Zutacore Logo White Update

ASRockRack_LOGO_RGB-1

Deploying NVIDIA HGX B300 at Scale

As AI server power densities increase with platforms like NVIDIA HGX B300, traditional cooling approaches struggle to sustain performance, reliability, and cost efficiency at scale.  

Together with ASRock Rack, we examine the real-world infrastructure and thermal challenges introduced by NVIDIA HGX B300-class AI systems. Through example see  how a combined server and waterless, two-phase direct-to-chip cooling architecture enables predictable performance, faster time-to-deploy, and lower operational risk.

Key Takeaways:

  • The technical and commercial challenges of deploying NVIDIA HGX B300-class AI infrastructure

  • Where air and single-phase liquid cooling increase cost, complexity, and deployment risk

  • How ASRock Rack’s NVIDA HGX B300 server architecture accelerates scalable AI system design

  • Why two-phase direct-to-chip cooling improves uptime, reduces throttling, and lowers TCO

  • How this solution compares to RDHx, immersion, and hybrid cooling from a cost and operability perspective

Guest Speakers

  • Andrew Tsao: Product Marketing & Development Manager, ASRock Rack.

  • Tobias Quaranta: Global Solution Architect Team Lead, ZutaCore

Watch On-Demand Now

Your Questions, Answered

Additional insights addressing questions raised during the live webinar.

 Q:  What liquid coolant does the system use? 

A: ZutaCore’s two-phase cooling solution uses R1233zd(E), a dielectric liquid with very low global warming potential (GWP) and excellent thermodynamic properties for high-power semiconductor cooling. Because the fluid boils directly at the chip, it removes heat extremely efficiently while operating at low system pressure and low flow rates compared with traditional liquid cooling methods.

Q: Is there any risk of  the liquid coolant pressure falling below atmospheric pressure when the system is idle in cooler environments (10–20°C)? 

A: No. The system is designed to operate above atmospheric pressure across a wide range of typical data-center temperature ranges. The liquid’s thermodynamic properties and the sealed system architecture ensure stable operation, preventing air ingress and maintaining system integrity even during idle or low-load conditions.

Q: How does two-phase cooling simplify deployment compared with other liquid cooling solutions? 

A: Two-phase direct-to-chip cooling simplifies deployment because:

  • Lower flow rates are required compared with single-phase liquid cooling
  • Smaller piping and pumps are needed due to latent heat transfer
  • Thermal performance is self-regulating through the boiling process
  • Infrastructure complexity is reduced, enabling easier scaling of high-density racks

This allows operators to deploy high-power AI infrastructure with less operational risk and simpler facility integration.

Q: Is the 1.2 MW CDU currently available?

A: Yes. ZutaCore ’s 1.2 MW End-of-Row CDU is part of the company’s cooling infrastructure portfolio designed for high-density AI deployments and supports large-scale rack deployments. 

Q: Are retrofit options available for existing air-cooled data centers?

A: Yes. ZutaCore solutions can be deployed in both new green and brown field environments.

Existing racks can be upgraded with direct-to-chip cold plates and CDUs, allowing operators to support higher-power AI workloads without fully rebuilding facility infrastructure

Q: When will the ASRock Rack + ZutaCore®  NVIDIA HGX B300 platform be available for testing? 

A: The ASRock Rack & ZutaCore NVIDIA HGX B300 is available now for testing.