News Intel experimenting with direct liquid cooling for up to 1000W CPUs - package-level approach maximizes performance, reduces size and complexity

[Admin]

Intel experimenting with direct liquid cooling for up to 1000W CPUs

Intel experimenting with direct liquid cooling for up to 1000W CPUs - package-level approach maximizes performance, reduces size and complexity : Read more

 

[Reverend_Clint]

I worked on a complete watercooled sappire rapids server that needed more power than we had in the lab. 2 cpus, 4 GPUs, Ram, and PSUs were liquid cooled.

 

[thestryker]

While I doubt the economic realities would never make sense this would be an interesting option for something like a KS part.

I imagine as chips shrink and cooling becomes more problematic the answer at the consumer level will probably be something along the lines of a vapor chamber IHS and/or thinner TIM.

 

[truerock]

My understanding is that when Intel (and AMD) went to liquid metal solder and stopped using thermal paste - that knocked off about 90% of the inefficiency of removing heat externally from a CPU. Perhaps some type of super-conductive-carbon-based-paste and carbon-based-lid might squeeze out another 5% to 9%.

To address heat in the future, I think they will need to work at the diode level inside of the CPU.

 

[thestryker]

My understanding is that when Intel (and AMD) went to liquid metal solder and stopped using thermal paste - that knocked off about 90% of the inefficiency of removing heat externally from a CPU.

What they both use is actually rather thick, but it has very good heat conducting properties. The thickness however is why consumer available liquid metal is more effective than what they use. It's entirely possible they could engineer a different type of TIM which is more effective.

Roman actually talked about this specifically in a recent video about TIM and conductivity:

Spoiler

View: https://youtu.be/1BfSx5IG42c?si=9hG4KtLEfCdziPPg

 

[Vanderlindemedia]

My understanding is that when Intel (and AMD) went to liquid metal solder and stopped using thermal paste - that knocked off about 90% of the inefficiency of removing heat externally from a CPU. Perhaps some type of super-conductive-carbon-based-paste and carbon-based-lid might squeeze out another 5% to 9%.

To address heat in the future, I think they will need to work at the diode level inside of the CPU.

We need solder, plus copper. That's pretty much the closest you can get. Anything "more" requires hefty tech such as direct die cooling and such. Not for consumers, not for enterprise.

 

[Alex/AT]

1000W CPU? Wow. Now you'll actually be able to bake an egg.

 

[TerryLaze]

1000W CPU? Wow. Now you'll actually be able to bake an egg.

Yeah yeah, funny.
But the point of cooling is to keep the CPU below 100° ,for servers probably even lower?! You'll have the same issues backing an egg as on a 200-300W CPU because any part of the CPU and the cooling system has to be below 100° to keep the CPU at below 100°.

 

[Alex/AT]

Yeah yeah, funny.
But the point of cooling is to keep the CPU below 100° ,for servers probably even lower?! You'll have the same issues backing an egg as on a 200-300W CPU because any part of the CPU and the cooling system has to be below 100° to keep the CPU at below 100°.

What I'm pointing at, with 1000W heat spreader really should be equipped with some frying pan or room heater forms to combine the function.

And if to become serious, yeah, maybe direct cooling approach is living its last days and vapor cameras / heatpumps should actually be used to make that difference between CPU heatsink and spreader temps indeed.

[then you'll also be able to bake without making CPU go over 100C]

 

[TerryLaze]

What I'm pointing at, with 1000W heat spreader really should be equipped with some frying pan or room heater forms to combine the function.
...
[then you'll also be able to bake without making CPU go over 100C]

NO PART OF A COOLING SYSTEM CAN GO ABOVE THE TEMP OF THE CPU.
If it goes above that temp then it can't cool anymore.
Any and all cooling is a room heater because where else is the heat supposed to go to.

To bake anything (better) you need a CPU that can go above 100 and use it without any cooling at all so that all of the heat goes into the thing you are baking.

 

[Alex/AT]

NO PART OF A COOLING SYSTEM CAN GO ABOVE THE TEMP OF THE CPU.

Okay man.

 

[jp7189]

NO PART OF A COOLING SYSTEM CAN GO ABOVE THE TEMP OF THE CPU.

This is true of ambient cooling. Heat pumps (TEC's, phase change, etc.) have a hot side and a cold side that don't have to follow that rule.

 

[TerryLaze]

This is true of ambient cooling. Heat pumps (TEC's, phase change, etc.) have a hot side and a cold side that don't have to follow that rule.

Could you get the hot side of an TEC/peltier whatever above 100° with the device working?! How cold would it make the CPU? Would it be able to work?

 

[jp7189]

Could you get the hot side of an TEC/peltier whatever above 100° with the device working?! How cold would it make the CPU? Would it be able to work?

Sure. You can get a TEC that operates up to 200C without getting exotic, and they are typically tuned to run the cold side slightly above expected condensation temp, though they can get far below 0C - even colder than standard phase change coolant. Speaking of phase change, the sky's the limit. You can (in theory) use almost anything for a phase change material as pretty much all matter has phases. It's just a matter (lol) of choosing one with phase temps within the range that's optimal for the application. Practically speaking, R134a is common and cheap and used in most hobbist coolers. It has a range of something like -40 to +60, so not the choice for the fried egg challenge, but I'm sure there are other coolants tuned for higher temp operation.

 

[TerryLaze]

Sure. You can get a TEC that operates up to 200C without getting exotic, and they are typically tuned to run the cold side slightly above expected condensation temp,

No I know that, but what would the temp of the CPU be like with the hot side at 200? At what wattage levels of the CPU could it operate and hold that temp?
Would sticking a frying pan (pretty big mass) to the hot side cool that side down enough to change the temp?
So many questions.

 

[jp7189]

No I know that, but what would the temp of the CPU be like with the hot side at 200? At what wattage levels of the CPU could it operate and hold that temp?
Would sticking a frying pan (pretty big mass) to the hot side cool that side down enough to change the temp?
So many questions.

Max deltaT depends on power input. If you really want to turn a cpu into a frying pan, I'm sure you can stack TECs together to be as cold as you'd like on one side and as hot as your like on the other. It would be a terrible waste of power though.

The pan would only sink until it's saturated. Surface area and air flow would make a big difference which in turn would affect cold side temps. The challenge is doable with enough engineering effort - if utterly impractical.

 

[Rob1C]

A portion of the cooling system can be very hot on the outside while the part touching the cooled section would remain very cold. Example:

Approximate temperatures of two sides of Webb Telescope. The sun facing Hot side of the sunshield ( 185°F / 85°C ) and the deep space facing cold side ( -388°F / -233°C ). Source: NASA, ESA, CSA, STScI https://www.nasa.gov/nasa-brand-center/images-and-media/