The NVIDIA GB300 NVL72 combines the groundbreaking performance of 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs into a single rack-scale, liquid cooled architecture. As AI agents are tasked with accomplishing increasingly complex goals, the growing size of large context windows required for AI reasoning demands far greater compute power – up to 100 times more needed than for single-pass inference. The NVIDIA GB300 NVL72 is a powerful solution to these challenges, offering up to 50 times greater AI factory output compared to an NVIDIA HGX™ H100.
Deploy nowCapacity available soon
Reserve capacity at Vultr now to access the latest in NVIDIA acceleration.
Key features
Combines NVIDIA Blackwell Ultra GPUs and NVIDIA Grace CPUs in a single architecture for up to 1,440 PFLOPS FP4 AI performance
Features 279 GB HBM3E GPU memory and 8 TB/s memory bandwidth per GPU, with a total of 37 TB total fast memory
Connected by NVIDIA 5th Generation NVLink™ 1.8 TB/s interconnect
Vultr's enterprise-ready infrastructure seamlessly supports any cluster size of NVIDIA GB300 NVL72. Whether you require a small cluster or a massive deployment, Vultr ensures reliable, high-performance computing to meet your specific needs.
Large clusters of NVIDIA Blackwell Ultra GPUs are available where you need them, thanks to Vultr’s far-reaching infrastructure. With 33 global cloud data center regions across six continents, we guarantee low latency and high availability, enabling your enterprise to achieve optimal performance worldwide.
Vultr ensures our platform, products, and services meet diverse global compliance, privacy, and security needs, covering areas such as server availability, data protection, and privacy. Our commitment to industry-wide privacy and security frameworks, including ISO, HIPAA and SOC 2 Type 2 standards, demonstrates our dedication to protecting our customers' data.
The NVIDIA GB300 NVL72 provides the computational acceleration required for modern AI training and inference – with up to 70x greater AI FLOPs and a 30x overall increase in AI Factory output performance for reasoning models like DeepSeek R1 compared to Hopper-generation GPUs.
No information is required for download
Vultr Clusters provides scalable NVIDIA GB300 NVL72 clusters with high-speed GPU fabric. Easily provision and configure clusters through the Vultr Console or API, and scale infrastructure up or down on-demand to match varying workload needs. Create, configure, and manage clusters without manual requests or reservations. Seamlessly add high-performance storage, monitor key cluster metrics, and schedule workloads with Slurm or Kubernetes. Both Vultr Cloud GPU and Vultr Bare Metal are compatible with Vultr Clusters.
No information is required for download
The NVIDIA GB300 NVL72 delivers rack-scale performance for AI agents – and more
The NVIDIA GB300 NVL72 on Vultr provides the compute capacity to accelerate the latest complex computational workloads. Leverage NVIDIA GB300 NVL72s on Vultr to power AI and HPC applications affordably and efficiently on a composable, global platform.
30x Greater*
30x Faster*
15x Lower*
*As compared to NVIDIA Hopper GPUs
NVIDIA GB300 NVL72
NVIDIA Blackwell GPUs
72
NVIDIA Grace CPUs
36
Total Fast Memory
37 TB
Total Memory Bandwidth
576 TB/s
FP4 Tensor Core
1,440 petaFLOPS¹
FP8/FP6 Tensor Core
720 petaFLOPS¹
GPU Memory
Up to 279 GB HBM3E per GPU
Interconnect
5th Generation NVIDIA NVLink™: 1.8 TB/s
¹Specification in sparse.
The NVIDIA GB300 NVL72 is a system that combines 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs in one rack-scale, liquid cooled architecture.
NVIDIA GB300 NVL72s provide breakthrough AI reasoning performance. They deliver up to 30x greater AI factory output performance, 30x greater energy efficiency, 15x lower cost per token, and 70x greater AI FLOPs compared to NVIDIA Hopper-generation platforms, along with 37 TB of fast memory per rack and 800 GB/s of data compression.
NVIDIA GB300 NVL72s are well-equipped to handle AI inference and training, agentic AI, and High Performance Computing workloads.
Growing AI complexity means multi-step reasoning has become a crucial step for AI workloads. Reasoning requires extremely large context windows, which require far greater compute capacity (up to 100x greater) than do single-pass inference queries. NVIDIA GB300 NVL72s are built to support these large context windows.
Vultr supports NVIDIA GB300 NVL72 clusters of any size across 33 global cloud data center regions, reaching 90% of the world’s population in 2-40ms. Vultr provides advantages in cost predictability, price-to-performance, ease of deployment, and management simplicity, allowing customers to deploy these advanced systems efficiently, affordably, scalably, and predictably.
Vultr GPU Enabled Images include prepackaged NVIDIA software like the NVIDIA CUDA toolkit. This makes setup and deployment even easier on Vultr’s simple control panel or API.
The NVIDIA GB300 NVL72 is a system that combines 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs in one rack-scale, liquid cooled architecture.
