"You had a great idea Elmor, but my pot is more advanced in every way" "How'd you solve the icing problem?" "...Icing problem?" with apologies to Iron Man (2008)
@@SkatterBencher so it's just increasing the surface are. A laudable improvement in cool down/heat up, but does this translate to increased headroom for overclocking and performance?
Wow, been waiting for these results since it showed up at Computex. Great performance there. Have you tried for longer sessions on it? As one potential issue that was raised was that there were some small channels which could become obstructed during long sessions. Otherwise the performance was really amazing.
We're tried it in a multi-hour session with the Raptor Lake system. We did see ice formation but so far didn't see any impact on performance. So the next step will be to try it in a "competitive" session to see how it stacks up when performance really matters. Maybe with the next gen chips
@@SkatterBencher One quick and easy way to test the effect of ice would be to simply repeat the pull down and heat up experiments, perhaps with a max temp of -30C so you're not melting the ice and as a realistic CBB temp. If ice is a problem, the -30C to -196C pulldowns will get worse and worse compared to the Volcano as you repeat. If it's not a problem they should both get a bit slower every time, but by similar amounts.
But performance will be less than what you can achieve with 3D printing. The goal here was really to see how far we can go with the performance. The answer you can see from te video: it becomes so performant that the heat spreader and the thermal interfaces actually become the major bottleneck in the system.
Amazing work. The design reminds me of coral! I imagine a future where our tech mimics nature. Biodegradable, self-evolving, who knows where we could take it... There has to be a way to sinter the material as it is being printed, varying levels of laser intensity perhaps? I feel as though printing the "tendrils" hollow may achieve this effect.
The relevant parts of the Volcano pot are copper, the only aluminium part is the extension to keep LN2 to make pouring easier and keep liquid from just spilling out.
microdiamond impregnated pure metal dust for SLS 3D printing is a possibility, but it would likely increase costs 10-fold for a 1.1x performance improvement otherwise, unless you can weld diamonds together or melt them (both physical impossibilities AFAIK) you're not looking at any kind of 3D printing with diamonds. The best you could do is CVD (i.e. how most lab grown diamonds are made), and trying to grow specific shapes or cut the larger diamond crystal instead... long story short, graphene is still better
IMHO this could have been achieved using traditional manufacturing or 3D SLS printing with a much simpler design of thin cylindrical processes (or really just copper wire soldered to a copper base and tensioned using an open lid at the top where the wires would be anchored). Performance hierarchy would probably be 3D-printed > milled > soldered cylinders > soldered wires (depending on the number of wires). However, the cost hierarchy would also probably be the same. The key difference between your "AI" design and Elmorlabs design is surface area. Assuming an equal mass of copper in both solutions, increased surface area would naturally speed up cool down or heat up. I don't think this would translate into much additional performance or overclocking headroom as once they are both cooled down they are at equal temperatures and the heat load from the CPU would be similarly equal. Still an interesting idea and interesting video... however one that seems to be a proof of concept rather than having any practical real-world benefit
If physics was as simple as maximizing contact area for heat transfer... we would be out of business quickly ;-) The reason why the AI design (let's leave out the " " shall we) outperforms the Elmorlabs design is neither mass, nor contact area (both are actually pretty similar), but it is the combination of the effectiveness of the heat transport through the copper structures (e.g. wider base, smaller tips) together with the highly efficient vapor bubble transport within the LN2.
@@iamdmc Having more contact area won't help because it is not the bottleneck in the heat transfer. The way the cooling structures are structured, maximizes the transfer from the copper to the LN2 to such an extent that the conduction inside the copper has become the bottleneck. Using (thinner) copper wires will make this even worse. Using a copper wire like you describe will therefore be far less effective than the current design.
@@Diabatix if that were true I imagine that you'd have seen a performance difference (improvement) with your design compared to the Elmorlabs design... at least according to the numbers in this video there was no performance difference at all
@@iamdmc For sake of discussion, let's indeed ignore all tests that demonstrated a dramatic improvement over the Volcano and focus on the one where performance was seamingly similar. It is easy explainable though: the temperature difference that is reported, is the temperature difference over the heat spreader, not the containers, so this test is actually measuring the performance of the heat spreader. Since in both cases the same heat spreader is used and the power is the same, the difference can only come from probe placement or thermal interface material variations. Another way to interpret the 2nd real life test is that the AI container only needs 9°C (=-196 - -187) to evacuate the 600W of power from base to LN2 while the Vocano requires 19°C achieve the same. This brings the performance gain perfectly in line with all other tests. That's also why a next step will be testing the container at a much higher power because these tests show we are still far from the limit it can handle.
With many heat cycles at extreme temperature I'd be worried the 3d printed part would fail or break down as the grain structure on a 3d printed part will be terrible. It is extremely cool though to see.
@@SkatterBencher he's just referring to temperature-related metal fatigue. It makes sense that the 3D printed version would have worse fatigue tolerance because of the way the crystal structure of the copper would be aligned in the 3D printed version vs a milled block of solid copper
I know the likelihood of you actually doing this is slim to none but is there any chance you could make a video going over the majority of bios settings for maybe the z390/z590 or newer boards and make it easy to understand?
but if you look closely, you can immediately see the error (incomprehensible) on the human side, it was enough to use any cooler with a large contact surface, I don't understand this hype, this is primary school physics. Moreover, the result obtained by AI can be easily improved using the same method = increasing the contact area of the cooling fluid with the solid material. eot.
There's much more to this, like the pathing of channels with such a low visc fluid, the least amount of exposed LN2 to air while still having it reach channels equally, the ability for hot gas to exit without harming the flow of cool liquid in, the ratio of thickness to surface area without harming heat transfer, stability of the copper under contraction. Ect. If you don't understand something don't hate on it, let people enjoy science.
Sort of because the design also minimizes the amount of LN2 exposed to air so it doesnt let it evaporate as fast as the Volcano which can be considered an almost "open" design. Yeah more surface area is going to transfer heat better but traditional pots built like that are also very fast and it's hard to control temperatures. In my opinion this is a very fast pot that is also good at retaining the ln2 inside of it.
It's not black magic. This isnt going to scam anyone but cost a lot for sure. The Ai container is basically a "technology demonstrator" it is not meant to be the perfect pot but definetly gives you an idea of what you can reach if you use the laws of physics and chemistry and push them to their limit. Obviously it would be the best if you gave AI much more liberty on the shape of the container, height limit and mounting; all in an "ideal' scenario which is very unlikely because in real life applications there are many external variables than can prevent you from getting the expected "ideal" result.
"You had a great idea Elmor, but my pot is more advanced in every way"
"How'd you solve the icing problem?"
"...Icing problem?"
with apologies to Iron Man (2008)
What is the mass difference between the two containers?
It's identical, both about 1.7kg
@@SkatterBencher so it's just increasing the surface are. A laudable improvement in cool down/heat up, but does this translate to increased headroom for overclocking and performance?
bare copper with huge surface area
Wow, been waiting for these results since it showed up at Computex.
Great performance there. Have you tried for longer sessions on it? As one potential issue that was raised was that there were some small channels which could become obstructed during long sessions. Otherwise the performance was really amazing.
We're tried it in a multi-hour session with the Raptor Lake system. We did see ice formation but so far didn't see any impact on performance.
So the next step will be to try it in a "competitive" session to see how it stacks up when performance really matters. Maybe with the next gen chips
@SkatterBencher that's great. Now to find a way to make it cost effective somehow, so we can all be blessed with such amazing performance.
@@SkatterBencher One quick and easy way to test the effect of ice would be to simply repeat the pull down and heat up experiments, perhaps with a max temp of -30C so you're not melting the ice and as a realistic CBB temp. If ice is a problem, the -30C to -196C pulldowns will get worse and worse compared to the Volcano as you repeat. If it's not a problem they should both get a bit slower every time, but by similar amounts.
That's a great suggestion. I'll try it next time there's access to LN2.
Feed the CNC constraints into the AI as well and get an optimized pot that can be relatively easily manufactured?
Yes, that's perfectly possible to do! 🙂
But performance will be less than what you can achieve with 3D printing. The goal here was really to see how far we can go with the performance. The answer you can see from te video: it becomes so performant that the heat spreader and the thermal interfaces actually become the major bottleneck in the system.
Amazing work. The design reminds me of coral! I imagine a future where our tech mimics nature. Biodegradable, self-evolving, who knows where we could take it... There has to be a way to sinter the material as it is being printed, varying levels of laser intensity perhaps? I feel as though printing the "tendrils" hollow may achieve this effect.
Whoa the cooldown time is seriously FAST!
Lhaa ketemu legend 😅
so we are talking about the differences between aluminum and purified copper...
well let's see if oney day there will be 3D diamond printing
The relevant parts of the Volcano pot are copper, the only aluminium part is the extension to keep LN2 to make pouring easier and keep liquid from just spilling out.
microdiamond impregnated pure metal dust for SLS 3D printing is a possibility, but it would likely increase costs 10-fold for a 1.1x performance improvement
otherwise, unless you can weld diamonds together or melt them (both physical impossibilities AFAIK) you're not looking at any kind of 3D printing with diamonds. The best you could do is CVD (i.e. how most lab grown diamonds are made), and trying to grow specific shapes or cut the larger diamond crystal instead...
long story short, graphene is still better
IMHO this could have been achieved using traditional manufacturing or 3D SLS printing with a much simpler design of thin cylindrical processes (or really just copper wire soldered to a copper base and tensioned using an open lid at the top where the wires would be anchored). Performance hierarchy would probably be 3D-printed > milled > soldered cylinders > soldered wires (depending on the number of wires). However, the cost hierarchy would also probably be the same.
The key difference between your "AI" design and Elmorlabs design is surface area. Assuming an equal mass of copper in both solutions, increased surface area would naturally speed up cool down or heat up. I don't think this would translate into much additional performance or overclocking headroom as once they are both cooled down they are at equal temperatures and the heat load from the CPU would be similarly equal.
Still an interesting idea and interesting video... however one that seems to be a proof of concept rather than having any practical real-world benefit
If physics was as simple as maximizing contact area for heat transfer... we would be out of business quickly ;-)
The reason why the AI design (let's leave out the " " shall we) outperforms the Elmorlabs design is neither mass, nor contact area (both are actually pretty similar), but it is the combination of the effectiveness of the heat transport through the copper structures (e.g. wider base, smaller tips) together with the highly efficient vapor bubble transport within the LN2.
@@Diabatix try the copper wire idea - which is likely to provide the maximum amount of surface area
@@iamdmc Having more contact area won't help because it is not the bottleneck in the heat transfer. The way the cooling structures are structured, maximizes the transfer from the copper to the LN2 to such an extent that the conduction inside the copper has become the bottleneck. Using (thinner) copper wires will make this even worse. Using a copper wire like you describe will therefore be far less effective than the current design.
@@Diabatix if that were true I imagine that you'd have seen a performance difference (improvement) with your design compared to the Elmorlabs design... at least according to the numbers in this video there was no performance difference at all
@@iamdmc For sake of discussion, let's indeed ignore all tests that demonstrated a dramatic improvement over the Volcano and focus on the one where performance was seamingly similar. It is easy explainable though: the temperature difference that is reported, is the temperature difference over the heat spreader, not the containers, so this test is actually measuring the performance of the heat spreader. Since in both cases the same heat spreader is used and the power is the same, the difference can only come from probe placement or thermal interface material variations.
Another way to interpret the 2nd real life test is that the AI container only needs 9°C (=-196 - -187) to evacuate the 600W of power from base to LN2 while the Vocano requires 19°C achieve the same. This brings the performance gain perfectly in line with all other tests. That's also why a next step will be testing the container at a much higher power because these tests show we are still far from the limit it can handle.
don't you have to disclose paid ads?
With many heat cycles at extreme temperature I'd be worried the 3d printed part would fail or break down as the grain structure on a 3d printed part will be terrible.
It is extremely cool though to see.
Interesting thought. Do you have any references that show 3D printed copper falls apart faster than regular copper?
@@SkatterBencher he's just referring to temperature-related metal fatigue. It makes sense that the 3D printed version would have worse fatigue tolerance because of the way the crystal structure of the copper would be aligned in the 3D printed version vs a milled block of solid copper
Dad?????
@@RsLays Yes, son?
Interesting, how about a pot designed specifically for helium?
Love the idea, but hard to execute given how rare and expensive helium sessions are. How would we test the performance? :D
I know the likelihood of you actually doing this is slim to none but is there any chance you could make a video going over the majority of bios settings for maybe the z390/z590 or newer boards and make it easy to understand?
I have a bunch of guides up with Z490 and Z590 motherboards. Just look for the Comet Lake and Rocket Lake CPUs in my video catalog
@@SkatterBencher awesome. Thanks guess I shoulda looked first lol happy 4th!
but if you look closely, you can immediately see the error (incomprehensible) on the human side, it was enough to use any cooler with a large contact surface, I don't understand this hype, this is primary school physics. Moreover, the result obtained by AI can be easily improved using the same method = increasing the contact area of the cooling fluid with the solid material. eot.
There's much more to this, like the pathing of channels with such a low visc fluid, the least amount of exposed LN2 to air while still having it reach channels equally, the ability for hot gas to exit without harming the flow of cool liquid in, the ratio of thickness to surface area without harming heat transfer, stability of the copper under contraction. Ect. If you don't understand something don't hate on it, let people enjoy science.
this is very interesting
Interesting 🥶
You should make a Ai water cooler, that would be way Interesting
Diabatix already made such an experiment a while back: www.diabatix.com/blog/elevating-cpu-cooler-efficiency-by-55-am-generative-design
Вау, это сарказм)) Ясное дело что ячейки где их до хрена и они мелкие будут быстрее передавать чем блямбы в правом углу экрана ))))
Sort of because the design also minimizes the amount of LN2 exposed to air so it doesnt let it evaporate as fast as the Volcano which can be considered an almost "open" design. Yeah more surface area is going to transfer heat better but traditional pots built like that are also very fast and it's hard to control temperatures.
In my opinion this is a very fast pot that is also good at retaining the ln2 inside of it.
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOooooooooooooooooooooooooo
Finally :D
Медные губки для посуды подорожают. 🎉🎉🎉 И при чем тут старая мочалка? 😆👹😆
Lol get scammed. Reminds me about audiophiles. U should just compare a block of wood and any metal. Feel free to chose any type of carcus as well.
It's not black magic. This isnt going to scam anyone but cost a lot for sure. The Ai container is basically a "technology demonstrator" it is not meant to be the perfect pot but definetly gives you an idea of what you can reach if you use the laws of physics and chemistry and push them to their limit. Obviously it would be the best if you gave AI much more liberty on the shape of the container, height limit and mounting; all in an "ideal' scenario which is very unlikely because in real life applications there are many external variables than can prevent you from getting the expected "ideal" result.
What the hell are you on about
Самая бесполезная дич.