Vultr is committed to delivering the latest in compute and GPU infrastructure. As each new generation of hardware provides an improvement in performance and capability, we understand the importance of accessing and leveraging the latest tools for supporting the ever-increasing demands of cutting-edge AI and high-performance computing (HPC) workloads. Companies need more than just infrastructure – they need a trusted partner dedicated to providing the resources necessary to host the most demanding workloads affordably.
As part of that ongoing commitment, Vultr is pleased to announce that preorders for NVIDIA GB300 NVL72 and NVIDIA HGX™ B300 early availability are now open.
NVIDIA Blackwell Ultra GPUs, used in both, are designed to meet the demands of complex inference queries, delivering support for larger models with exceptional efficiency. By choosing NVIDIA Blackwell Ultra GPUs on Vultr, get access to the most powerful NVIDIA offerings available through a simple, composable platform with global reach and transparent, predictable pricing.
Rack-scale architecture for AI reasoning: the NVIDIA GB300 NVL72
Containing 72 NVIDIA Blackwell Ultra GPUs and 36 NVIDIA Grace™ CPUs in an interconnected rack-scale solution, the NVIDIA Grace Blackwell Superchips provide unprecedented performance, including up to 1,440 FP4 PFLOPs.
Built to handle the computational demands of AI reasoning, the NVIDIA GB300 NVL72 features 37 TB of total fast memory and 576 TB/s total memory bandwidth for AI reasoning’s long context windows. Each system delivers up to a 30x overall increase in AI Factory output performance for large AI models compared to NVIDIA Hopper-generation GPUs, and up to 50x greater AI factory output, all with up to 30x greater energy efficiency.
Also available: NVIDIA HGX B300
Each NVIDIA HGX B300 contains eight Blackwell Ultra GPUs, connected by high-performance networking, providing 7x more AI compute than an NVIDIA HGX H100. With more than 2 TB of HBM3E memory, the NVIDIA HGX B300 is well-equipped to accelerate training and inference for the growing size and complexity of AI models, delivering up to 1.5x more dense FP4 FLOPs than an NVIDIA HGX B200.
The right cloud provider matters
All clouds are not made equal. Vultr stands out with its affordable, transparent pricing, unmatched price-to-performance, an extensive operating history spanning over a decade, and a comprehensive list of compliance certifications and attestations, confirming Vultr’s diligent attention to providing secure, trusted infrastructure.
Provisioning advanced hardware such as the NVIDIA GB300 NVL72 on Vultr is a smooth process with Vultr’s intuitive console or API. Vultr GPU Enabled Images, containing the latest NVIDIA drivers and CUDA software, accelerate the process. Vultr’s open, composable platform provides the flexibility to deploy the infrastructure needed without concerns about vendor lock-in. With a global presence, Vultr can offer availability wherever required, reaching 90% of the world’s population in 2-40ms. NVIDIA GPUs on Vultr can be consumed through Vultr Cloud GPU and Vultr Bare Metal, as well as through Vultr Kubernetes Engine. Clusters are available at any size.
When it comes to infrastructure, a trusted, mature platform matters.
Reserve the latest in NVIDIA GPUs today
These advanced new NVIDIA offerings will be available through Vultr soon. Ready to reserve? Get in touch to preorder for early NVIDIA GB300 NVL72 or NVIDIA HGX B300 availability on Vultr today.