Pushing the bounds: Fans are at all times on the lookout for an edge on this planet of overclocking and excessive cooling. On this wild take a look at, the group sought to find out whether or not superior GenAI and 3D printing may assist them squeeze out extra efficiency from right this moment’s processors. To reply this, they constructed a liquid nitrogen (LN2) container in a complete new manner – and arrived at some fascinating conclusions.
The project introduced collectively specialists from throughout the ecosystem – Skatterbencher who’s famend for overclocking prowess; Diabatix, specializing in generative AI for thermal options; 3D Methods for additive manufacturing; and at last ElmorLabs for overclocking gear.
The group took ElmorLabs’ present Volcano LN2 container as a reference level, then tasked Diabatix’s ColdStream Subsequent AI to generate an improved design. 3D Methods then introduced that digital blueprint to life, 3D printing a prototype utilizing oxygen-free copper powder. Shockingly although, the cutting-edge course of commanded a steep $10,000 price ticket – a far cry from the $260 value of the unique Volcano.
The AI/3D printed design confirmed promise in early testing, specializing in three key metrics: cool-down time from room temperature to -194°C, heat-up time from -194°C to twenty°C below a 1250W load, and the bottom temperature achieved utilizing 500mL of liquid nitrogen.
It blew previous the Volcano in cool-down pace, chilling from 28°C to -194°C in below a minute in comparison with the Volcano’s 3-minute tempo. Warmth-up efficiency was higher too, with the AI container warming up 30% sooner. Effectivity additionally favored the AI design – utilizing 500mL of LN2, it hit -133°C, whereas the Volcano stopped quick at -100°C.
Nevertheless, since these checks don’t signify real-world efficiency, the group determined to run three extra utilizing the Intel Core i9-14900KF Raptor Lake processor. First, they fired up Cinebench 2024 to seek out essentially the most steady most CPU frequency.
“We discover that each LN2 containers can deal with the Core i9-14900KF with P-cores clocked to 7.4 GHz with none challenge. It appeared the AI-generated design may maybe maintain 7.5 GHz only a tad longer. However that may simply be run-to-run variation,” they famous.
Within the second take a look at, they checked the CPU temperature deltas between the warmth spreader and cooling container base to evaluate actual warmth switch capabilities. There was additionally an all-out stress take a look at, pushing over 600W by way of the chip for a number of minutes.
Whereas the AI container did pull forward a smidge, the beneficial properties had been comparatively muted in comparison with the theoretical take a look at outcomes. Temperature deltas between the CPU warmth spreader and container base had been tighter on the 3D-printed mannequin, however not by an earth-shattering margin. Even the efficiency uplift in Cinebench was pretty modest, as seen above.
After crunching the numbers, the group decided that whereas technically spectacular, the AI/3D printed design at present does not pencil out from a price/profit perspective for modest overclocking eventualities. Not with that $10,000 price ticket.
Nevertheless, they don’t seem to be completed but. Whereas “nothing concrete” is in hand, the group says they may look into efficiency and value optimizations. The design of the LN2 container does not essentially should be round, for instance. They’re additionally exploring new designs for higher-power CPUs just like the Ryzen Threadripper or Intel’s Xeon 6.
All in all, the feasibility examine could have uncovered some limitations, but it surely additionally proved generative AI has higher makes use of than merely churning out six-fingered models.
