Frore Introduces Diamond-Infused Coldplates for High-Power AI Chips
Frore Systems launches LiquidJet Diamond coldplates that use diamond wafers to lower GPU temperatures by up to 22C for next-gen AI servers.
Key points
- Frore Systems releases LiquidJet Diamond coldplates using diamond wafers to manage heat on high-power AI accelerators.
- The solution claims to reduce GPU temperatures by 16C to 22C compared with conventional coldplates.
- Diamond integration allows handling of hotspot power densities up to 770 W/cm², surpassing previous copper-based limits.
Frore Systems has introduced the LiquidJet Diamond series of liquid-cooling coldplates designed for next-generation artificial intelligence accelerators. According to Tom's Hardware, the bespoke cooling components utilize diamond surfaces to address precise thermal hotspots on processors such as Nvidia's Rubin and AMD's Instinct MI400 and MI500-series chips.
The company states that these new coldplates can reduce GPU temperatures by up to 22 degrees Celsius compared with conventional cooling methods. Frore claims this thermal improvement leads to a 30% to 35% increase in AI token throughput per watt. The solution targets the escalating power demands of modern data center hardware, where heat management is becoming a critical bottleneck for performance and efficiency in high-density computing environments.
Engineering and Thermal Performance
LiquidJet Diamond combines technology from Frore’s original LiquidJet product with innovations from the LiquidJet Nexus bespoke coldplate used for Nvidia Blackwell and Rubin servers. While Nexus integrates multiple coldplates into a single large assembly for several processors, LiquidJet Diamond functions as an individual, custom coldplate for each GPU. This design allows for more targeted cooling strategies tailored to the specific power distribution of individual processor units.
The internal architecture employs a three-dimensional, ultra-short-loop, multi-stage structure. Coolant channels are customized to match the power distribution profile of the specific processor being cooled. This approach ensures that cooling resources are directed precisely where heat generation is highest, rather than relying on uniform cooling across the entire chip surface. Such precision is essential for managing the intense localized heat loads found in advanced AI accelerators.
Diamond Integration and Conductivity
The primary innovation in LiquidJet Diamond is an additional heat spreader integrated into the bottom of the coldplate. This spreader incorporates diamond wafers positioned directly above the GPU’s hottest areas. These wafers distribute heat more efficiently before it reaches the liquid-cooling structure. Diamond, which is carbon-based, possesses exceptionally high thermal conductivity that can exceed 20 W/cm·K.
In comparison, copper has a thermal conductivity of approximately 4 W/cm·K. According to Frore, this material advantage allows LiquidJet Diamond to handle hotspot power densities up to 770 W/cm². This is a significant increase from the approximately 600 W/cm² limit of the original LiquidJet coldplates. The company has not disclosed the specific manufacturing or integration methods used to embed the diamond wafers into the coldplate assembly.
What should you do?
Data center operators and AI hardware engineers should evaluate current thermal headroom against projected power densities for upcoming chip generations. Teams planning deployments of Nvidia Rubin or AMD Instinct MI500-series hardware should review cooling specifications with vendors to ensure compatibility with high-density workloads. Organizations should monitor industry developments in liquid cooling technologies to assess whether diamond-infused solutions offer necessary performance gains for their specific operational requirements.
What to do and how to stay safe: Frore Systems
- Audit current server cooling infrastructure to determine if it can handle power densities approaching 770 W/cm².
- Consult with cooling solution vendors about the availability and compatibility of bespoke coldplates for next-generation AI accelerators.
- Review thermal performance metrics and token throughput efficiency claims when evaluating new hardware procurement options.
- Establish clear thermal management protocols for high-power GPU deployments to prevent overheating and ensure sustained performance.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How much cooler do Frore's new coldplates make GPUs compared to standard cooling?
Frore claims the LiquidJet Diamond solution reduces GPU temperatures by 16C to 22C compared with conventional coldplates.
What chips are compatible with the LiquidJet Diamond coldplates?
The coldplates are designed for AMD's Instinct MI400 and MI500-series, Nvidia's Rubin, and other next-generation AI accelerators.
Why does Frore use diamond in these coldplates?
Diamond has higher thermal conductivity than copper, allowing the coldplates to handle higher power densities of up to 770 W/cm².
