A startup’s cooling chip may surge processor power | Hacker News Reader

A startup’s cooling chip may surge processor power

allaboutcircuits.com

114 pointsWaitWaitWha54 comments

An inkjet nozzle for air. Not enough for a 100 W CPU die of that size, but it sounds like an interesting idea.

Dissipation and consumption both seem to scale linearly with area:

The datasheet says the Mini has 5.25 W heat dissipation with 1 W max consumption. Size is 27.5 x 41.5 x 2.8 mm. Area 1141 mm².

The Pro has 10.5 W dissipation at 1.75 W draw. Size is 31.5 x 71.5 x 2.8 mm. Area 2252 mm².

We have passively cooled ultracompact laptops and tables on the market with sustained power dissipation of 10-15W. The AirJet Pro dissipates 10.5W and uses a significant amount of power itself. These things are also HUGE. Unless one can improve the dissipation by a large factor I don't see much practical use for this kind of product.

Suspicously absent in the diagram.. dust and particles.. i guess its a closed inner air circuit transporting the heat towards some large scale shedding surface.

Also, is this something similar to the https://en.wikipedia.org/wiki/Venturi_effect ?

I went through their marketing materials and according to them, the flow is strong enough to allow putting a filter there.

The air is full with fats that will penetrate any filter, and long term build up there... With the extra heat contributing to the issue. Personally I don't think it can work in an open loop system.

Maybe it will come with a sticker saying "spray inlets with alcohol periodically".

..unexplained fires are a matter for the courts.. coolandnarrow.. coolandnarrow..

I'm just really glad its not some Peltier junction solution. It looks like it could be a boon for low power devices that throttle easily such as tablets, maybe even phones and ultra thing laptops. Might even make cooling for non-cpu chips easier.

I feel like these are not as common/popular as they were back in the day. Now it's all water cooling with those all-in-one setups. I always thought the Peltier coolers were cool (pun not originally intended, but now it is).

I was very young, probably as a teenager back in the 90s, when I discovered the existence of Peltier elements for the first time, and that one had been part of one of my former CPU coolers for all the time.

After an upgrade, I had this now unused CPU fan on my desk, and took it apart for whatever reason (because it and the screwdriver were there, I guess). And there I noticed that as part of the assembly there had been... another thin plate sandwiched between cooler and fan, separate and just held in place with the screws, and I now saw that it also had two wires for power that were spliced into the fan's power cable? Optically, this was clearly just a thin, static slate of something, maybe ceramic. Definitely no moving parts, why in the world does that need power.

I curiously applied some voltage, and was amazed that it turned cold! I played around with it a bit, and unwittingly touched the other side of it. Hot, surprisingly hot! It wasn't touching anything, that was just from the air around it!

I was mesmerized, good times.

It would have been far less interesting had you accidentally touched the hot side first. :)

Would it though? I guess I'd have been extremely confused, "why is there a thing that generates heat on my cooler", and hopefully I'd have also touched the cold side at some point, and I'd still have been mesmerized... I think?

My university surplus store was selling a huge batch of peltier elements that were used in cloud chamber research for very cheap. I bought them and connected some to a homemade water cooler loop circa 2006. I could generate temperatures around -20f in my dorm room IIRC. Condensation, and then frost would start to form quickly. It was an interesting project, for sure!

Sadly they’re extremely inefficient when compared to radiators and fans, whether directly on the heat source or across a water loop.

In the olden days of Pentium 3 (or Pentium!!! if you prefer), one guy graphed his Peltier element's performance vs. voltage and found a 115% efficient point (with assistance of his fan, ofc.)

Limiting voltage and coupling it with a very efficient cooler for its time, he overclocked his CPU a great deal.

Looked for the article, but it's possibly gone down with the downfall of old internet, because it was a personal webpage.

The real issue with Peltier junctions is that unless you shed the heat away from the "hot" side very effectively, the "cold" side will heat up even faster. That makes them kind of a non-starter.

See, I don't understand why there aren't strange shaped Peltier cells in order to increase the surface area on the hot side vs. cool side. Then it might be worth it...

> The AirJet datasheets report a power dissipation of 5.25 W and 10.5 W using the AirJet Mini and Pro, respectively.

Ah, beans. Was hoping for something more applicable to desktop processors. Neat idea though.

Edit: I missed the callout about laptops and tablets as I was skimming.

If you are interested in cooling, I suggest that you replace your thermal paste.

In my research I have found that the thermal paste is, paradoxically, the least thermally conducting part of a PC heat pipeline. So, use as little of it as possible, to avoid adding thickness.

Or use one made of a metal alloy:

https://www.youtube.com/watch?v=hdTsra-uLBI

> In my research I have found that the thermal paste is, paradoxically, the least thermally conducting part of a PC heat pipeline.

Well, the second least conducting after what is meant to replace: Air.

And it makes sense, if it were more thermally conducting, then you should just make the entire cooler out of thermal paste.

Indium gallium dries out. It needs annual repasting.

Also, using too much thermal paste is not an issue. What matters is mounting pressure and geometry. Replacing the bendy Independent Loading Mechanism (ILM) with an aftermarket square bracket. If you want to go further you can lap the IHS or even delid for direct-die cooling.

Delid for liquid metal TIM in between the IHS and die made sense for a few years, but Intel has switched backed to solder, so the only gain to be had with delidding these days is with direct die.

I have liquid metal of 5y old and it's just as new - the thing does not 'dry out', but it can be pumped out during thermal expansion and contraction.

Aside from Indium Gallium paste drying out as another post mentioned, it also has significant downsides. A big one is that it's electrically conductive so if it gets on the motherboard or oozes off the cpu, it can short out things on the motherboard. Another one is that the gallium in liquid metal will react with metals like aluminum, copper, and nickel and over the long term might cause bad things to occur.

Gallium will not really react with Nickel, in the worst case it will be tarnished which has no measurable impact on the thermal conductivity. Also the metals don't dry out (as in evaporate), but they can be pumped out.

> ...use as little of it as possible, to avoid adding thickness.

You might be interested in watching this: <https://www.youtube.com/watch?v=EUWVVTY63hc> and maybe also this: <https://www.youtube.com/watch?v=nWu2tcm4wL8>. Do note that the videos contain graphs which summarize the findings.

If this is true, why not have coolers built into the CPU? Like a block of metal attached? Is it a small improvement gained? Or manufacturing nightmare?

neat. something like an integrated heat spreader? that's a pretty good idea

It's called direct die cooling and that's pretty much what happens in the laptops. The reason why the IHS was introduced some 20 years back was the cracking of the silicon while mounting a cooler on top of it was quite easy.

This one does[1]. It have a bunch of pistons on springs touching each chiplet, all in mineral oil. Price bit out of range for typical desktop user tho

* https://www.youtube.com/watch?v=xQ3oJlt4GrI

You could mount the cooler directly on top of the die with the Athlon XP. Turns out people kept damaging the CPU, hence the switch back to heatspreaders.

Permanently attaching a cooler to the CPU itself doesn't really seem like a viable option to me. Mounting it to the motherboard would be a nightmare, and you'd end up creating a one-size-fits-nobody cooling solution. Nobody would be happy with the cooler you end up choosing.

In reality, cooling paste really isn't that big of a deal.

Oof, I remember how terrifying it was to mount one of those Athlon coolers, having heard so many stories about chipping the CPU or cracking a corner off.

Doesn't extra paste just squeeze out the sides?

Yes. This myth needs to die.

The advice in the subsection is just hilarious.

It squeezes out and potentially causes a short-circuit as some pastes are electrically conductive.

Very few modern thermal pastes are conductive.

> So, use as little of it as possible, to avoid adding thickness.

Don't do this. Far far worse than having too much thermal paste is having too little. Assuming you are using non-conductive paste(which most these days are), be liberal with the thermal paste. As long as you stay within reason it pretty much doesn't matter thermally how excessive you apply, especially when applying on a heat spreader.

I don't understand why they are blowing air over the chip - why not use a https://en.wikipedia.org/wiki/Heat_pipe ? They can be made extremely thin and work much better than air.

Maybe even combine the piezoelectric idea with a heat pipe to increase thermal flow.

The diagram is misleading. The use case seems to be to directly vent air from the device which already has a passive cooling design. Drawing that air from near the chip maximizes the temperature delta and thus efficiency.

Not sure how novel it is considering things like https://www.curiejet.com/en/product/micro-pump/air-pump-and-... already exist.

Interesting. I wonder if it's audible when running? Maybe the piezos operate in a range higher than human hearing, or it's just not very loud in the first place?

High frequency also means it's easy to filter out by sheer mass of the system

Considering the outrageous price of macro-scale piezoelectric fans, I hope this is one area where scaling makes it much cheaper.

Hmm. I'd rather have the industry try to reduce the power consumption.

But I guess it would be fine for mobile devices.

That’s one part of the problem..

Why not make cooling / heat dissipation part of the cpu?

You can have horizontal or vertical planes to guide the heat to the top or sides. chips are not 2D anymore, and this might allow for a slightly cooler core/inside temperature

No way this can make economical sense. Every bit of wafer surface has much more value for transistors than improving thermal performance. The current state with a heat conductive plate atop of the silicon is certainly the best tradeoff, since thermal dissipation is the second most significant way to improve ship performance after actual transistors layout and conductivity.

Without thermal performance, your transistors will have to stay mostly powered down ("dark silicon"). At some point, the tradeoff will go the other way.

Sure, for now. But it might create more stable ICs, or might even break some limitations due to heat runining up no the center of the CPU.. if a SoC can suddenly be x times as thick / have more layers, you can have shorter paths to different components. Which is faster again. The area of a wafer would remain the same, but I’m not sure about the economics of silicon wafers (thickness vs area). Less area mean less surface area for imperfections.

This idea or effect has been around for Astro and even 10 years old video camera. What is new ?

For some reason I thought battery life was more constraining than cooling in mobile devices?

Different constraints. Better cooling allows the processor to throttle less and achieve higher average performance, which incidentally would also reduce your battery life if you had the processor running at 100% load the whole time.

Higher performance finishes the race to idle quicker and thus counterintuitively saves battery.

not at all: higher performance equals higher freq, which requires higher voltage. Power scales w/ frequency linearly and the with the voltage squared. Then any active cooling will have its own power consumption.

Best battery/power savings are achieved running as low as possible voltage for as high as possible frequency. At such voltages the CPUs/APUs are less likely to throttle to begin with.

Seems like the kind of tech that Apple would potentially buy a company for. ;)

How many days til this clogs in fire season?

Reply on news.ycombinator.com