AMD 8 Core vs Intel 8 Core

[arkie87]

As for TIM being a bottleneck, if that's the case, the only way to combat it is active cooling. You can increase the heat flow through TIM by increasing temperature gradient. Since you want CPU to stay cool, you need to make external temperature positively frigid. You can either use thermoelectric cooler between CPU and liquid cooler, or you can put your liquid through a refrigerator. Both are done occasionally for extreme overclocking.

By active cooling, you mean refrigeration and/or cryogenic fluids? Liquid cooling with a pump is also "active" when comparing it to passive heat pipes.

[steve_v]

Well, if you want to look at it that way adding a fan to the heatsink is also "active"

In this case however, I think "active cooling" means reducing the temperature of the "cold side" of the interface to below ambient - with something like a thermoelectric (peltier), phase-change (refrigeration) or evaporative (liquid nitrogen etc.) system.

Once the thermal resistance of the heat spreader / TIM becomes the bottleneck, it's the only way to improve heat transfer further.

Edited September 1, 2015 by steve_v

[K^2]

By active cooling, you mean refrigeration and/or cryogenic fluids? Liquid cooling with a pump is also "active" when comparing it to passive heat pipes.

I mean expending energy to move heat from colder surface to a hotter one by means of Peltier Effect or mechanical heat pump. Refrigeration would be an example. Thermoelectric cooler, as I've mentioned, another.

Simply helping heat flow down a temperature gradient is still passive cooling in my books.

[arkie87]

Well, if you want to look at it that way adding a fan to the heatsink is also "active"

In this case however, I think "active cooling" means reducing the temperature of the "cold side" of the interface to below ambient - with something like a thermoelectric (peltier), phase-change (refrigeration) or evaporative (liquid nitrogen etc.) system.

Once the thermal resistance of the heat spreader / TIM becomes the bottleneck, it's the only way to improve heat transfer further.

If TIM becomes limiting factor, then you can simply remove TIM

It's called embedded cooling

It is actually the project i am working on

[Camacha]

http://www.intel.com/content/www/us/en/architecture-and-technology/hyper-threading/hyper-threading-technology.html

Allows the running of two threads on a single core for applications designed to take advantage of it.

That is the simplified version of what I posted yesterday, yes. The extra thread is what allows the CPU to switch tasks quickly, because other instructions are already at hand to execute as soon as the previous one got done or ran out of things to calculate. It is a tiny bit is doubled pipeline, which makes the OS thinks is it a full extra core. All this is why the gains are typically between none and ~15%, despite the extra thread. No actual calculative power is added.

*Wonky comparison alert*: see it as a shredder with two hoppers. Rather than having to reach back and pick up another log, you have one already loaded to push in each time the previous one gets pushed through. Or maybe compare the situation to a double clutch gearbox, where two clutches allow for much faster gear changes, since the gearbox is used closer to full capacity. In both cases, less time is wasted making logistical movements.

If TIM becomes limiting factor, then you can simply remove TIM

Or, as is more common, delid and/or lap your CPU. The latter will eliminate one instance of TIM, the former might two, though lapping a bare chip takes some bravery or a thick wallet.

Embedded cooling looks pretty cool though, I am curious if and when that will percolate through to consumer grade hardware.

Edited September 1, 2015 by Camacha

[arkie87]

Or, as is more common, delid and/or lap your CPU. The latter will eliminate one instance of TIM, the former might two, though lapping a bare chip takes some bravery or a thick wallet.

Embedded cooling looks pretty cool though, I am curious if and when that will percolate through to consumer grade hardware.

Lapped CPU's still have thermal interface resistance; thus, they still need TIMs, albeit a much smaller volume (probably not even possible to deposit that little).

And as a side, most silicon is already lapped and polished; its the heat sink and case that needs lapping.

My guess is it will take a while before it is necessary for consumer CPU's. Right now, its mostly applicable to other high flux electronics, like laser diodes and power amplifiers. CPU's have been getting more efficient (less heating per flop), so that bypasses the thermal limits.

Edited September 2, 2015 by arkie87

[Camacha]

Lapped CPU's still have thermal interface resistance; thus, they still need TIMs, albeit a much smaller volume (probably not even possible to deposit that little).

That is the thing - if the TIM is isolating your CPU worse than the remaining air gaps to, you are better off without any. Since some people manage to lap their CPU in such a fashion that it will stick to the cooler purely by atmospheric pressure (vacuum suction), I would say there is little air left between cooler and chip.

In general, I would not bother with lapping at all, though. It is tedious, hard to do really well and you might ruin your gear. It yields only a handful of degrees if done correctly, and properly voids any warranty. If you want to build a PC as a practical working machine, not as a project, I would just pick a good cooler, some good TIM, apply those properly and be happy. Intel's current focus on power efficiency means those degrees are less likely to make a relevant difference anyway. Pretty much the same as can be said about water cooling. They are just not reasonable options for anyone wanting a practical machine, or most bang-for-buck.

[arkie87]

That is the thing - if the TIM is isolating your CPU worse than the remaining air gaps to, you are better off without any. Since some people manage to lap their CPU in such a fashion that it will stick to the cooler purely by atmospheric pressure (vacuum suction), I would say there is little air left between cooler and chip.

This isnt correct. Even if they are so flat, that they float on top of each other and stick together, only a small fraction of the surfaces are actually touching, which is what thermal interface resistance is. Roughness is still on the order of 100-200nm even after lapping, and i havent heard of anyone who can go lower (surely not if you are hand lapping the chip yourself). Of course, it is better than if the surfaces werent lapped, but it is still too large (and not better than unlapped but with TIM).

In general, I would not bother with lapping at all, though. It is tedious, hard to do really well and you might ruin your gear. It yields only a handful of degrees if done correctly, and properly voids any warranty. If you want to build a PC as a practical working machine, not as a project, I would just pick a good cooler, some good TIM, apply those properly and be happy. Intel's current focus on power efficiency means those degrees are less likely to make a relevant difference anyway. Pretty much the same as can be said about water cooling. They are just not reasonable options for anyone wanting a practical machine, or most bang-for-buck.

I agree. Just go with commercially available products, unless you want to make your own project out of it (and possibly destroy your cpu).

[Camacha]

This isnt correct. Even if they are so flat, that they float on top of each other and stick together, only a small fraction of the surfaces are actually touching, which is what thermal interface resistance is. Roughness is still on the order of 100-200nm even after lapping, and i havent heard of anyone who can go lower (surely not if you are hand lapping the chip yourself). Of course, it is better than if the surfaces werent lapped, but it is still too large (and not better than unlapped but with TIM).

I think this is the point where someone needs to find some measurements I agree that even airtight lapped hardware will have a fair remaining roughness, but about it being worse than TIM I am not sure. A common problem with hot CPUs is someone putting on too much TIM, effectively putting a somewhat isolating layer between the cooler and CPU. Air is, of course, much worse, but I can imagine that TIM is less effective between those very flat surfaces and forms a much, much thicker layer.

[cantab]

Soldering the die to the heatspreader seems to be held up as the ideal. But supposedly it's hard - and thus expensive - to avoid cracking the die especially with the smaller dies as process nodes shrink. Intel restrict it to their expensive E-series processors now.

[Camacha]

Soldering the die to the heatspreader seems to be held up as the ideal. But supposedly it's hard - and thus expensive - to avoid cracking the die especially with the smaller dies as process nodes shrink. Intel restrict it to their expensive E-series processors now.

For stock CPUs, you are right. For enthusiast overclocking, internal TIM is probably better, because it allows for easier removal of the heatspreader.

[waterlubber]

Guys, guys. Just get an air conditioner, find the evaporator coils, and run them over your computer components. Turn on the AC to max. Bam, freon cooling.

[arkie87]

I think this is the point where someone needs to find some measurements I agree that even airtight lapped hardware will have a fair remaining roughness, but about it being worse than TIM I am not sure. A common problem with hot CPUs is someone putting on too much TIM, effectively putting a somewhat isolating layer between the cooler and CPU. Air is, of course, much worse, but I can imagine that TIM is less effective between those very flat surfaces and forms a much, much thicker layer.

Properly applied TIM is always better. With lapped surfaces, it is probably very difficult to apply the "proper" amount (to fill roughness but no more). TIM conductivity is still 100x larger than air (and solder is 10-100x higher than TIM). And yes, it is pretty easy to apply too much TIM, even for unlapped cpus

[Camacha]

Guys, guys. Just get an air conditioner, find the evaporator coils, and run them over your computer components. Turn on the AC to max. Bam, freon cooling.

They invented (or applied) Peltier, nitrogen and helium cooling for a reason A lot of folks have worked very hard on perfecting the art of cooling when it comes to computing, do not think you will improve something so easily.

Properly applied TIM is always better. With lapped surfaces, it is probably very difficult to apply the "proper" amount (to fill roughness but no more).

That is what I am saying. With properly lapped components, it is probably impossible to apply the right amount. You would need specialist equipment and if you have access that, there are probably better approaches to the matter (like CNC milling the two pieces to perfection or soldering/welding them in some fashion). Ergo, applying TIM might very well not be in your best interest.

Besides, like I said, when you take contact area and the thinner layer into account, the better TIM conductivity might not turn out so advantageous.

[arkie87]

Besides, like I said, when you take contact area and the thinner layer into account, the better TIM conductivity might not turn out so advantageous.

Not sure what you mean here.

If you could (magically) apply the minimum layer to fill the roughness, then TIM would be better, even for lapped surfaces. And it should reduce total interfacial resistance significantly (at least a factor of 2).