Advancing AI With Breakthrough Performance and Scalability
At GTC 2025, NVIDIA introduced the Blackwell Ultra AI Factory Platform, designed to power the next generation of AI reasoning, agentic AI, and physical AI. Built on the foundation of Blackwell architecture, this platform enhances AI training and inference, enabling models to scale dynamically for higher-quality responses and improved efficiency.
NVIDIA Blackwell Ultra
The NVIDIA GB300 NVL72 connects 72 Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs in a rack-scale design, creating a single massive GPU optimized for test-time scaling and AI model reasoning. This architecture enables AI-driven problem-solving with improved accuracy and multi-step processing capabilities.
Key performance improvements include:
- 1.5x More AI Performance than the NVIDIA GB200 NVL72.
- 11x Faster Inference on large language models compared to previous generations.
- 7x More Compute Power and 4x Larger Memory than Hopper-based systems.
The NVIDIA HGX B300 NVL16 platform further enhances AI model performance, minimizing latency and optimizing large-scale inference workloads.
Optimized Networking for AI at Scale
To support its growing AI infrastructure, Blackwell Ultra seamlessly integrates with NVIDIA Spectrum-X™ Ethernet and Quantum-X800 InfiniBand. These networking platforms offer 800 Gb/s throughput per GPU, ensuring smooth communication between GPUs and eliminating bottlenecks in AI model processing.
Additional Features:
- NVIDIA ConnectX®-8 SuperNIC for high-speed, low-latency GPU interconnects.
- NVIDIA BlueField®-3 DPUs for multi-tenant networking, AI-driven data acceleration, and real-time cybersecurity.
NVIDIA Dynamo
Alongside hardware advancements, NVIDIA introduced Dynamo, an open-source AI inference framework designed to maximize performance and reduce AI inference costs. Dynamo optimizes GPU utilization across large-scale deployments, accelerating reasoning AI models while lowering operational overhead.
Build AI-Optimized Infrastructure With AMAX
As an NVIDIA Elite Partner, AMAX delivers DGX and HGX solutions equipped with the latest Blackwell-based GPUs for AI and high-performance computing (HPC).
Visit AMAX at GTC 2025
Visit AMAX at GTC 2025, Booth #2018, from March 17-21, showcasing the latest advancements in AI and liquid cooling technologies. Attendees can experience a live pumped two-phase liquid cooling demonstration and learn how to build a secure on-prem AI chatbot with NVIDIA AI Workbench, presented by AMAX’s engineering team.