Two-Phase Direct-to-Chip Cooling
Product · Hardware / device by Orbital Industries
The thermal architecture for next-generation AI silicon.
Two-Phase Direct-to-Chip Cooling is a high-density thermal architecture for AI data centers that uses phase-change refrigerants to cool chips exceeding 2,000 W TDP. It is designed for next-generation accelerators like NVIDIA Vera Rubin, offering higher heat absorption and lower flow rates than single-phase water cooling.
Category: Data Centers
- Performance
- Supports chip TDP of 2,000 W+
- Operating conditions
- Boiling point matched to target GPU operating temperature
- Cooling Method
- Two-Phase Direct-to-Chip (2P-D2C)
- Heat Transfer Physics
- Latent heat of vaporization
- Target Chip TDP
- 2,000 W+
- Fluid Type
- Dielectric refrigerant
- Supports chip TDP of 2,000 W+
- Uses latent heat of vaporization for phase-change cooling
- Dielectric refrigerant prevents short-circuit risks
- Lower flow rates reduce component wear and plumbing strain
- Engineered for NVIDIA Vera Rubin and future AMD accelerators
Overview
Two-Phase Direct-to-Chip Cooling (2P-D2C) is Orbital IT's proprietary liquid cooling architecture designed for next-generation AI data centers. It solves the physical limits of air and single-phase water cooling by using a dielectric refrigerant that boils directly on the GPU cold plate. This phase-change process absorbs heat at a near-constant temperature using the latent heat of vaporization, enabling cooling for chips with thermal design power (TDP) exceeding 2,000 W. The system is engineered to support the highest density GPU racks, such as those required for NVIDIA Vera Rubin and future accelerators, while operating at lower flow rates than traditional liquid systems. Orbital IT provides these solutions to data center operators and infrastructure buyers who need to support the thermal requirements of future AI silicon.
Thermal Engineering
- Latent Heat of Vaporization: The system uses a dielectric refrigerant that boils directly on the GPU cold plate to absorb heat through phase change.
- 2,000 W+ TDP Support: The architecture is engineered to support chip thermal design powers (TDP) exceeding 2,000 W.
- Dielectric Fluid Safety: The cooling fluid is electrically insulating, allowing it to contact GPU components directly without short-circuit risk.
- Reduced Flow Rates: The system operates at lower flow rates at the chip compared to single-phase systems, reducing component wear and plumbing strain.
System Architecture
- GPU Roadmap Alignment: The architecture is purpose-engineered for next-generation AI silicon, including NVIDIA Vera Rubin and future AMD accelerators.
- Integrated Hardware Components: The solution includes proprietary cold plates, manifolds, and CDUs redesigned for phase-changing fluids.
- Infrastructure Integration: The system is designed to fit into data center infrastructure engineered for high rack densities.
- Chemical Stability: The refrigerant is selected to survive thousands of boiling cycles without breaking down or fouling cold plates.
Use cases
- Data center operators deploy 2P-D2C to support NVIDIA Vera Rubin and Rubin Ultra accelerators with TDPs over 2,000 W.
- Infrastructure engineers use the dielectric fluid system to eliminate short-circuit risks during high-density liquid cooling deployments.
- Facility managers implement two-phase cooling to reduce pump sizes and plumbing strain through lower volumetric flow rates.
Specifications
- Performance
- Supports chip TDP of 2,000 W+
- Compatibility
- NVIDIA Vera Rubin, Rubin Ultra, Feynman, and next-generation AMD accelerators
- Operating conditions
- Boiling point matched to target GPU operating temperature
- Cooling Method
- Two-Phase Direct-to-Chip (2P-D2C)
- Heat Transfer Physics
- Latent heat of vaporization
- Target Chip TDP
- 2,000 W+
- Fluid Type
- Dielectric refrigerant
- Electrical Property
- Electrically insulating
- Hardware Components
- Cold plates, manifolds, CDUs, quick-disconnects
How it works
- Technical Scoping: Orbital IT conducts a technical scoping session based on the customer's deployment timeline, site constraints, and target rack density.
- Custom Proposal: A proposal is developed with a cooling architecture designed for the specific high-density AI silicon being deployed.
- Infrastructure Integration: The system, including proprietary cold plates, manifolds, and CDUs, is integrated into the data center infrastructure.
FAQ
How does two-phase cooling differ from single-phase liquid cooling?
Two-phase cooling uses a dielectric refrigerant that boils on the chip, absorbing heat through the latent heat of vaporization at a near-constant temperature. This allows it to handle much higher heat densities than single-phase water cooling, which relies on sensible heating (temperature rise).
What is the maximum chip power this system can cool?
The system is engineered for thermal design powers (TDP) of 2,000 W or more per chip. This is intended for future accelerators like NVIDIA Vera Rubin and Rubin Ultra where single-phase cooling reaches its physical limits.
Is the coolant safe for electronic components?
Orbital IT uses a dielectric refrigerant that is electrically insulating. This ensures that if the fluid contacts GPU components directly, there is no risk of a short circuit.
What are the infrastructure benefits of lower flow rates?
Two-phase systems operate at lower flow rates at the chip compared to single-phase systems. This results in smaller pumps, less component wear, and less strain on the plumbing infrastructure.
Added by AntSpace from public sources (orbitalindustries.com).