

[merlins_wraith]

Ok next question:

What are the heat transfer properties of ceramic versus commonly used metals (ie. copper aluminum brass)?

- Isn't ceramic used as an insulator?
- If ceramic essentially 'soaks' up heat then why don't we have ceramic heatsinks?

Just a couple of thoughts that popped into my head ... wait a few and they will pass.

[cjtune]

merlins_wraith (Jul 19, 2001 03:09 a.m.):
Ok next question:

What are the heat transfer properties of ceramic versus commonly used metals (ie. copper aluminum brass)?

- Isn't ceramic used as an insulator?
- If ceramic essentially 'soaks' up heat then why don't we have ceramic heatsinks?

Just a couple of thoughts that popped into my head ... wait a few and they will pass.


The ceramics used for the CPU core dies are specially formulated and used so that they are the best compromise between heat conductivity, mechanical strength, and electrical non-conductance. In other words, AMD, Intel and other chip makers have no choice but to use ceramic although its thermal properties are way inferior to that of metal.

How much thermal energy a material 'soaks' up depends on the specific heat capacity, that is, how many joules or BTUs of thermal energy a gram of the material can hold before rising in temperature. On the other hand, this is of no concern in heatsink design as the cooling we are all concerned about is a steady state process. Imagine a cold heatsink. When you first turn on your PC, the heatsink will get heated up and for some time the temperature will rise until it can establish its max heat transfer to the ambient. When that is achieved, the heatsink (and the CPU) will reach a steady temperature -and it'll be like this till you turn off your PC hours or days later. Thermal conductivity (ie. how much heat transferred per temperature difference per thickness) is much more of a concern to heatsink designers.

[cjtune]

Hoot (Jul 18, 2001 08:01 a.m.):
People poo-poo this because it causes more of the core heat to take a different path (through ceramic) to the HS. I believe that any parallel heat conducting path, no matter how resistive, is a plus in reducing the core temp.

Hoot

Agreed, more parallel paths (or parallel thermal resistances) permit better heat transfer although maybe just a fraction of the waste heat of a CPU passes through the ceramic base to the heatsink. The 'shim effect' is most probably due to the poor contact between the die and the heatsink due to the improper installation of a shim or the use of a poorly dimensioned shim (ie. a bit too tall).

[cjtune]

AmbientFiction (Jul 18, 2001 09:58 a.m.):
I wasn't so worried about the temp as I was the core in my thought process think about the size of the new copermines the core is flippin tiny.

Maybe we've discovered another upper limit to the number of transistors they can fit into a tinier die -the problem of throwing off waste heat.

[YahooDave]

Hoot (Jul 18, 2001 08:01 a.m.):
I don't recall the exact value, but I believe the core only sits about .036 inches above the ceramic substrate. That is not a lot of additional surface area. All the same, I also toyed with the prospect of doing some precision milling to not only inset the core, but also all the surface mount components, so that the HS baseplate mated with the core at the same it met the ceramic.

People poo-poo this because it causes more of the core heat to take a different path (through ceramic) to the HS. I believe that any parallel heat conducting path, no matter how resistive, is a plus in reducing the core temp.

Consider you car engine. The design is for the heat in it to get removed by the water cooling system. Some still gets wicked away in the motor oil, transmission, etc. Those secondary paths help lower the temp a little bit and in the case of the CPU, every little bit helps. It certainly would screw with your in-socket thermistor readings, but who's worse off? The person who's thermistor reads too low, or reads too high?

This point is mostly moot and for the same reason, I did not pursue it. Namely, how many OC'ers have access to a decent mill? Documenting and writing it up for publishing as an article would not benefit many people while constituting a lot of work.

Now don't crowd each other, stepping up to take a swing at my logic. ;D

Hoot


I think your biggest problem would be getting enough pressure on the core. As soon as you mill the HS and drop it arounfd the sides of the core you will start to loose pressure on the top of the core. I don't think the disipation of the heat through the sides and the ceramic would be enough to compensate for the loss of pressure.

[merlins_wraith]

Well I would think that you could modify the clip, or build a custom solution to fix that problem. Just as long as the milled hole was not deeper than the height of the cpu die.

[William]

someone asked about the properties of ceramic, it sucks at absorbing heat, but when it does, it holds it in pretty well.

[diehrd]

I personally think a heat sink needs to be thicker at the contact area of the processor and needs to thin out as it moves to the outer edges,Further finn design would bnifit from thinner fins at the center and thicker finns as it moves to the edges.

Some one think hard on this idea because I think it would produce a huge stepp foward in air cooling,So big a difference that water cooling would be unneeded.

[asmodean]

Well, looks like AMD is moving on to carbon-based casing for their chips. That'll scrap the whole ceramic heat question, atleast for desktop Palominos. And it'll scrap a LOT of heatsinks too (It's not as tall as the current AMD CPUs). Thank god I'm using water...

I was thinking of that recession too few weeks ago, but didn't bother so much about it, as the easiest solution to try that is getting a bit too short (eg. not enough height) shim and put some AS2 & superglue or AS thermal adhesive on it. Then just align it perfectly on the chip and put the HSF on it and apply enough pressure.

Preferably use an Intel chip or AthlonMP to test, as they both got an on-die thermistor. Can't trust that in-socket one much, it gets affected by air in the socket too much even if it's touching the bottom of the chip.

[cjtune]

diehrd (Jul 22, 2001 09:46 a.m.):
I personally think a heat sink needs to be thicker at the contact area of the processor and needs to thin out as it moves to the outer edges,Further finn design would benefit from thinner fins at the center and thicker finns as it moves to the edges.


Did you get this idea from one of them new Globalwins? I did see one with a thicker base right over the CPU core and that thins out away from the core. Be aware that a thin base constricts heat transfer horizontally and this inhibits the distribution of heat to the outer fins that, incidentally, gets the most airflow from the fan(s), ie. heat doesn't spread well. Maybe the base is made thinner so as to give the outer fins more height relative to those in a block of metal of same dimensions but with all fins with the same height. A taller fin has more surface area and the cross-sectional area from the thinner base may just be adequate at letting enough heat through to that taller fin.

I think that, ultimately, without refridgeration (peltier or otherwise), we will run into a upper limit of heat transfer even water cooling let alone air cooling. The total C/W of any heat removal system, is basically a summation of a series of many thermal resistances:
1.) Conductive thermal resistance of CPU core ceramic.
2.) Contact/interface thermal resistance between CPU core and bottom of heatsink or waterblock.
3.) Conductive thermal resistance of heatsink metal and geometry.
4.) Convective thermal resistance of airflow and heatsink/waterblock geometry.

Once we have 'taken cared' of thermal resistances (3) and (4), we are still pretty much stuck with (1) and (2), which depends on how much available area of the core is in contact with the heatsink and quality of the mating. In the end, to ensure the superfluous mating of the CPU core to the heatsink, it may have to be done in-house at the plant and will be out of the hands of third-party HSF providers.



Tcpu----/VVVV--------/VVVV-------/VVVV-------/VVVV---- Tambient
1 2 3 4

[most deaf]

they would have to design different ones for different chips, cost more

[sillyunclemark]

I vote for more cooling paths used the better.

Any thoughts on a gasket around the perimiter of the cpu and heat sink and filling the voids with a non conductive thermal compound, mineral oil or or such. Maybe put small fins on the underside of the hs around the core. I know the more metal and less goop the better.

Just a meandering thought. :)

[KaHNZa]

Whoa, this thread got brought back from the grave! :eek: It is an interesting concept though.

[madcow235]

coppermines :)

[DManeKid]

lol it was only a year ago

[Malpine Walis]

Originally posted by Hoot


This point is mostly moot and for the same reason, I did not pursue it. Namely, how many OC'ers have access to a decent mill? Documenting and writing it up for publishing as an article would not benefit many people while constituting a lot of work.

Hoot

And if it is not custom machined, you would have a hard sell to get even one manufacturer to make a different heat sink for each die size out there.

On the other hand, perhaps there could be an aftermarket for a shim that picked up the sides of the core.

Like the red in this pic:

[arhines]

I seriously doubt that anyone could effectively (not to mention inexpensively) make this... I guess if you were really hardcore you could make a shim, extend the copper from it to the side, and attach another heatsink to that....getting REALLY excessive though, if you ask me.

[CryptokiD]

WOW this is the prefect thread to exlpain my setup.

basicly what i did was put plastic wrap tightly around the heatsink base, and i poured thermal epoxy all over my processor so that it covered the ceramic from the sides all the way to the core, then i put the heatsink on and let it dry 24 hours.
then i removed the heatsink and took the processor and lapped that thing untill the thermal epoxy was dead flat with the core's top, i mean flat enough that a razor blade didnt catch a line.

my reasoning is that the core of a processor throws heat off in all directions, not just twards the heatsink, so now i can perfectly cool the sides of the core as well as the back of it. because the back of the core tramsmits heat to the ceramic which is now perfectly flat with the core's top. this lets me activly cool 100% of the sides of the core.
hows it working?

on a passivly water cooled athlon (no fans) i am running at 10C over ambient, at 1.95V and 1600mhz 10x320

any questions, email me at josephsmith@directvinternet.com, no spam please:)
ill also check this thread again later,i hope i helped someones noggin spark an idea:)

ohh i forgot to meantion my cpu is stock@ 1.33ghz ahyja"9" t-bird

[sillyunclemark]

CryptokiD
Do you have before and after numbers on that thermal epoy mod?

Thinking about trying the epoxy with a copper shim.