This New Technology will Keep Moore’s Law Going!
ฝัง
- เผยแพร่เมื่อ 8 ก.ย. 2024
- In this video I explain the Cooling Technologies of the Future.
Check out HotChips Conference ➜ hotchips.org/a...
The video I mentioned ➜ • New Disruptive Microch...
Support me at Patreon ➜ / anastasiintech
Let's connect on LinkedIn ➜ / anastasiintech
My Deep In Tech Newsletter ➜ anastasiintech...
More about me: anastasiintech...
Let me know what you think! 🙂
Absolutely terrible pun. Love it.
cool stuff
You stated that silicon is more conductive than copper (6:53), which is not true. I otherwise loved the video and the puns!
Are there risks of any buildup that can block those channels the water flows through? Or any kind of corrosion?
I think you made a mistake at around 6:58 when you say that Copper is 4 times less heat conductive than Silicon... its the opposite.
The fan club talks about how cool they are, but they're just blowing hot air.
Are you expecting the chips to tank?
the fan club joke was marvelous
She is the president of the CPU fan club.🤣
I still don't understand why this channel doesn't have more subscribers. It's a great cutting edge tech channel.
Part of what makes it good is that every damn video doesn't have a thumbnail with her mouth hanging open and a title like "new cooling tech supercharges your computer."
Also, I suspect many people either don't want this level of detail or don't want the intro explaining it all. It's great content but it's a bit lengthy if you know the background (that's fine for me but that's a consideration).
I don't watch a ton of CPU content but I suspect those people form a majority of subscribers to CPU channels.
@@petergerdes1094 On point
I already have a big subscription list and don't want to add new channels to it (so i made an extension to add channels to popup list)
@@petergerdes1094 It's perfect. We want IQ 115s to trickle down knowledge from here, not IQ 94's to dumb-down a rich information source. She's doing a great job on the balance-beam where she's at.
190k IS A LOT!
You should have been a Bond girl scientist during the cold war Bond movies
00:49 This cooling episode was so cool that the whole club chilled xD
Here is one of the best technology channels that there is you are clear you always have the best and most innovative tech and I really appreciate these videos
Her energy with cpus is contagious !!!!
@@FULLTILTSWIFF Sorry, I don't believe in fairy tales ...
loosening up in your style there Anastasi, telling jokes, laughing, having fun.......i like it
Indeed. She seems in a really good mood
Yes great
Audio really good this time, every word nice and clear with no audio adjustment 👌
That's my kind of humor! My wife tells me I'm weird, but I believe that dad jokes are a rite of passage in life.
Its fact! Don't let anyone tell you otherwise.
@@Zucr_ If you're not weird you're just consuming energy without putting new information vectors into the collective space.
@@freecivweb4160 My machinations that envelop the vectors for which I possess don't incessantly need to be superimposed onto the collective space.
@@freecivweb4160 The machinations that envelop the vectors for which I possess don't incessantly need to be superimposed onto the collective space.
@@Zucr_ I feel like my membership in the collective space just got superimposed upon by your response vector.
As a simple homelab guy, wondering how we can harness this energy for cooling and heating? I live in Michigan, a state getting a ton of new love from Data Centers due to the amount of fresh water we have. I can see a data center in Detroit whose excess heat is delivered to low income households, to large apartment structures. Heat Pumps and Enterprise IT need to have a meeting.
They're already doing exactly this with LUMI in Finland - it's the fastest supercomputer in Europe and it provides 20% of the district heating to the city of Kajaani
@@felinefireaudiodistrict heating is underrated
At 6:52 "four times less conductive than silicon" ? A chilling statement, upended my understanding of materials. 🙃
Indeed 😅
@@biggityboggityboo8775 You must be from a large family ! 🖖
its actually 4 times MORE conductive...
@@WacKEDmaN I know, and I see some already pointed the blup out.
Pressure and flow rate come to mind as the biggest challenges with this. Super tiny cooling channels could also be super easy to destroy with pressure from the coolant.
Cooling implies heat which implies energy waste an inefficiency from obtaining power via higher energy use.
So the interesting part of the thesis that cooling = Moore's law going ahead, is the assumption of energy as not being a scarce resource, and the only problem is how to disperse the heat.
What this really represents is more-than-linear increase in entropic heat consumption, or "chaos animals chewing up orderly stored energy and dispersing it into heat and chaos." This definitively puts a ceiling on Moore's Law but we seem to keep breaking through the ceilings just in time to keep it going. But what's also still going is the higher-than-linear (not exponential) increase in entropic heat-generation order-destroying chaos consumption.
As a philosopher I know these are new ideas to you you're just the right person to spend 22 seconds in a video comment to kickstart the idea.
I think the option of having phase change cooling just build into the CPU is a good option, so like a vapour chamber, but built into the CPU with the capillaries going right down into the CPU, then you shouldn't even need water cooling because you will be able to transport heat to air cooled heat sinks so efficiently it's no biggy, in cases were you want to have a heat spreader you can integrate the condenser into the heat spreader.
Like in a data centre you would still cool the air with water because it means you can efficiently transport heat outside the building meaning you can pack thermal density more tightly, but going with air cooling within each machine isn't about being cheaper for the hardware it's about the ongoing maintenance being more expensive over a 3 - 5 year period that is the big deal, the cost of the hardware compared to that is a none concern.
Anastasi getting more cooling halfway through the video.
I noticed that too
Encapsulation of a phase change material in microfluidics directly in the silicon perhaps in place of the through silicon wires might be an interesting approach to moving heat out from the center of thicker vertically stacked chiplets to the heat spreader without as much need for radical changes to the rest of the current cooling setup.
Subscribed, and i have told so many people i know who operate at this level of intelligence about your work. Thanks so much for all i have learned from you.
Thank you!
I found this video very captivating and I am utterly impressed!
this type of liquid cooling has been seen in nuclear reactor technology and in computers and science-fiction. It's quite exciting to finally see this coming to reality.
You’re adorable. Love how you cracked yourself up with your own joke. That was awesome. Hope you do that more often.
This is why engine blocks in cars have so many holes for surface area cooling. So interesting.
Ha, I was worrying about cooling in a previous video of yours. This feels like it was made for me. Thanks!
I literally had this coolant idea at the start of your video. It's easy to have a good idea, much harder to implement, even harder to implement at scale and make cost effective.
Great video Anastasi, it's interesting that cooling technologies could be the neccesary breakthrough we all need to advance in chips, not the chips architecture, not software layer, even ai implementation, but cooling. Temperature is computation final challenge.
I think that's clear for a very, very long time. Probably Seymour Cray was the last systems designer who was advocating the principle of more compute power by consuming more electric power. To give a slightly more realworld example than Cray supercomputers, BIT Technologies did develop the MIPS R6000 processor in about 1988. It ran at for the time insane 80 Mhz. A 3 CPU R6000 system did consume about 3.6 kW or 16A @ 220 V. The crazy heat prevented higher integration but higher integration is required for higher clocks, larger caches, FPUs and specialized accelerators. Aside of an even more successless attempt at building an ECL PowerPC that meant CMOS won the battle. And within the realm of CMOS the Pentium 4 was one of the lastt speed daemon designs. It's not that lower clocks are iinherently superior but all these designs since at least the 80s in the end werre dominated by how one can get the heat out of the chip.
And once the heat has been pushed out from the system the next challenge is to get it out of the building. I know data centres built in former cold war underground bunkers were any change to the building in order to improve cooling is extremely expensive and the operators have opted to rather leave some floorspace unused. That difficult cooling can become!
Battery powered devices are at the other end. Cooling there tends to be even more limiting (want your phone to burn your hand?) as is the energy supply.
Whatever, software is also in charge to optimize power consumption, optimize cooling. Challenging times but I like it 🙂
Thanks for the vid, Anastsi 🙂
Fluorinert is the trademarked brand name for the line of electronics coolant liquids, used on the Cray 2 in 1985.
I love the joke and seeing you laugh. My wife and I love your channel. Take care.
That wafer scale chip taking 15-25kW is crazy. For comparison, that converts to 20.1-33.5 horsepower which is about how much power it takes to drive a car at highway speeds.
I've been water cooling most of my technology for many years and i see a few issues with some of the things mentioned. Using deionized water (0PPM) will over time leach metals and other material in the loop leading to corrosion. Also cleaning water blocks is a pain in the butt, i can't see how cleaning the microchannels on a chip is going to work out.
What some people think is impossible may take a little longer to solve. No one fully understands an atom, but we can still use it to our advantage.
Years ago I worked for a silicon wafer manufacture as an electrical maintenance tech. It was one of the most interesting, challenging and rewarding jobs in my entire career history. The stuff you cover is fascinating to see where the industry is headed for the future. Cooling at a transistor level , they would of laughed in the past.
11:21 with closed captions is crazy
Ha. Loving all the dad humor in the comments. I didn't know most of this stuff about our cooling technology and it's so amazing! Things are going to begin getting very weird. "begin". I, for one, am here for it. Great video well written and edited. I'll be checking out your others.
Your attention to video editing and synchronization with the audio is next level ...
08:29 So here it is, how strange, i was thinking the same like Transistor level cooling Today before you posted this video, idk how this idea only got into my brain today 🤔
With 1 kW GPU, your computer is a kettle. You played a game for 10 minutes, you brewed 4 mug of coffee or tea.
The performance of the biological compute of our brains is so impressive, we think up all of these advances using only a fraction of the power our inventions need. We then cool our brans with blood flow. The more one thinks about humans the more impressive we are. Great to see how the leading players are planning to cool their inventions. Thank you for sharing!
A fantastic and informative video
Thank you
Make sure to give the likes folks she works hard and deserves them
Awesome and informative video yet again, Anastasia! Thank you
Wow! The information density of your content is amazing! So many mega interesting current and historical developments in chip cooling delivered superbly in just a little over 19 minutes. Super great!
Actually thinking of joining your channel because i think its great.
Great presentation. So good I will watch it twice!
The fan club joke made me hit the like button 😂
Subscribed. I thought I was for months now but your prompt made me double check. Oops 😂 I'm that guy.
Your jokes are soo cool as well, you have your own fan club either! 😀I loved that joke 🤣🤣🤣
Thank you
I'm an electronic engineer and I like the way explain it
Thank you
You are very beautiful and super smart 🤓
Thanks Anastasia..
You're amazing. Very good content, deserving of far more subscribers..
Keep it up 👍
Paragraf is a company making inroads into graphene based semiconductors (not at chip level yet but certainly FETs; potential for better heat management and performance density using the same manufacturing process.
Like you, the video was perfect. Thanks.
Wow that's amazing 😻🤩
So Much Love From Nepal 🇳🇵🇳🇵
This EXACT Cooling Method is used by the Mainframe Computer In the Movie Sunshine .. Which came out some years ago !! This one also had a high degree of A.I. and A.V. input & feedback .. (pretty wild) 😮
Such a cool video, ty Anastasi
Plug =[□]= and play, keeping chips cool is an ever changing challenge!
As cooling is actually transferring the heat (on chip or other) elsewhere; the ancillary opportunity is to turn that heat back into electrical energy for powering the cooling systems or the data centers. I know this become feasible only if the cost makes it so.
Just one addition: the problem is not the power or generated heat itself. A power plant with e.g. 1200 MW power output has roughly 1800MW of additional power that is simply heat (assuming a 40% efficiency of process)
The issue is power density and how to transfer this high density power output to a larger area where it in the end is air cooled.
You are bringing me up to date
While watching this video about liquid cooling the CPUs that maybe a closed loop system can be set up to create steam to generate electricity to power the CPUs. Probably not the first person to come up with that idea.
Or steam to froth some milk
Wow definitely in love with this technology
Only the technology?
These improvements in cooling would also allow clock speeds to increase considerably. And if graphene comes on the scene (which is possible now that we can make semiconductors with them) that could get even faster.
Cool! Thanks for the conference link!
I estimate that compared to the 5nm technology node, we can look forward to a performance gain of approximately 1.8 times or a power consumption reduction of 2.6 times as we move into the chiplet phase. Therefore, I believe it is imperative to focus on improving the efficiency of mathematical calculations to resolve pertinent problems in software.
Amazing video! Thanks for that!
very interesting. Even if, it seems clear to me that the real next huge lap in computing technology will be make computing power not lineary related to the amount of heat produced by the conputing chip.
6:50 Copper's thermal conductivity is 3.4x that of silicon.
Copper: 395 (W/m)/K @80°C
Silicon: 117 (W/m)/K @80°C
I retired recently and this helps me stay up-to-date. 👍
Learning something and having a laugh...I wish my teachers would have been like this
I knew radio techs in the early 70's using "Can-tenna" products as dummy loads for r.f. outputs. A 1 gal. paint can could dump a kilowatt of radio energy output into mineral oil while the set was tested and adjusted.
I thoroughly enjoyed your updating archaic me, on what are clearly critical information regarding the requirements for our new personal super computers. Thank You 😁🌹
Using a coolant other than water might be more efficient. Something like gallium or rubidium. However, these have much higher viscosity so you'd need bigger pumps. They're also very air sensitive so the transport system would need to be designed around oxygen impermeable materials, increasing cost and complexity. But for a server farm these would be worthwhile investments.
Fascinating. At first I thought you were integrating peltier electric cooling into the chips.
At the beginning of 1970’s, the communication satellite dishes (for telecom companies) had cryogenic cooling attached to the rear to cool the LNA (low noise amplifiers). By the later 70’s the newer dishes used semiconductor LNA’s that did not require cooling.
Smoking hot video 🔥
Thank you for these amazing insights
Love the video! Been catching up again. Got to say I've been thinking of acoustic resonance cooling combined with a jules Thompson cooling effect to the integrated heat spreader for a while. It would use a "balanced armature" (very small diaphragm pump) that would help with power delivery through soaking up emi areas and collecting free errant electrons during expansion. Basically it's shaped to bend during heating, but also has these tiny capacitors they just developed that have power get delivered from these tiny coils placed around transistor areas to soak up emi and z style electromagnetic fields produced via current delivery. They pump coolant around like this talks about but it changes the stability of band gap frequencies as it flows because it's conductive (slightly) as it changes pressure during flow.
Which captures more electron hole potentials and electro magnetic/static fields into itself. Granted it becomes more resistive to follow as it does this and gains more heat. So it has to go into a much larger expanse area into the heat spreader to release it energy, electromagnetic potential energy, and it's kinetic energy so I figured a passive diaphragm capture for acoustic resonance would be good to do as it then places a phonoic force wave through the material to keep "still", I guess, the material but gains reflective disorder and kinetic energy back to it that can be dissipated into a regular heart sink/spreader.
But I was also getting into light fiber optics for cooling that can be conductive, if we could reflect light in through the liquid that take advantage of the pumps, them we could create a photon cooling area into transition area that works off of a grid reflection pattern. This would in turn help with the acoustic resonance cooling, while being able to help deliver switching help. Either make a switch happen, or reduce it's chance. But then conductive fiber optics get into the conversations with that for power delivery for old school tube TV tech that can alter frequency, direction, reflective properties, placement on target, photon resonance emittance, and so on. It becomes more of a hybrid qbit & regular transistor chip design. Something I put up on my other TH-cam channel but it's long winded for some I'm sure.
You’re amazing!
At some point the cooling tech used in quantum computers may have to be applied to advanced "conventional" cpu's and gpu's. YES !?!?!?
How do you want the normal PC user to use such cooling solutions at such cold temps.
I highly doubt, cos that level of cooling is crazy
The current quantum cooling is for early human first gen quantum computers. State of the art quantum computers require the same cooling as the leaf, one of nature's quantum computing devices. When the brain is better recognized as a quantum computer, the only cooling that may be needed is a cap on a sunny day.
That dense micro-channels won't work, water molecules will create friction, and mechanical force on the molecules of the Chip; also water oxides Copper, another liquid element needs to be used. So, more spaced micro-channels, would work but the Volume for useful transistors will be reduce, so Dies need to be Bigger, to keep the density, making less Chips/Wafer, increasing the price a lot, maybe for those huge Chips. I think a variation would work: Submerging the Big Chips in that liquid that bubbles but stays cool, so the liquid will be picked by small tubes that will be shaped in the form of the current air coolers, so they'll be cooled with air there; from there it'll go to a Chiller, and back to the Big Chip
What if, someday, we could harness quantum tunneling to teleport heat away from a chip, just enough to alleviate the issues?
This issue is the same as faced by the home audio industry back in the 70's when regulations required that the advertised output be able to be maintained for, say, 1/2 hour, rather than momentary peak output. Heat sinks grew exponentially and started including heat pipes and thermosiphons for the output transistors.
Very fun and Kool as always, Thank you.
The challenges of managing increased thermal design power are well-explained. The use of through-silicon vias and heat corridors is an innovative approach.
Anastasia is the kind of presenter that could make videos about why moths circle lightbulbs, and it would be a hit
It becomes clear that light will have to replace electricity for switching. Interference, both destructive and constructive will provide the 1s and 0s.
Power can be provided by the results of the SAFIRE experiment and the adoption of the Structured Atom model (SAM) of the nucleus.
OK then. subscribed 👍
Thank you for the amazing video 🔥🔥
great video ana
Great video well explained keep it up
I am subscribed! One of the 25% 🙂
immersion cooling is basically phase change cooling. Finding replacement for coolant should be as easy as varying the pressure. They could use water at 1/10 bar, which boils at 40C. They could also vary the pressure for different cooling needs
I also think so. However, could it be that keeping water from interfering with electrical networks is tough? That, even de-ionised water isnt working?
@@catman4859 Sure could. There's also a question of how different materials interact with the electronic components at these phase change points. Maybe liquid de-ionized water doesn't present any issues but its gaseous form corrodes them.
@@mikelord93 yeah. So there are probably lots of complexities here that we are unaware of. Otherwise, such a problem would have been already solved.
Yeah I thought that using phase change inside the chip would be what this video was about. Not just water.
Isn't liquid pretty thick relative to the transistors! How thick are these channels?
I thought water had a capillary limit of .0001mm. Or 100 nanometers.
I remember seeing a solid state fan demo. Think it was called AirJet. I wonder how that company was doing. I guess this technology blows it out of the water.
When water becomes gas - that spot stops being cooled until its covered with water again, because gas has bad heat conductivity. I think its hard to predict and control hotspots in such a setup. When boiling water in a pot, you can see air bubbles forming on the bottom and staying on one place for a long time. These bubbles will cause overheating of a chip, if not moved quickly.
Díky!
Maybe combating overheating batteries could help solve this with graphene which has the best thermal qualities .maybe sandwich the component's ,but then their might be electrical run off,where a similar material with better resistance could work
That amd Athlon chip brings back some good old memories:)
Interesting cooling tectechnology maybe the best solution may be a hybrid idea combining the best positive features into one.
That was a truly fascinating walk through the heat challenges. There's an interesting issue here - when you're dealing with in-chip channels this thin you end up with a lot of resistance, however low your coolant viscosity is. It gets worse if you use phase conversion (liquid to gas and back) to pick up heat energy, you will need quite high pressures to maintain sufficient flow. That again works its way back into a need for robustness of these channels to handle that. It would be interesting to know how that is dealt with in design as well as the external facilities to keep that coolant moving.
All that heat is a valuable recourse if it can be captured and recycled into electricity. It could help compensate for the large power needed to run the computers in the first place.
I'm just a Philosopher. And I don't know Badiou extensively or follow the Set Theory construct, but rather wraggle a Hegelian one at worst if to belong anywhere. No less, Badiou said something somewhere about a Philosopher's duty to know at least a little bit about everything. Which I don't totally comply with, but the best example of a person who actually did this would be Krishnamurti. The whole time I was thinking, "what's in the liquid?". And the honesty of at least commenting about it's after affects was admirable, on the environment. Before Philosophy, I was interested in the second law of thermodynamics and Entropy from the mid-way of the Novel, especially T. Pynchon. So I thought today was really cool. Salut.
Mineral oil is nonconductive but is very messy especially if you have to replace something before you are replacing your whole system. And as far as the masturbating that can hit a whole office building with that kind of heat. Maybe they should look into heat for office buildings with that much heat. Btw u r so intelligent and pretty it makes me think about going straight!
Good job Anastasi!!
Great video, very informative, glad I watched it.