NVIDIA GB300 NVL72

Built for the Age of AI Reasoning

Revolutionary AI inference at scale with unprecedented performance and efficiency.

img

NVIDIA GB300 NVL72

NVIDIA GB300 NVL72 is the cornerstone for modern enterprise AI

The NVIDIA GB300 NVL72 platform, featuring a fully liquid-cooled rack-scale design, integrates 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace CPUs into one powerful solution optimized for AI reasoning. Offering extraordinary throughput and responsiveness, the GB300 NVL72 transforms AI factories, delivering a 50x increase in inference output compared to NVIDIA Hopper™ platforms.

img
img

Performance

Performance Highlights

Experience breakthrough compute performance with cutting-edge GPU acceleration, ultra-fast interconnects, and optimized power efficiency—engineered to tackle the most complex AI, data science, and high-performance computing workloads.

AI Factory Output: 50x increase compared to Hopper

User Responsiveness (TPS per user): 10x faster

Throughput (TPS per MW): 5x more efficient

Storage Capacity & GPU-to-GPU Bandwidth

Features

Key Features

The NVIDIA GB300 NVL72 is designed for the most demanding AI and HPC workloads, delivering exascale performance with next-gen NVLink interconnects, massive GPU memory, and seamless multi-node scaling—ideal for training trillion-parameter models and real-time inference at unprecedented speed.

Blackwell Ultra Architecture

Massive HBM3e Memory

Next-Generation Networking

Advanced Liquid-Cooling Design

Specifications

NVIDIA GB300 NVL72 Specifications

FeatureSpecification
Configuration72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs
NVLink Bandwidth130 TB/s
Fast MemoryUp to 40 TB
GPU Memory & BandwidthUp to 21 TB / 576 TB/s
CPU Memory & Bandwidth18 TB SOCAMM LPDDR5X / 14.3 TB/s
CPU Core Count2,592 Arm Neoverse V2 cores
FP4 Tensor Core Performance1,400 PFLOPS (with sparsity), 1,100 PFLOPS (without)
FP8/FP6 Tensor Core720 PFLOPS
INT8 Tensor Core23 PFLOPS
FP16/BF16 Tensor Core360 PFLOPS
TF32 Tensor Core180 PFLOPS
FP326 PFLOPS
FP64 & FP64 Tensor Core100 TFLOPS

Preliminary specifications subject to change.

Applications

Ideal Applications

  1. AI reasoning and real-time inference at massive scale
  2. Large language model (LLM) inference workloads
  3. Generative AI deployment and scalability
  4. Advanced scientific research and simulations
logo

Built on years of expertise in digital innovation, we have established ourselves as a trusted name in AI infrastructure.

Address

P.O.Box 108093 Al Moroor street Abu Dhabi University Building

Email Address

info@centeraivision.com

Number Phone

+971 50 108 0066

© 2024-2025, All Rights Reserved