Testing the AI-generated 3D Printed CPU LN2 Container

The AI-generated & 3D-printed CPU LN2 container delivers a 3X cool down speedup, 1.2X heatup speedup, and +20% LN2 efficiency.


We’re trying out the US$10,000 AI-generated and 3D-printed LN2 container. It was on display at the G.SKILL Computex 2024 booth earlier this month. We’ll go over the basic design and manufacturing processes and also evaluate the performance in a real-world extreme overclocking scenario.

The performance is quite spectacular, so let’s get right to it.

Project Scope

As I outlined in a different page on this blog a couple of weeks ago, the idea is simple: let’s utilize cutting-edge technologies like generative AI design and additive manufacturing to create a thermal solution for extreme overclocking with liquid nitrogen.

Generative AI design and additive manufacturing are still relatively new technologies. They are revolutionizing many industries by helping to optimize thermal performance, reducing time to market, and lowering costs. However, their application in designing and manufacturing LN2 containers for achieving peak compute performance using liquid nitrogen cooling has yet to be explored.

So, this project aimed to achieve two key goals:

  1. First, a feasibility study: we want to know if it’s even possible to produce a high-performance LN2 CPU container using these technologies
  2. Second, if it’s feasible to produce a sample, conduct a performance evaluation and see how it stacks up against one of the best LN2 containers available on the market.

Feasibility Study

Our first objective is to see if it’s even possible to build an LN2 container with generative AI and additive manufacturing techniques. For this purpose, we teamed up with three industry-leading companies: Diabatix, 3D Systems, and ElmorLabs.

Diabatix ColdStream Nxt Generative AI Technology

Diabatix is a Belgian company leading the charge of utilizing generative AI technology to develop thermal solutions. They do this with their cloud-based ColdStream Nxt platform.

The platform features a physics-reinforced approach enabling design optimization for maximum heat transfer and efficiency while minimizing material usage and energy consumption. The cloud-based platform boasts an impressive 80% reduction in engineering time and a 20% improvement in cooling performance, all while delivering a user-friendly experience.

For instance, Diabatix previously demonstrated a 55% improvement in water block cooling efficiency by combining generative AI and additive manufacturing techniques. However, designing an LN2 container presents unique challenges compared to traditional air and liquid cooling solutions.

diabatix ai generated water block

The approach to thermal design using generative AI is slightly different than traditional methods. Essentially, you set up a generative design internal loop which is defined by a certain design target and usually multiple design constraints.

  • The design target for the LN2 container is an objective function which essentially says the temperature should be as low as possible.
  • The design constraints include the manufacturing constraints but can also include other constraints like cost (although that wasn’t the case for this project

The ColdStream engine will then generate an initial design followed by a performance evaluation. Based on the evaluation, the software will iterate the design to further improve the design target within the design constraints.

The output of the generative AI process is impressive. It is immediately obvious, however, that this design cannot be manufactured using traditional processes. We need metal 3D printing!

ai generated ln2 cpu container

3D Systems Direct Metal Printing Technology

Founded in 1986, 3D Systems is a pioneer in the 3D printing industry, pushing the boundaries of additive manufacturing innovation. They offer a range of technologies, including Direct Metal Printing or DMP.

DMP is an additive manufacturing technique that builds complex, high-quality metal parts from 3D CAD data. A high-precision laser selectively melts metal powder particles layer by layer. That enables the creation of intricate geometries that are impossible to construct with traditional subtractive or casting methods. DMP offers a wide range of functional metals for printing designs, from prototypes to production runs of up to 20,000 units. For this project, 3D Systems utilized their DMP Flex 350 Metal 3D Printer.

3d systems dmp flex 350

In addition to the 3D Systems printer, we also used a 3D Systems certified oxy-gen free copper powder. The powder is a crucial element for designing a thermal solution because thermal conductivity is the key property. Maintaining the purity of the copper powder during printing is of utmost importance as any oxygen in the Cu matrix has a detrimental effect on its thermal conductivity.

ElmorLabs Volcano CPU LN2 Container

ElmorLabs is perhaps the most well-known name in extreme overclocking and PC DIY enthusiast circles. The ElmorLabs Volcano CPU LN2 container is considered one of the top LN2 containers used by many of the top extreme overclockers. It comes as no surprise then that it served as the reference design for this project.

elmorlabs volcano

Featuring a copper base and an aluminum body, the Volcano LN2 pot is designed for maximum cooling efficiency and stability, ensuring reliable and consistent performance every time.

Modern CPUs and GPUs boast significantly higher power usage and density than in the past. To achieve peak performance, the heat generated by these chips needs to be transferred as efficiently and quickly as possible, which is a major challenge for many outdated CPU LN2 container designs. Some high-performance chips can dissipate nearly 2000 watts of power!

Therefore, effective LN2 containers must address several critical design constraints:

  • Mass: The container needs enough mass to maintain a stable and cool LN2 reservoir.
  • Surface Area: The design must maximize the surface area in contact with the CPU for efficient heat transfer.
  • Leidenfrost Effect: The design should mitigate the Leidenfrost effect, a phenomenon that can hinder heat transfer at boiling temperatures.

To make comparing the AI-generated and Volcano LN2 containers, we re-used the Volcano mounting mechanism.

With the efforts and resources of these three industry-leading partners, we were able to design and manufacture a CPU LN2 container. Feasibility, check! But what about thermal performance? Let’s put the container to the test.

Basic Performance Tests

Our first three tests are purely focused on the performance of the container without heat load. We conduct three basic tests:

  1. Cool down time
  2. Heat up time
  3. Efficiency

Cool Down Time

With the cool-down time test, we measure how long it takes for the LN2 container to get from an ambient temperature to the boiling temperature of nitrogen, which is about -196 degrees Celsius. Having a fast cool-down temperature is important because the faster you can cool down the LN2 container, the faster you can get started chasing those peak clocks and benchmark records.

The ElmorLabs Volcano LN2 container is known to be one of the fastest containers on the market. However, the AI-generated design utterly destroys the traditional design in this test as we see a 3X speed up in cool-down performance. It takes the AI-generated design less than a minute to cool down to -195C whereas it takes the Volcano about 3 minutes.

ai generated ln2 container 3X cool down improvement

Heat Up Time

With the heat-up time test, we measure how long it takes for the LN2 container to get from a “full pot” scenario back to an ambient temperature. Being able to heat up quickly can be important in case you need to swap parts or deal with a particularly challenging cold bug or cold boot bug.

Again, the AI-generated design is quicker than the traditional design of the ElmorLabs Volcano. However, the performance difference isn’t as dramatic as the cool-down test scenario. We see a speedup of 1.21X as our AI LN2 container heats up 30 seconds faster than the Volcano.

ai generated ln2 container 1.2X heatup improvement


In our next test, we measure the temperature delta when using a fixed 500ml of liquid nitrogen. A more efficient LN2 container design will cause less of the nitrogen to evaporate and more of it to be used for actual cooling. Having an efficient design can help save liquid nitrogen which can be important if you have a limited supply.

Again, the AI-generated design shows a significant performance improvement as it can cool down to -133 degrees Celsius with half a liter of liquid nitrogen. The Volcano also achieves below minus 100 degrees Celsius. The total efficiency improvement is about 20%.

ai generated ln2 container +20% efficiency

Practical Performance Tests

So clearly the AI-generated and 3D-printed CPU LN2 container shows a lot of potential. But the proof is in eating the pudding. So, we put the LN2 container to the test in a real system with the Intel Core i9-14900KF Raptor Lake processor.

We conducted three tests:

  1. Performance in Cinebench 2024 when overclocked
  2. Check the CPU Pot to IHS temperature delta in high-load scenarios
  3. Push the power consumption of the CPU to the maximum

Cinebench 2024 Performance

Our first check is the most practical of all our tests conducted as we try to find out what’s the maximum stable CPU frequency. Of course, a key limiting factor is the CPU’s overclocking capability. Still, a well-designed container can help maintain higher loads for longer.

We find that both LN2 containers can handle the Core i9-14900KF with P-cores clocked to 7.4 GHz without any issue. It seemed the AI-generated design could perhaps hold 7.5 GHz just a tad longer. But that might just be run-to-run variation.

Either way, the AI design passes with flying colors.

Temperature Delta

In our second test, we want to verify the delta between the CPU container base and the CPU heat spreader. The idea is that good designs will minimize the delta as it transfers the heat more efficiently.

Here we find that the traditional design has a smaller delta of only 15 degrees Celsius, whereas our AI-generated design has a delta of 23 degrees Celsius. However, the CPU heat spreader temperature is pretty much the same. So, there might have been an issue with the heat spreader temperature probe placement. We will have to repeat this test in the future.

July 1st Update: the experts at Diabatix chimed in with additional information that helps us better understand the odd behavior with the temperature delta.

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 Volcano 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.


Maximum CPU Power

Last but not least, I also wanted to do a “glory run” and push the CPU to its maximum power consumption. For this purpose, we ran the OCCT CPU Stress Test with AVX2 and small data set. This stresses not only the CPU cores but also the CPU’s caches.

We pushed the CPU over 600W for a couple of minutes! However, the AI-generated and 3D-printed CPU LN2 container had no issues at all. Another impressive result.

600w raptor lake


The performance of the AI-generated 3D-printed LN2 container exceeded our wildest expectations. Before testing the LN2 container, most of us were skeptical about the design. But our opinion quickly changed, the moment we witnessed the incredible cool-down performance.

Overall, this design beats a traditional design on all relevant performance metrics. However, the impact in a real-world scenario is less pronounced. When testing with a highly overclocked Intel Raptor Lake processor, the traditional design performs on par with the AI-generated design. Furthermore, it costs significantly less than the AI-generated design.

So, what are the next steps for this LN2 container? Well, there’s nothing concrete I can share with you today. However, we have a couple of options.

On the design side, we could look into performance and cost optimizations. For example, the design of the LN2 container doesn’t necessarily need to be circular. That shape is the preferred choice to cost-optimize for CNC machining. We could, in theory, print any 3D shape.

Also on the design side, we could look into an LN2 container design for even higher-power CPUs such as the AMD Ryzen Threadripper or Intel Xeon 6 processors. Elmor has demonstrated an overclocked Intel Sapphire Rapids can go up to nearly 2000W!

We’d also love to find a way to commercialize the design and make it available to anyone. Of course, the manufacturing cost will be significantly higher than a CNC-machined LN2 container. But, perhaps we can bring it to market at a reasonable cost if we find a way to optimize the supply chain.

ai generated ln2 container next steps


Anyway, that’s all for today! I hope you enjoyed this video.

I want to thank Lieven from Diabatix and Bram from 3D Systems for getting on board with this project. Without their enthusiasm, know-how, and support, we wouldn’t have been able to try out an AI-generated and 3D-printed LN2 container. Be sure to check out their websites and consider them for your own AI and 3D printing projects. Also, I want to thank Jon from ElmorLabs for his support in this project. It takes guts to be open to having your designs challenged.

Lastly, I want to thank you for watching and the Patreon supporters for supporting my work. If you have any questions or comments, please drop them in the comment section below. See you next time!

Leave A Comment