Pushing the limits: Enthusiasts are always looking for an edge in the world of overclocking and extreme cooling. In this wild test, the team wanted to find out if advanced GenAI and 3D printing could help squeeze even more performance out of today's processors. To find the answer, they built a liquid nitrogen (LN2) container in a completely new way and came to an interesting conclusion:
The project brought together experts from across the ecosystem, including Skatterbencher, known for its overclocking prowess, Diabatix, which specializes in generative AI for thermal solutions, additive manufacturing 3D Systems, and overclocking gear company ElmorLabs.
The team took inspiration from ElmorLabs' existing Volcano LN2 container and turned to Diabatix's ColdStream Next AI to design an improved version. 3D Systems then brought the digital blueprint to life, 3D printing a prototype using oxygen-free copper powder. But to their surprise, this cutting-edge process came with a hefty price tag of $10,000, a far cry from the original Volcano's $260 price tag.
The AI/3D printed design showed promise in initial testing, focusing on three key metrics: cooling time from room temperature to -194°C, heating time from -194°C to 20°C at a load of 1250W, and minimum temperature achieved using 500mL of liquid nitrogen.
It outshone the Volcano in cooling speed, going from 28°C to -194°C in under a minute versus 3 minutes for the Volcano. Heating performance was also superior, with the AI container warming up 30% faster. Efficiency also favored the AI design, reaching -133°C using 500mL of LN2 versus the Volcano stopping at -100°C.
However, these tests do not reflect real-world performance, so the team decided to run three more tests with the Intel Core i9-14900KF Raptor Lake processor: First, we launched Cinebench 2024 to find the most stable maximum CPU frequency.
“We found that both LN2 containers were able to handle a Core i9-14900KF with a P-core clocked to 7.4GHz without any issues. It seems the AI-generated design can probably sustain 7.5GHz for a bit longer. But that could be run-to-run variation,” they noted.
The second test assessed the actual heat transfer capabilities by checking the CPU temperature differential between the heat spreader and the cooling container base, and also included a full-scale stress test where over 600W of power was passed through the chip for several minutes.
The AI Container had a slight lead, but the difference was relatively modest when compared to theoretical test results. The temperature difference between the CPU heatspreader and the container base was smaller for the 3D printed model, but not by a huge margin. As mentioned above, the performance gains in Cinebench were also fairly modest.
The team crunched the numbers and determined that, while technically impressive, their AI/3D printed design is currently not cost-effective in modest overclocking scenarios – not at the $10,000 price point.
But they're not done yet: While there's “nothing concrete” yet, the team said it could look at optimizing performance and cost — for example, the LN2 container design doesn't necessarily have to be circular — and they're also looking at new designs for more powerful CPUs like Ryzen Threadripper and Intel's Xeon 6.
Overall, while the feasibility study may have revealed some limitations, it also proved that generative AI has applications beyond simply mass-producing six-fingered models.
