Frore Systems claims diamond cooling cuts AI GPU temps by 22C and boosts efficiency 75%

Frore Systems revealed LiquidJet Diamond, a cooling solution that could reshape how data centers manage extreme heat from next-generation AI chips. Buyers and operators should care because the technology targets the thermal limits of GPUs that are pushing past 2,000 watts. This shift matters for anyone building or running large-scale AI clusters that currently struggle with power density.

Diamond-embedded cold plate targets 2,300W AI accelerators with unconfirmed performance gains

The product is a direct-to-chip cold plate that embeds diamond spreaders into its bottom surface. Frore designed this hardware specifically for future AI accelerators, including Nvidia's Rubin Ultra and Feynman, as well as AMD's MI455 and MI500. The company demonstrated the cooling system using a 2,300 W Nvidia Rubin GPU at the OCP Global Summit 2026.

LiquidJet Diamond Specifications

• Hotspot Density Handling: 770 W/cm²
• Heat-Transfer Efficiency: 75% greater than conventional coldplates
• GPU Temperature Reduction: 16–22°C lower
• AI Tokens per Watt: 30–35% more
• Demonstration Power: 2,300 W Nvidia Rubin GPU

Frore claims the diamond spreaders allow the cold plate to handle hotspot densities of up to 770 W/cm² under specific test conditions. The company states this design offers 75% greater heat-transfer efficiency compared to conventional multi-inlet skived coldplates. These efficiency gains reportedly result in GPU temperatures that are 16 to 22°C lower than standard solutions.

Beyond temperature control, the cooling technology reportedly delivers 30% to 35% more AI tokens per watt. These performance metrics remain unconfirmed by independent benchmarks, as the figures come directly from Frore's internal testing. It is also unclear at this time how widely the industry will adopt LiquidJet Diamond.

Frore Systems has positioned LiquidJet Diamond as a critical component for the next wave of high-power AI hardware. The company has shown the technology works with 2,300 W GPUs, but independent validation and market adoption rates are still pending.

Source: NotebookCheck