NVIDIA and ST Present New Delivery Boards for 800VDC Architectures

Why It Matters

These boards provide data‑center designers with ultra‑compact, high‑efficiency converters, reducing footprint and energy loss in next‑gen compute platforms. Their performance accelerates adoption of 800 VDC architectures across AI and high‑performance computing workloads.

Key Takeaways

• 6 kW board delivers 800 V→12 V at 97.5% efficiency
• Power density reaches 2500 W per cubic inch
• 20 kW stacked LLC converter outputs 6 V at 96.5% efficiency
• Secondary side uses low‑voltage MOSFETs for compact design
• Enables efficient power for GPUs, CPUs, memory modules

Pulse Analysis

The shift toward 800 VDC power architectures reflects the industry’s drive for higher efficiency and lower thermal budgets in data‑center environments. GaN transistors, with their fast switching and low on‑resistance, enable converters that operate at hundreds of kilohertz while maintaining tight voltage regulation. By pairing high‑voltage GaN devices on the primary side with low‑voltage MOSFETs on the secondary, NVIDIA and ST achieve a balance between performance and component cost, a formula that has become critical as AI workloads demand ever‑greater power density.

The two delivery boards address distinct segments of the power‑delivery chain. The 6 kW, 850 kHz LLC module targets mid‑range subsystems, converting 800 V to 12 V for logic and peripheral rails, while boasting a 97.5% peak efficiency and a remarkable 2500 W per cubic inch. Its smaller counterpart, the 20 kW stacked LLC converter, focuses on high‑current, low‑voltage rails, delivering 6 V at 96.5% efficiency to power GPUs, CPUs, and memory modules. Both designs eliminate the need for intermediate conversion stages, simplifying board layouts and reducing overall system mass.

From a market perspective, these converters lower the total cost of ownership for hyperscale operators by cutting energy waste and freeing up board space for additional compute resources. The compact form factor aligns with the trend toward modular, blade‑style servers where every cubic inch counts. As AI accelerators continue to scale, the ability to efficiently step down from 800 VDC to low‑voltage rails will be a differentiator, positioning NVIDIA and ST as key enablers of the next wave of high‑performance computing infrastructure.