CoreWeave launched its NVIDIA GB200 NVL72 cluster, establishing itself as a key player in AI cloud infrastructure. This deployment marks one of the industry’s earliest implementations of the NVIDIA GB200 Grace Blackwell Superchip, enabling the cluster to handle large-scale generative AI workloads, simulations, and real-time analytics. Capable of delivering up to 1.4 exaFLOPS of compute power per rack, the system offers unprecedented efficiency with features like liquid cooling and 13.5TB of high-bandwidth GPU memory per rack.
In addition, CoreWeave introduced new GPU instance types, including NVIDIA GH200 Grace Hopper Superchip, L40, and L40S GPUs. The GH200 combines high-bandwidth memory with a CPU-GPU integration that eliminates PCIe bottlenecks, while the L40 and L40S deliver balanced performance and cost efficiency for AI and visualization tasks. CoreWeave also announced the preview of its AI Object Storage service, an exabyte-scale solution designed to accelerate data-intensive AI workflows with proprietary Local Object Transport Accelerator (LOTA) technology, achieving up to 2 GBps per GPU in throughput.
These offerings are now available in CoreWeave’s US-based regions and integrate seamlessly with its Kubernetes-based management services. The company plans to expand availability and refine its data solutions in 2024 and beyond.
• NVIDIA GB200 NVL72 cluster: 1.4 exaFLOPS of compute per rack, 13.5TB of GPU memory, liquid cooling for energy efficiency.
• NVIDIA GH200 instances: 96GB HBM3 memory, 4TBps bandwidth, 72 Arm CPU cores, 7TB NVMe storage.
• L40 and L40S GPUs: Enhanced AI inference and visualization; FP8 support on L40S boosts training throughput by up to 1.4x.
• AI Object Storage: Exabyte scale, 2 GBps per GPU throughput, enterprise-grade features like encryption and S3 compatibility.
• Global expansion: Available in US-EAST-04 and RNO2 regions; additional rollouts planned for 2024–2025.
CoreWeave CTO Brian Venturo commented, “Our latest innovations deliver the power and efficiency needed to accelerate AI research and development at scale.”
