HeadsUpAI

NVIDIA releases open source software to maximize AI factory energy efficiency

NVIDIA released two open-source software components for its DSX platform. The new DSX MaxLPS suite optimizes token performance per megawatt—AI output per unit of power—using liquid cooling and rack-level management. It launches alongside DSX OS, a modular operating system that automates the lifecycle and health of massive GPU clusters.
GPU Capacity Increase
Up to 40%
Cooling Specification
45-degrees-Celsius liquid cooling
Primary Metric
Token performance per megawatt
Software License
Open source
Early Adopters
CoreWeave, Lambda, and others

This update extends the Vera Rubin DSX architecture by providing the software layer to manage the power demands of the next-generation Vera Rubin platform. As data centers hit physical power limits, running up to 40% more GPUs within a fixed megawatt budget directly lowers the cost of inference (running a trained model to generate outputs).

Infrastructure builders can adopt these modular libraries to standardize multi-tenant operations. Cloud providers like CoreWeave and Lambda are already deploying these components to accelerate capacity rollouts. The software is available through the DSX ecosystem, supported by systems from partners including Dell Technologies and Supermicro.

View the full update on nvidianews.nvidia.com
NVIDIA Newsroom
NVIDIA Newsroom
@nvidianewsroom
X

Announced at #NVIDIAGTC Taipei: New DSX MaxLPS and DSX OS software for the NVIDIA DSX platform. This gives AI factory builders open-source software to improve energy efficiency and manage infrastructure at scale. ⬇️ https://t.co/0me4c4JJp0

28retweets165likes
View on X

Still wondering? A few quick answers below.

NVIDIA DSX MaxLPS is a software suite designed to maximize the number of AI tokens generated per megawatt of power. By combining 45-degree-Celsius liquid cooling with rack-level optimizations, it allows data center operators to run up to 40% more GPUs within their existing power and cooling constraints.

NVIDIA DSX OS is an open-source, modular operating system purpose-built for managing AI factories at scale. It provides essential infrastructure services including lifecycle management, intelligent workload scheduling, health automation, and multi-tenant operations, ensuring that massive GPU clusters remain resilient and perform consistently during production AI workloads.

The software improves efficiency by identifying the most energy-efficient operating point for GPUs within a fixed power budget. By using advanced liquid cooling and in-rack technologies to manage thermal loads, DSX MaxLPS minimizes the performance impact of power constraints, effectively lowering the total cost per AI token.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Share this update