Visuals courtesy of Switch

‍

At CoreWeave, we don’t follow a traditional roadmap for building AI infrastructure. We’re pioneering AI  infrastructure while engineering faster and smarter. Building on our legacy as the first AI cloud provider to provide access to the NVIDIA HGX H100 system, NVIDIA H200, and NVIDIA GB200 NVL72, we continue to accelerate the pace of AI innovation with an industry-first bring-up of NVIDIA’s latest cutting-edge platform, NVIDIA GB300 NVL72, which is housed within Dell’s integrated rack scale system.

Setting a new standard for AI reasoning and agentic performance

The NVIDIA GB300 NVL72 represents a significant leap in performance for AI reasoning workloads, delivering up to a 10x boost in user responsiveness and a 5x improvement in throughput per watt compared to the previous generation NVIDIA Hopper architecture. That translates into a staggering 50x increase in output for reasoning model inference, empowering you to develop and deploy larger, more complex AI models that are exponentially faster than ever before. Accompanied by 1.5x more dense FP4 performance and 2x higher NVIDIA Quantum-X800 InfiniBand speeds, the GB300 NVL72 enables unprecedented speed for inference that will define the next generation of AI applications.

“At Moonvalley, we are building the future of generative filmmaking to empower creators with tools that bring their imagination to life," said Naeem Talukdar, Co-Founder and CEO of Moonvalley. “The bringup of the GB300 on CoreWeave is incredibly exciting. This platform and CoreWeave's ability to quickly bring it to market via its AI cloud platform, are a major enabler for AI-driven filmmaking to become a reality for our industry."

CoreWeave enables customers to  harness the full potential of the GB300 NVL72 with its purpose-built cloud platform, allowing customers to accelerate their AI development. Built for the demands of frontier-scale AI, CoreWeave empowers teams to train, optimize, and deploy multi-trillion parameter models with maximum performance, reliability, and resiliency.

We’ve gone beyond the hardware to ensure every component in the CoreWeave AI cloud, from our Kubernetes service (CKS) and Slurm on Kubernetes offering (SUNK) to our deep observability and custom-designed Rack LifeCycle Controller (RLCC) is optimized to maximize GB300 NVL72 performance. This now includes deeper visibility with our new Cabinet Wrangler and Cabinet Details dashboard for insights at a rack scale. In addition, hardware-level data and cluster health events are now streamed directly through Weights & Biases models, empowering AI researchers to instantly correlate infrastructure issues with their training runs, rapidly resolve disruptions, and get back to pushing the boundaries of AI.

CoreWeave’s unique approach to infrastructure has consistently proven to amplify hardware performance. Earlier this year, CoreWeave published multiple groundbreaking performance benchmarks for NVIDIA GB200 NVL72-based instances on the CoreWeave cloud. This hyperfocus on AI workloads is how our GB200-based instances delivered 2.86x better per-GPU inference performance compared to H100-based instances. By integrating the GB300 NVL72 into our cloud portfolio, backed by our best-in-class infrastructure and engineering expertise, we provide an unparalleled platform experience that is fine-tuned for the most demanding AI and HPC applications.

A look inside the NVIDIA GB300 NVL72 

The CoreWeave GB300 NVL72 is a rack-scale solution, liquid-cooled platform designed to meet the most challenging AI reasoning demands. It unifies 72 NVIDIA Blackwell Ultra GPUs, 36 Arm-based NVIDIA Grace CPUs, and 36 NVIDIA BlueField-3 DPUs  into a single, powerful platform, fueled by a wide range of NVIDIA’s industry-leading innovations: 

• NVIDIA Blackwell Ultra GPUs: These GPUs feature 1.5x more AI compute FLOPS compared to Blackwell GPUs.
• Expanded HBM3e Memory: With up to 21TB of high-bandwidth GPU memory per rack, GB300 NVL72 boasts 1.5X more GPU memory compared to GB200 NVL72 and allows for larger batch sizes and larger models, which is crucial for maximizing throughput in AI reasoning.
• Fifth-Generation NVIDIA NVLink: The latest iteration of NVLink provides a staggering 130TB/s of aggregate bandwidth, ensuring seamless, high-speed communication between every GPU in the system for peak-performance AI models.
• Next-Generation InfiniBand Networking: The platform is equipped with NVIDIA Quantum-X800 InfiniBand switches and NVIDIA ConnectX-8 SuperNICs, providing a massive 800 gigabits per second (Gb/s) of dedicated network connectivity to each GPU. This ensures best-in-class remote direct-memory access (RDMA) for maximum efficiency in large-scale AI workloads.
• Secure, Multi-Tenant Cloud Networking: The  NVIDIA DOCA software framework, running on NVIDIA BlueField-3, accelerates AI workloads by providing line-speed tenant networking up to 200Gb/s and high-performance GPU data access.

This system is engineered to handle the massive computational demands of test-time scaling inference, a critical component for deploying state-of-the-art AI models. For our customers, this translates to a significant leap in performance for AI reasoning workloads. 

Built in collaboration with our partners

Pioneering the future of AI infrastructure is a team sport. While our teams at CoreWeave work tirelessly to engineer and integrate these platforms, our success is only possible because of our diverse partner ecosystem, which operates with the same levels of urgency and vision. Our joint collaboration with Dell Technologies, Switch, and Vertiv on this deployment of the NVIDIA GB300 NVL72 is fundamental to our speed and agility, providing the critical support we need to turn groundbreaking technology into a reality for our customers at an unprecedented pace.‍

The future of AI is now

This first deployment of the NVIDIA GB300 NVL72 on CoreWeave’s cloud is just the beginning of a much bigger story. As AI models continue to rapidly grow in size and complexity, the need for purpose-built AI infrastructure will only continue to grow at the same pace. We are excited to drive this transformation, empowering the planet’s leading AI labs and enterprises to continuously push the boundaries of what’s possible.

Stay tuned for more updates as we move towards broader availability. The next generation of AI is here, and it’s being built on CoreWeave.