{"id":6486,"date":"2025-09-12T06:25:33","date_gmt":"2025-09-12T06:25:33","guid":{"rendered":"https:\/\/openzeka.com\/en\/?page_id=6486"},"modified":"2025-09-12T06:26:19","modified_gmt":"2025-09-12T06:26:19","slug":"nvidia-gb300-nvl72","status":"publish","type":"page","link":"https:\/\/openzeka.com\/en\/nvidia-gb300-nvl72\/","title":{"rendered":"NVIDIA GB300 NVL72"},"content":{"rendered":"

NVIDIA GB300 NVL72<\/strong><\/h1><\/div>

Built for the age of AI reasoning.<\/h2><\/div><\/div><\/div><\/div><\/div>

NVIDIA GB300 NVL72<\/strong><\/h2><\/div>

Built for the age of AI reasoning.<\/p>\n

<\/div><\/h2><\/div><\/div><\/div><\/div><\/div>

NVIDIA GB300 NVL72<\/strong><\/h2><\/div>

Built for the age of AI reasoning.<\/h2><\/div><\/div><\/div>
<\/span><\/div><\/div><\/div><\/div><\/div>

Overview<\/strong><\/h4><\/div>

Designed for AI Reasoning Performance<\/strong><\/h1><\/div>

The NVIDIA GB300 NVL72 features a fully liquid-cooled, rack-scale design that unifies 72\u00a0NVIDIA Blackwell Ultra\u00a0GPUs and 36 Arm\u00ae-based\u00a0NVIDIA Grace\u2122 CPUs\u00a0in a single platform optimized for test-time scaling inference.\u00a0AI factories\u00a0powered with the GB300 NVL72 using NVIDIA Quantum-X800 InfiniBand or Spectrum\u2122-X Ethernet paired with ConnectX\u00ae-8 SuperNICS provide a 50x higher output for reasoning model inference compared to the\u00a0NVIDIA Hopper\u2122 platform.<\/p>\n<\/div><\/div><\/div><\/div><\/div>

Performance<\/strong><\/h4><\/div>

Scaling AI Factories to Unprecedented Levels<\/strong><\/h2><\/div>
<\/span><\/div>

DeepSeek R1 ISL = 32K, OSL = 8K, GB300 NVL72 with FP4 Dynamo disaggregation. H100 with FP8 in-flight batching. Projected performance subject to change.<\/p>\n<\/div>

Experience next-level AI reasoning performance with the NVIDIA GB300 NVL72 platform. Compared to Hopper, the GB300 NVL72 delivers an impressive 10x boost in user responsiveness (TPS per user) and a 5x improvement in throughput (TPS per megawatt (MW)). Together, these advancements translate into a remarkable 50x leap in overall AI factory output.<\/p>\n<\/div><\/div><\/div><\/div><\/div>

Features<\/strong><\/h4><\/div>

Technological Breakthroughs<\/strong><\/h2><\/div><\/div><\/div>
<\/span><\/div>

AI Reasoning Inference<\/strong><\/p>\n<\/div>

Test-time scaling and AI reasoning increase the compute necessary to achieve quality of service and maximum throughput. NVIDIA Blackwell Ultra\u2019s\u00a0Tensor Cores\u00a0are supercharged with 2x the attention-layer acceleration and 1.5x more AI compute floating-point operations per second (FLOPS) compared to NVIDIA Blackwell GPUs.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div>

288 GB of HBM3e<\/strong><\/p>\n<\/div>

Larger memory capacity allows for larger batch sizing and maximum throughput performance.NVIDIA Blackwell Ultra GPU\u2019s offer 1.5x larger HBM3e memory in combination with added AI compute, boosting AI reasoning throughput for the largest context lengths.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div>

NVIDIA Blackwell Architecture<\/strong><\/p>\n<\/div>

The NVIDIA Blackwell architecture delivers groundbreaking advancements in accelerated computing, powering a new era of unparalleled performance, efficiency, and scale.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div>

NVIDIA ConnectX-8 SuperNIC<\/strong><\/p>\n<\/div>

The\u00a0NVIDIA ConnectX-8 SuperNIC\u2019s\u00a0input\/output (IO) module hosts two ConnectX-8 devices, providing 800 gigabits per second (Gb\/s) of network connectivity for each GPU in the NVIDIA GB300 NVL72. This delivers best-in-class remote direct-memory access (RDMA) capabilities with either\u00a0NVIDIA Quantum-X800 InfiniBand\u00a0or\u00a0Spectrum-X Ethernet\u00a0networking platforms, enabling peak AI workload efficiency.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div>

NVIDIA Grace CPU<\/strong><\/p>\n<\/div>

The NVIDIA Grace CPU is a breakthrough processor designed for modern data center workloads. It provides outstanding performance and memory bandwidth with 2x the energy efficiency of today\u2019s leading server processors.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div>

Fifth-Generation NVIDIA NVLink<\/strong><\/p>\n<\/div>

Unlocking the full potential of accelerated computing requires seamless communication between every GPU. The fifth-generation of\u00a0NVIDIA NVLink\u2122\u00a0is a scale\u2013up interconnect that unleashes accelerated performance for AI reasoning models..<\/p>\n<\/div><\/div><\/div><\/div><\/div>

NVIDIA GB300 Grace Blackwell Ultra Superchip<\/strong><\/h2><\/div>

As a building block for the NVIDIA GB300 NVL72 rack-scale solution, the NVIDIA GB300 Grace Blackwell Ultra Superchip features four NVIDIA Blackwell Ultra GPUs, two Grace CPUs, and four ConnectX-8 SuperNICs. Through\u00a0NVIDIA NVLink Switch\u00a0technology and NVIDIA BlueField\u00ae-3 DPUs, 18 superchips combine into one giant GPU, purpose-built for the age of AI reasoning.<\/p>\n<\/div><\/div><\/div>

<\/span><\/div><\/div><\/div><\/div><\/div>

Specifications<\/strong><\/h2><\/div>

NVIDIA GB300 NVL72\u00b9<\/strong><\/h2><\/div>\n
\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Feature<\/strong><\/th>\nSpecification<\/strong><\/th>\n<\/tr>\n<\/thead>\n
Configuration<\/td>\n72 NVIDIA Blackwell Ultra GPUs, 36 NVIDIA Grace CPUs<\/td>\n<\/tr>\n
NVLink Bandwidth<\/td>\n130 TB\/s<\/td>\n<\/tr>\n
Fast Memory<\/td>\nUp to 40 TB<\/td>\n<\/tr>\n
GPU Memory | Bandwidth<\/td>\nUp to 21 TB | Up to 576 TB\/s<\/td>\n<\/tr>\n
CPU Memory | Bandwidth<\/td>\nUp to 18 TB SOCAMM with LPDDR5X | Up to 14.3 TB\/s<\/td>\n<\/tr>\n
CPU Core Count<\/td>\n2,592 Arm Neoverse V2 cores<\/td>\n<\/tr>\n
FP4 Tensor Core<\/td>\n1,400 | 1,100\u00b2 PFLOPS<\/td>\n<\/tr>\n
FP8\/FP6 Tensor Core<\/td>\n720 PFLOPS<\/td>\n<\/tr>\n
INT8 Tensor Core<\/td>\n23 PFLOPS<\/td>\n<\/tr>\n
FP16\/BF16 Tensor Core<\/td>\n360 PFLOPS<\/td>\n<\/tr>\n
TF32 Tensor Core<\/td>\n180 PFLOPS<\/td>\n<\/tr>\n
FP32<\/td>\n6 PFLOPS<\/td>\n<\/tr>\n
FP64 \/ FP64 Tensor Core<\/td>\n100 TFLOPS<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

1. Preliminary specifications. May be subject to change. All Tensor Core specifications are with sparsity unless otherwise noted.
\n2. Without sparsity.<\/p>\n<\/div><\/div><\/div><\/div><\/div>

Get contact with us<\/p>\n<\/div><\/div><\/div>

CONTACT US<\/span><\/i><\/a><\/div>