How to Cool a 2600 Watt CPU or GPU 🔥🔥🚒

[@Serachja]

a vapor chamber direct on the chip with groves would be fun to see. That way you could spread the heat to bigger surfaces, that could be cooled traditional

[@gamingonthespectrum]

This actually seems like a really good idea. Have the underside of a IHS basically be a vapour chamber... The IHN doesn't even need to be like they are today. They could mushroom out to offer a greater surface area to sink heat from...

[@cornchipzzzz]

Imagine gunking up your KW GPU with contaminants. You'd be so sad

[@RonnieMcNutt666]

the pure physics of something small as an i9 10900k creating as much heat as it can in such a small area (350+ watts) is fascinating

[@nickmudd]

Im just flat out amazed at what phone and tablet SOC's can do with such little power

[@ChimpyChamp]

5.8lpm is incredibly slow, a problem I can see is that the etched cooling fins on the chip are so small that the liquid really needs to be at a very high pressure to get even that flow.

[@tommihommi1]

definitely lots of room for optimization, this was mostly a proof of concept with very simple geometry

[@glenwaldrop8166]

Yeah, no matter what you do at that scale the surface tension is going to create issues with heat transference, the heat density of the CPUs vs everything else we cool (even engines, etc), hot spots creating air bubbles, etc... Pressure and coolant "stickiness" is going to be a big factor. I honestly wonder if in die/on die cooling will work, especially as it is only going to get denser. It may need another layer to diffuse the heat a bit so it can't create air bubbles and block cooling in specific areas. Even with a mirror finish, tiny imperfections give a focal point to generate air bubbles when heating liquid.

[@mart-erikmaeots3518]

It sounds like a lot to me, if you are trying to push 5.8 liters through what is essentially a microfluidic device instead of a liquid cooler like we have today. You might have insane back pressure, which would need better pumps, stronger connectors all around. Sounds expensive(I only know microfluidics though, not PC cooling)

[@kasuha]

I believe the trenches are only to increase surface area and the liquid goes all over the surface of the chip, not just through the trenches.

[@glenwaldrop8166]

@@kasuha yeah, but if the heat is intense enough it'll still cause bubbling and cavitation. 90C is 194F, bubbles would cause enormous problems, runaway heating.

[@wargamingrefugee9065]

The camera is over exposed and the audio is out of sync, but, on this channel, that doesn't really matter; we're here to hear what the man has to say and inform ourselves.

[@TechTechPotato]

Audio should be fine - it is locally?

[@tommihommi1]

@@TechTechPotato can confirm it's fine on my end

[@loafbred]

@@TechTechPotato No audio problem here, but the exposure is hot.

[@karehaqt]

@@TechTechPotato It's ever so slightly out of sync on the audio but not enough to warrant a re-do.

[@wargamingrefugee9065]

@@TechTechPotato The audio is fine now. The sync issue must have been on my end. It persisted after a refresh, so I thought it was the video. Sorry for the confusion.

[@glenwaldrop8166]

Trenches vs grooves vs fins vs tubes... The hotter it gets the more pressure it needs and the larger the spaces will have to be to prevent air pockets from forming at the hottest points. We deal with this in the automotive world and I imagine it won't be long until we have a stickier coolant that raises the boiling temp similar to antifreeze and the various "water wetters" to prevent air pockets and hot spots in CPU cooling. I haven't seen air bubbles form inside an engine but they're also cooled on an entirely different scale than the CPU. I have seen air bubbles form in other situations that make me wonder if liquid cooling can even work at the silicon die level...

[@SebLukaTech]

Prosiphon elite much? :P

[@glenwaldrop8166]

@@SebLukaTech Nice. That doesn't get past the cavitation issues if the hot spots are hot enough on die. In context I was talking specifically about having the coolant on the surface of the silicon or running through tubes/nanotubes/grooves in the silicon itself. Simply having a layer of silicon or metal on it will inhibit the cooling somewhat but may prevent air bubbles *if* the heat generated is intense enough in certain locations.

[@rapsod1911]

That diagram @13:01 shows Relative increase of TDP, if you increase flow to 6 L/m and you use pillars (I assume intersecting trenches you can dissipate 2.6 times more power and hold same temp.

[@EyesOfByes]

13:35 Gaming outdoors 🇸🇪🥶

[@denvera1g1]

I have a pair of Xeon X5690, and at one point i had the pair OCd to 4.2Ghz, IIRC each processor pulled 160w at default, and 220w OC, so 440w just on the CPU(s), add to that the 12 fully buffered DDR3 ECC RAM at 14-22w each and before you even add a drive or video card you could be looking at over 750w

[@SpeedrunnerG55]

i can only imagine what dissolved solids would do to the silicon over time. would the water need to be pure?

[@tommihommi1]

That's why the silicon oxide "TIM" solution is appealing, the oxide acts as a diffusion barrier.

[@SquintyGears]

But water never stays pure in a loop. And at the scale those channels are I'm ready to bet just erosion is a big issue. In regular open loop watercooling the nickel plating (on plated block obviously) gets rubbed off in a a year or two of daily use. It's not much thickness but when the whole package stack is 12 micro meters (for that 3D example) i don't think you can afford to lose much more. Current silicon basically never dies if it is kept in spec. Do we want to move to a solution that makes it a consumable for the sake of performance? I wouldn't think that's an improvement when we know pushing more power through a chip to get more clocks is pretty sharp diminishing returns. It's not possible to push every single component that far to match.

[@FireStormOOO_]

@@SquintyGears I think that's why he was saying this will probably never come to the consumer space. For specialized chips that will be obsolete inside 2-5 years (e.g. performance/watt is no longer competitive with a newer repacement) I'm not sure that'd be a dealbreaker. Those specialized chips aren't hot because of overclocking, there's just that much crammed in there; think like GPUs that run hotter than CPUs even at half the clock speed.

[@glenwaldrop8166]

@@SquintyGears It's like automotive cooling systems though, you make sure the grounding is properly done between the radiator and the engine, try to make the coolant as non-conductive as possible. Most of the oxidation and corrosion tends to be caused by electrolysis, even just the static difference between the engine and body if the grounding isn't adequate, it's still electrolysis, just really, really slow. Eliminate that and 90% of your corrosion issues are gone. Using the same materials, or at least non-reactive materials, further reduces the issue.

[@SquintyGears]

@@glenwaldrop8166 think about the scale though. Engine cooling is done on a massive chunk of metal. Here even without any issues with conductivity (which wouldn't be in question anyway if they don't use waterbased solution) I think the waterlfow would grind away the chip physically.... To keep things in perspective a DDC pump does about 16L/m and TSMC's table goes from 1 to 7L/m. But people usually don't run their pump maxed out and you still see visible errosion on the nickel plating within the lifespan of the PC. I obviously have no way of testing this conjecture but that's a reasonable thing I imagine happening for this concept.

[@Mireaze]

Gimme them 1.21 gigawatt CPUS. Imma compute so hard I can transcend time and space

[@zunriya]

glad to see lucky noob on gskill OC event on this channel

[@TechTechPotato]

When we get travelling again, I'll meet up with Alva one Computex like I usually do, and we'll do a video together

[@wtflolomg]

It comes down to packaging. If AMD and Intel wanted to use direct cooling, they certainly could, on a mass scale. I think it is likely they will, too. I built my first PC... no fans required, on the chips or motherboard. Today, there are easy solutions for people to use all-in-one liquid cooling radiator loops. I can see a day when our desktop chips come with water cooling ports directly in the packaging and we have standard kits (or they are included) to hook them up.

[@unitedfools3493]

Direct cooling doesn't actually work very well - hence it not being used much. You can pump the coolant at high pressure though tiny holes aimed at the die to make it better but at that point it's a lot cheaper to attach a block of some kind.

[@agenericaccount3935]

My concern with it would be the end user. The lowest but widest common denominator doesn’t maintain their loops regularly as is. You think they will maintain something intricate like this? I feel like they wouldn’t, and that would probably steer the decision to implement it en masse.

[@CorDawgYT]

That's sound's like a good idea and all, but then if you have any problem with your cooling, and/or your internal fins are all clogged up because the end user cooling fluid is not chemically perfect, the entire chip ends up having to be scrapped.

[@Veptis]

When I was really young I had a book about modern houses. And how in the future we will have heat and cold access ports in every room. So you can move heat from the freezer, back into a radiator or hair dryer, putting your computer on that loop wasn't sketched out back than. Also using a heat pump outside. If power(=heat) becomes less of a concern, will we just push the frequency higher and higher for performance? Also the efficiency on the smallest end for phones maybe? My next computer will be small form factor and it will have liquid cooling for CPU and GPU.

[@Dahoon]

Not quite there to those degrees but at least here in Denmark it is required by law in any new house/apartment that heat and air gets circulated around in ways that puts damp air into dry rooms, hot air into cold rooms and the other way around. Any home from 2020 and newer all have this. But while plugging your pc or refrigerator into this system would be quite easy it would not make any difference at all. The loss and insulation needs alone would make it unfeasible. A better system would be moving all the heat generating systems into the same space and then move air from there to could rooms. We have done pretty much this in our house.

[@Lauren_C]

For consumer/enthusiast use, I hypothesize that a grooved chip built into a vapor chamber (basically, the vapor chamber would take the place of the traditional heat spreader) would see some significant benefits in temperature without increasing build complexity or maintenance for the end user.

[@SBA_poiko]

You mean basically like the vapour chamber body sorta slots into the silicon? Like wood working joints?

[@alesksander]

@@SBA_poiko Yeah i think that also. Much cheaper to use as sealed soldered heat spreader.

[@spuchoa]

I learn a lot from your videos, thanks and great video Mr Potato.

[@khhnator]

just cool them from both sides, a fan and a heatsink on each side :D twice the cooling! twice the TDP!! because that's how that works right?

[@Zarcondeegrissom]

I keep thinking they have a giant power connection that would be a great thermal connection (esp on op-amps, buffers, VRM power stages, etc), and they're starting to fuss with vertical interconnects that would also work great as thermal vias, then I think of what NVIDIA would do with that on consumer cards and drop the idea out of fear of how hot my house would get, lol. all that heat eventually ends up in the room the PC is in. great if you live in the arctic, not so great if you live in a temperate zone or the tropics.

[@JohnDlugosz]

@@Zarcondeegrissom Yea, I already have that problem with my home office getting toasty. Maybe my next custom case will hook up to a standard dryer vent rather than having exposed fans on top.

[@isaacx593]

Your right, the future solution is to have a cold plate heat sinc on each side of the chip for better cooling. We already going there with these flimsy plates on back of GPUs

[@suntzu1409]

Yeah but how would you supply power and data efficiently to the chip?

[@Zarcondeegrissom]

@@suntzu1409 well, on the PCB the traces go out to the sides anyway, so data on the outer perimiter edges, cooling and power from the top and bottom. common on the top heatsink, Vcore from the bottom, and the VRM can even plug into the back-side of the socket like the old CRAY SV2 processing elements. no it's not 'ideal' for some things, yet for others, it could work. like most things, it's just tradeoffs of one thing for another.

[@robertsneddon731]

Back in the distant past (the 1980s) the discussion about future processors was about the HSG (Hairy Smoking Golfball) CPU package. It would be as small as a golfball to reduce cross-chip signal latency down to a few dozen nanoseconds, it would have lots of hair-thin wires coming out of it for the 32-bit or even (gasp!) 64-bit data and address buses and it would be smoking from the amount of heat it would need to dissipate as it ran at hundreds of MHz, maybe even as much as one GHz! for super-high-end designs. We laughed then. We're squirming uncomfortably on our chairs now.

[@leisterj]

I was at purdue back i think it was 2001 a EE friend was telling me about a cooling competition they were having. And the winner delided the cpu etched the silicon and then put liquid N2. On it. Its interesting to see people now claim that these are new ideas. This class was also the class that had the who can get a charcoal grill up to temp the fastest. And the winner used powdered charcoal and liquid O2. I thi k there's a youtube video on that. Kids used to be able to have fun.

[@DigitalJedi]

Homie melted the grill. Jesus. I remember getting to watch LO2 go to town with some rocket-science demos once. That stuff is no joke with a powdered fuel.

[@JeoshuaCollins]

For reference, that graph at 12:30 seemed to imply almost 3x TDP at 6 LPM. The "1x" point is at 2 LPM trenches, which if you check the previous graph at 11:11, was somewhere around 2 KW. If that is to be taken at face value, that's nearly 6 KW TDP!

[@nathanlowery1141]

Somebody has been overclocking a reactor?

[@JeoshuaCollins]

@@nathanlowery1141 Kind of low for a reactor, but that is literally the wattage of a fairly sizeable solar panel crammed into a computer chip.

[@anonymous_coward]

The max power you should draw off a typical surge protector is 1800 watts. So 1800 watts - monitor power = Max PC Power.

[@daemontus]

Depends... a lot of modern European apartments come with surge protectors somewhere in the 3-3.5kW range (as far as I know), and houses can sometimes go into the 5-6kW range (especially if the design explicitly includes something like a shed with a workshop). Anyway, with the rate at which car chargers are becoming commonplace, we might be gaming in our garages soon enough.

[@richardyao9012]

I have been wondering when Intel would introduce the first 1kW CPU for years. I imagine someone else might beat them to it, but I still am hopeful for Intel.

[@niwasox3]

They'll still label them 95W TDP, though.

[@RonnieMcNutt666]

if you consider the 56 core one a single cpu, that one and the w-3175 both go well over 1kw if overclocked heavily lol

[@RonnieMcNutt666]

intel did some silly 5ghz 28 core overclock on the w3175 using 1500 watts i think

[@LordZordid]

Great video. Thanks. Your ending statement was pure gold.

[@Michael-OBrien]

Okay, but what about more advanced applications of existing tech, such as vapor chambers. They excel at heat spreading, i.e. high power density use cases benefit the most. Any way we can implement a vapor chamber directly onto a substrate with the bottom of the chamber being the die and substrate?

[@denvera1g1]

I'm waiting for Novec to become afordable, i'd love to have a completely passively cooled computer with phase change immersion, and then a massive, passive condenser on the top/back of the case

[@PainterVierax]

I also think phase shift liquid cooling is more convenient for such an application. After all this is the convergent evolution of copper heatpipe tech and good old oil immersion cooling.

[@denvera1g1]

@@PainterVierax I think the.... ice giant? uses phase change fluid in this way, which is a much more economical use of a fuild that last time i checked costs ~$3000 for a 5 gallon bucket, which i'd probably need half of that for a regular ATX build with enough room to put the GPU and CPU blocks on display

[@PainterVierax]

@@denvera1g1 I don't know about pricing and budget but from what I've read from experiments on water, mineral oil, novec or even helium and hydrogen cooling, fluid replacement or refilling is always needed, as it ages or it leaks even in a sealed containment/circuit.

[@denvera1g1]

@@PainterVierax IIRC helium, cannot be contained completely because the atom is so small it literally slips between the atoms of the container, not just thrrough the seals, but every part

[@SirCrest]

I absolutely love these types of videos.

[@JohnDlugosz]

I remember when I got a Pentium Pro, something that was unusual for consumers. My friends online teased me about the CPU taking 40W of power and how I could use an extra drive bay as a pizza warmer. Today that's still in the "Mobile" category!

[@DigitalJedi]

High end mobile chips blow right by 40W. I regularly see 65W going to my 9th gen i7, and I hear the 10th gen ones are pretty toasty as well.

[@tommihommi1]

Isn't IBM pushing some insane power numbers in their mainframe CPUs?

[@TechTechPotato]

400W ish

[@tommihommi1]

@@TechTechPotato huh, I expected more, seems like everyone in the industry who doing normal-ish hardware is in this ballpark at the moment

[@TechTechPotato]

@@tommihommi1 I asked Lisa Su about this once. She said it was easier to draw the mega GPU power in GPUs because of dense repeated units. With CPUs it's harder to get everything working all the time, even when cranked with voltage and frequency. She didn't say the word cranked, that's my approximation :)

[@tommihommi1]

@@TechTechPotato that's why you put 64 cores into the CPU :)

[@Zarcondeegrissom]

my two farthings worth. there is also a point where more volts for a smidge more MHz gets you diminishing returns vs allocating those VRM watts to a lower clocked FPU/ALU for another thread per core. becomes a slightly different game with supercomputers running massive-parallel/massive-vector simulations vs a low-threaded gaming CPU/GPU. you can always cut the atmosphere up into smaller cubes to run it with more threads on a bigger machine, it's kind of difficult to do that with the keyboard/mouse input driver, lol.

[@plonk420]

that "28 cores" part of the thumb reminded me that i wonder what a 32-40 E-core (or 8-10 * 4-E core modules) ADL CPU would look like in the way of performance, considering Intel is boasting of "Skylake-like performance" xD

[@TechTechPotato]

I'll take two.

[@MeriaDuck]

9000 what!?

[@organichand-pickedfree-ran1463]

It’s a dragon ball reference. The Sajajin power measure from their smart glasses

[@goodyKoeln]

Over 9000 watt?! (Fixed it for you)

[@william_SMMA]

Good one 🤣

[@khhnator]

kzbin.info/www/bejne/inPXoaybgd2Joas back to a time when the memes were pure and for the lulz

[@YounesLayachi]

@@organichand-pickedfree-ran1463 This isn't a Dragonball reference though, this reference can only be found on the English adaptation of the anime adaptation of Dragonball Z *saiyajin *scouter

[@powertoker5000]

Single stage is impractical and causes condensation. My P6T / W3570 / OCZ Cryo-Z were notoriously hard to keep running, cryo z was made for sub 130W CPUs so the Xeon gave it a rough time.

[@NUCLEARARMAMENT]

If you have a chillbox/"purge box" you can remove all the humidity from the sealed case and eliminate the possibility of condensation. There's also water/glycol mix that allows chillers to get below negative temperatures without freezing in the loop. It might prove more flexible than direct-die phase change cooling, if at a slight loss of efficiency and increase in insulation efforts.

[@bmgjet]

Posting from my rig which under Benchmarks uses 580W CPU (10980XE @ 5.1ghz) and 600W GPU (3090 with 1KW XOC Bios). Water cooled obviously.

[@Flojer0]

I like where nerding out can take this, my only fear is that chiplet CPUs like threadripper as they exist today aren't far from tripping the electricity I have. If something pulls more clock speed out, maybe a lower process to pump more power through as well, then things could get crazy at home.

[@daguerref1]

I want my hot water heated by a massive computation

[@jrherita]

Good video! Would like to hear more about cooling limits with shrinking and stacked transistors.. thermal density seems like a long term problem.

[@Zarcondeegrissom]

not sure why this makes me thing of a cover I wrote years ago, lol. ZDG - World of High-end game-'n'. (Inspired by ZZ Top, World Of Swirl) yea. Yea! I got new graphics card, had a Game in mind. It's look-'n' for a kill-a-watt, and the PSU died. It's getting a little crazy, as the case is turn-'n' red. Got to find a cooler for the inside outside. Smoke-'n', from the melting fans. Screaming, from the Power-supply. In the world of High-end game-'n'. Everything is smoldering, ten feet from the case. Isn't it amazing, how it glows without any lights. It's so unpredictable, when the breaker will explode. Got-a escape the flames, coming out the side window. Smoke-'n', from the melting fans. Screaming, from the Power-supply. In the world of High-end game-'n'. (Oh $#!+, the extinguisher is empty again) Ramming in more cards, you gotta keep on gaming. Heaping on that halon, just to keep it running. Beyond thermal, exceeding plasma. Ready for fusion if it had the pressure. Tumbling, between Crossfire and SLI. conflagration, inside the computer. In the world of High-end game-'n'.

[@boneappletee6416]

Thank you for the cat tax at 15:00 :)

[@okkomp]

In Finland the fins for water cooling are awesomer

[@tobiwonkanogy2975]

my 3200g is currently reporting 495w on package and 1277177.0 volts on gpu . definitely a bug but funny timing

[@TechTechPotato]

got a screenshot?

[@Zarcondeegrissom]

don't give GPU makers ideas about 2.6 killa-watt 'gaming' chips, they will make them and we will all suffer the consequences, lol. my minimum spec, a PC that doesn't need a 40-foot roof-top chiller just to keep my house from melting it's way down to the earths core, lol.

[@JohnDlugosz]

Not a roof-top chiller; rather, it doubles as the hot water heater for the home.

[@deilusi]

So its friction welded onto the chip. NICE. I would still use heatpipe instead of groves and direct liquid cooling take that 300mm of area cooling should do better than 80mm, even if heat transfer will be lower in case of high area solution. I am not fine with conductive fluid going so close to such power. I can get it its, the most hardcore option, but for me heatpipes replacing IHS, and expanding it to cover whole chip or even more, is where I will draw my line.

[@nickglazzard2385]

How about total immersion cooling using Freon (or similar). Probably not … but the Cray 3 managed to dissipate 88kW in a 405 cubic inch package … which sounds impressive at first sight, anyway. That used unpackaged Gallium Arsenide chips (naked on PC boards).

[@countach27]

Can’t wait for a 5kW RTX 10900 Ti that needs a three phase connection

[@Steamrick]

I wonder when we'll get the first CPU with G1/4 threading cut right into the heat spreader?

[@tripzero0]

Also, condensation.

[@Joemama555]

i wanna see a phase change heat pipe where the back of the Si is exposed to the refrigerant..

[@vcjester]

This would be tough to implement with mainstream machines, because I've wandered the intarwebz enough, to develop a theory that people don't know how to keep organics out of a water loop.

[@Johnrich395]

Thank you for answering a very nagging question about if you could directly push coolant over the die. I was calling it “Direct impingement cooling” but you have answered the question.

[@orlovskyconsultinggbr2849]

My dream one day to use a complete made from gold heatsink, yes it would be expensive , but gold is the most heat absorption material.

[@whoknows7513]

Nice explanation. Thanks

[@sandwich2473]

Absolutely remarkable!

[@vanhetgoor]

What if you would start with a block of solid coper, make holes in it and let water run trough this. Then on that solid coper block the silicon is placed on whitch the actual chip is working. Then on top of that another solid coper block is sandwiched. The other possibility is switching over to ARM.

[@aroncanapa5796]

This channel is so much better than any college course

[@Nobe_Oddy]

this is AMAZING!!! I think if this does come to fruition then the water entering and exiting the chip should be ON THE REAR OF THE MOTHER BOARD and a waterproof coating applied to the back of the waterboard. And this should be a standard if this comes to the home desktop PC's. My reasoning is that regular people (non 'enthusiast" users) would most likely have no idea how careful they need to be as to ned get the computer wet and to COMPLETELY DRY IT before trying to run the machine. PLUS we know those 'BIG NAME" PC manufacturers make there products as CHEAPLY AS POSSIBLE and their "garbage" likes to easily break over time. So WHEN IT DOES, the water would only be able to get onto the back of the MoBo and it would have a nice heafty coating so there is essentially no risk of a short. PLUS throughout the ENTIRE HISTORY of the PC we have seen only a SMALL HANDFUL of manufacturers utilize the rear of the MoBo. Adding an extra M.2 slot is really the ONLY thing that comes to mind, besides that one MoBo that was features on LTT where the processor was on the rear of the board so it could attach the ENORMOUS heatsink that was 100% passive (with no fan) that was TOO BIG to put on the front of the board because it would be in the way of EVERYTHING that goes on the front. Other than these two examples, I really can't think of anything else that has been done on the rear of DESKTOP Motherboards (sbc's and specialist boards are not part of this convo as they are NBOT considered DESKTOP PC MOTHERBOARDS) - AM I WRONG ABOUT THIS??? DO ANY OF YOU KNOW O F THE REAR OF THE MoBO BEING USED FOR SOMETHING??? PLEASE LEAVE A REPLY ABOUT IT. So using the REAR of the board is a GREAT PLACE for water cooling for the regular consumer because they wont accidently bump the pipes or tubes if they upgrade their RAM or use an ADD-ON card or something... they might think (hey this dumb 'bar' is making it hard to add my new sound card... it's probably there to protect the computer from shipping bumps or being kicked while under my desk... and those twisty rings MUST mean it's meant to be detached so you can put more cards in so here goes nothing... *SPLASH!!!*) This would be MUCH MORE FREQUENT if processor will REQUIRE water cooling in the future. SOOOOOoooo Dr. Cutress, if you ever find yourself in a position to make some suggestions to anyone about water cooling the chips directly, PLEASE FEEL FREE to use this idea if you find it useful. It wouldn't take much for manufacturers to FLIP THE PROCESSOR OVER TO FACE THE REAR OF THE BOARD and this OPEN A WHOLE WORLD OF DESIGN for PC builders. PLEASE DO SHARE THIS WITH ANYONE YOU CAN. (that is if you like the idea) It might make a difference in the world. :) 💗💗(no homo lol)

[@clocktower1164]

Instead of DWC (Direct Water Cooling) they can 1. Pump liquid metal through the troughs to pull out even more heat than plain water and, 2. Chill the heck out of the liquid metal (water can get frozen at 0°c but liquid metal remains in liquid form at 90 Kelvin or -183.150°c).

[@srpenguinbr]

Amd remove air humidity from the environment to prevent condensation

[@Speak_Out_and_Remove_All_Doubt]

Loved this video! I really want to make a PC submerged in 3M Novec fluid but crazy expensive. We are already needing better water cooling options for a large portion of the enthusiast market as most of Intel's chips are currently heat limited and I can only see 3D chips making this an even more pressing issue.

[@Mr.Morden]

It'd be interesting to see if this could work with heatpipes fused to the silicon instead. I'm sure they could work up a way to deposit a capillary structure inside the channels. It'd cost less than water cooling and be more reliable because there's no need for a pump.

[@TechTechPotato]

Well, the fusing is part of the problem. It adds thermal resistance to the system. Fusing to the silicon still creates a non uniform barrier

[@ragincaveman]

"At 280W you're getting close to the limits of what air can do" - ICEGIANT ProSiphon: hold my beer

[@YounesLayachi]

Actually the quote is 8000 but in the English adaptation of the anime they made a mistake and said 9000 instead of 8000

[@kendokaaa]

If we go even higher, it starts becoming a problem for the space containing the PC. Pretty hard to cool a room when there's a 1kW space heater running

[@reconciliation86]

I'm actually letting my 2080 Ti run Furmark rn because the heating built into my place is not enough. I for one... :)

[@JohnDlugosz]

I'm thinking that the case ought to use the ventilation commonly sold for dryers. You can buy these and DIY kits to install on the side of your house already. So, build a case that has the fans on the radiator exhaust into a manifold with a standard connector as used for clothes dryers, rather than being exposed on the side of the case.

[@DigitalJedi]

To answer the intro question: My i7 9750H is rated at 45W. I have regularly seen 65 for longer than the longest boost duration could be. My 1660ti is a desktop SKU, and runs a mild overclock and pulls up to 140W in bursts, but sits around 90W for most of it.

[@Mranshumansinghr]

Your Table is a bit Wobbly. Your information is solid.

[@kayakMike1000]

Yeah, there's a limit. If you dump energy into the silicone faster than 148 W/m K .... The chip will eventually burn out. But that's a bit silly to say because you're not dumping energy into the whole chip, but just parts of the chip to decode instructions and do ALU calculations, maybe flip things between the registers and L1 cache, branch prediction, that sorta thing... So it's safe to say there's a maximum amount of power tiny sections of the chip can work at... I don't think silicone CPUs can get much faster than 10GHz... However, using CARBON in the diamond structure could see absolutely insane speeds well beyond 10GHz.

[@pweddy1]

You should talk about Cray water cooling. They submerged entire system boards in a non conductive fluid!

[@TechTechPotato]

I've been writing about that (it's called multi-phase immersion cooling) for years. There are companies dedicated to it in the datacenter. But super high upfront costs and lots of components have to be redesigned.

[@BRUXXUS]

I’m surprised you didn’t touch on the AMD patent from a while back which used on/in silicon Peltier/TEC between stacked layers to quickly move heat around and out. Don’t know if they’ll actually ever use it, but a neat idea.

[@shinokami007]

it was horribly inefficient.

[@BRUXXUS]

@@shinokami007 that specific patent/paper? From what I remember it was way better than normal TEC.

[@jeanmacdobea2614]

max-preformence-chips without the need for air-coolers or liquid-coolers is what i wish for ..

[@DigitalJedi]

So a GaN Threadripper or what? Those things can get cozy at upwards of 200C iirc. Should be fine with a passive heatsink on there provided it doesn't desolder itself from the inside out.

[@TheHalo294]

What power does my cpu draw? About 20W idle, ~55W low demanding gaming, ~90W high demanding gaming, 100W rendering (set as ppt) ;)

[@wile123456]

Over 9000 is such an old meme it is now just a cultural quirk people mention every now and then. It's not even a punchline or a joke most of the time, it's just a number with cultural value.

[@lucysluckyday]

Given heat naturally rises, it makes the concept of flip chips seem counterintuitive since the heat is being trapped underneath. Seems like the conductivity components should still be on the upper outside surface (classic bonded style) - I do wonder if flip-chipping was more about protecting IP than about removing wire bonds. Maybe we should return to the classic face-upward chip design, then to achieve stacking we add draft-cavities with ceiling layers held up by conductive pillars/bus (producing inner spaces like in underground crystal caves) between layers with top-to-outside venting tubes for the rising heat from lower caverns. You could probably run coolant "rivers" through the caverns if the conductors have all been sealed with a protective (oxide?) layer. LOL, I'm just visualizing the concept! If you could coat the cavern walls with glass then I wonder if you could flow mercury through them, and would mercury flow friction free (they say mercury switches never wear out - but not sure how long "never" is in that context).

[@Ormaaj]

"heat rising" usually refers to differential buoyancy in fluids / gases whose density changes with temperature causing convective motion. It's not applicable to thermal conduction within solids and not really a significant effect with actively pumped liquId here.

[@NomenNescio99]

It would have been a great comedy cut if there had been an angry cat scream and some more crashing sounds directly after he threw the first cpu cooler!

[@esra_erimez]

der8auer has been direct die cooling for a while. and, he has a delidding tool.

[@tobiwonkanogy2975]

silicon rust is better than all interfaces . soon maybe for data center and then a few years for enthusiast

[@kayakMike1000]

Limited by the amount of energy that would vaporize silicone faster than it can dissipate that energy.

[@Steamrick]

What kind of pump did TSMC use? 2LPM (120 L/h) is on the high end for a PC custom water cooling loop as is and makes me question what kind of pressures are necessary to force 354 L/h through those tiny channels cut into the silicon.

[@TechTechPotato]

Laing D5, a favorite for home liquid cooling, does 1500L/h.

[@Steamrick]

@@TechTechPotato That's a very theoretical value. I have a Laing D5 and it used to top out at 160 L/h in my admittedly rather large loop. Then I added a bunch of quick disconnects for ease of maintenance and now I max out at 100 L/h, though I throttle it into the 60s for noise reasons. 1500 L/h is what you get when you have a pump, a reservoir and a measuring device with a minimum of tubing in between them.

[@jtrevathan33]

But why do we need to cool CPU's? Is it just coincidence that silicon stops working at a few dozen degrees above room temp? Could we make a CPU that is perfectly happy at 500C, thus allowing for a much larger temperature gradient to help with heat dissipation?

[@gerryjamesedwards1227]

Not as long as materials such as solder, and gold wires are used in the chips, and even then the silicon structure itself degrades with high voltage and high temperature. I'm by no means sure of the relationship between the two, but I would guess that as temps go higher, less voltage is required to begin degrading the structure. There are lots of metallic layers embedded in the silicon, too, and those will melt way before the silicon does. If they can come up with a way of making the conductors out of graphene, the temperature resistance will go up hugely

[@victoralander1398]

en.m.wikipedia.org/wiki/Landauer%27s_principle The energy needed to flip a bit is directly proportional to temprature, or you have to shout to be heard in a crowd (more energy) to be heard (represent a bit) in noisy(hot) environments making the area lowder (hotter) so others have to shout to.

[@tommihommi1]

Thermal Noise is one of the fundamental principles of the universe

[@tuhaggis]

It gets exponentially mode difficult to control things at the quantum level when you increase the energy in the system (every 10 degrees celcius/kelvin doubles the rate of reaction), plus the electrical resistance increases with temperature so you get less and less efficient at the same workload as your temp goes up, plus you also have to now insulate your chip from the neighboring components and you also need to now think about how different parts of your system may expand at different rates based on the large differences in temperature between them.

[@khhnator]

I'm no specialist, but i wonder what thermal expansion might do to semiconductors since they are basically very thin layers upon layers of different materials over silicon

[@Mobile_Dom]

next video, ian sends a chip to a milling co to use a 0.1mm milling bit to cut micro grooves in a 3700X to do direct die liquid cooling

[@trueriver1950]

Lots of cpus on a motherboard seems a better solution to me. We used to do that with single cored 486 so why not have dozens of 16core cpus? We would need to keep related threads on the same chip, but affinity coding can do that already

[@utubekullanicisi]

How long until this technology is available on consumer CPUs/GPUs and helps OC teams break the 10GHz barrier with liquid helium?

[@P4GrAnGeR]

I always thought the visible area of a exposed die was Silicone filler, not silicon, being silicon just what's at the bottom behind all the gunk. This information says very much otherwise

[@Fergunator]

Vegeta, what is his power level?

[@paulvanmeveren7488]

9000 watts at 120 volts in number US would require 75 amp wire which would be number 3 wire So this and above has be server-grade.

[@FinneousPJ1]

So do you imagine this in the server space any time soon? And then trickle down to consumer?

[@josuad6890]

a DWC cooling solution would probably need you to buy the chip packaged along with you cooler, right? I mean, there's no way you can mount a cooler by yourself since it's directly on the silicon, and it's rather fragile. and because of that, is it safe to assume a DWC chip is simply not feasible on consumer market?

[@MichailAgustusSolomonic]

What about Heavy Water and pool case?

[@philosoaper]

I don't suppose immersion cooling fluids like novec 7000 is quite destined for home use yet then.. I was just wondering what the size of the molecules in that is compared to water..

[@hariranormal5584]

4:23 what's that music haha

[@Johnrich395]

Thought: how small can you make heat pipes? Is it possible to use that same technology to have in die cooling (or at least faster conduction away from the heat source)?

[@ПётрБ-с2ц]

I wonder why do they not normalise the temperature delta (I assume it's water to heat source).

[@allenshepard7992]

Dumb question - why only cool one side ? Any numbers on thermal expansion or stress where one side is cold and the other side is hot. I've seen the demo of using Freon directly on the chips.

[@LordOfNihil]

look at all the kitties!

[@TechTechPotato]

Lana'i Cat Sanctuary, Hawaii. It's worth a visit

[@LordOfNihil]

@@TechTechPotato my desk is a cat sanctuary. i got 2 up here now. but if i ever find myself in hawaii il go take a look.

[@guanyu210379]

Sooner than you think, we would start seeing 1000W PSU to be the minimum needed for mid-classed gaming PCs.

[@davidgunther8428]

If we can get high band- gap logic chips, then they could run at higher temperature and be easier to cool. 🤓 200C? Sounds preachy! Also, you could heat usable water or steam with those temperatures and reclaim some energy. I like how these chips with integrated water cooling will probably be like a nuclear power plant, with inner and outer cooling loops. 😄 The water in the micro channels needs to be very clean, no pastel loop pigments! 😛

[@gamingonthespectrum]

Have you done a video on DOJO?

[@pegasusted2504]

I have generally wanted more performance out of my machine but I am NOT going to freeze my nuts off just to get that lol, guess it is just water cooling then and not moving to the arctic circle ;~) I am also wondering if they will ever incorporate graphene into the heat removal process.

[@Sipu79]

I saw an article once about a server warehouse where they had racks and racks of pools with the entire motherboards submersed in heat conductive (but not electrically conductive) liquid and the heat would just boil off the chips, obviously that's not practical for home use

[@Sipu79]

kzbin.info/www/bejne/j6queo2Gqsqjopo

[@TechTechPotato]

I've written about those and met a number of companies doing it :) It's pretty cool (pun intended), but base cost is +++++++++

[@Sipu79]

@@TechTechPotato do you have any numbers on what kind of heat those liquid pools can handle, in terms of comparable wattage to the rest of the solutions mentioned in your video?

[@quantuminfinity4260]

Is it just me, or is the sound ever so slightly out of sink. Just a little bit.

[@jihadjoe]

Please do a video on the wafer-scale stuff!

[@wskinnyodden]

If you have free electricity and simply do not care for power efficiency you can also use Peltier's...

[@bgtubber]

I'd love to see a 32-core Threadripper on 5nm running at 5.0 Ghz all-core daily with this new direct water cooling tech.