CoreWeave Introduces Cutting-Edge NVIDIA GB200 NVL72 Systems for AI Advancements
CoreWeave has taken a significant step forward in the realm of cloud computing by becoming one of the first cloud service providers to launch NVIDIA GB200 NVL72 systems on a large scale. This development is already being leveraged by forward-thinking AI companies like Cohere, IBM, and Mistral AI to enhance the training and deployment of next-generation AI models and applications.
CoreWeave’s pioneering move to make NVIDIA Grace Blackwell systems generally available marks a critical advancement in cloud technology. The company has demonstrated remarkable performance in MLPerf benchmarks with the NVIDIA GB200 NVL72, a robust computing platform designed for complex reasoning tasks and AI applications. As a result, CoreWeave’s clients are now gaining unprecedented access to thousands of NVIDIA Blackwell GPUs.
Mike Intrator, the CEO of CoreWeave, highlighted the collaboration with NVIDIA, emphasizing the swift provision of the most advanced solutions for AI model training and inference. He stated, "With the new Grace Blackwell rack-scale systems, many of our customers will be among the first to experience the innovative benefits and high performance offered by these AI technologies on a large scale."
These NVIDIA Blackwell GPUs are transforming raw data into actionable intelligence at an extraordinary pace, with more systems set to come online soon. CoreWeave is actively supporting its customers as they adapt to the rapid deployment of cloud-based systems built on NVIDIA Grace Blackwell technology. These systems are evolving into what are being referred to as "AI factories," capable of producing intelligence on a large scale and converting raw data into real-time insights with remarkable speed, accuracy, and efficiency.
AI Applications and Personalized AI Agents
A leading AI company, Cohere, is already harnessing the power of its Grace Blackwell Superchips to develop secure enterprise AI applications. Their enterprise AI platform, North, is designed to enable teams to build personalized AI agents that can automate workflows, surface insights in real-time, and more. With the NVIDIA GB200 NVL72 on CoreWeave, Cohere is experiencing up to three times the performance in training for models with 100 billion parameters compared to previous-generation NVIDIA Hopper GPUs—without needing specific Blackwell optimizations.
The enhanced performance is attributed to the large unified memory, FP4 precision, and a 72-GPU NVIDIA NVLink domain of the GB200 NVL72, which allows every GPU to operate in harmony. This results in significantly higher throughput and more efficient inference processes.
Autumn Moulder, Vice President of Engineering at Cohere, expressed satisfaction with the seamless integration of their workloads on the NVIDIA Grace Blackwell architecture, stating that it "unlocks incredible performance efficiency across our stack." She further noted that this integration allows their vertically integrated North application to run on a single Blackwell GPU and scale training tasks across thousands of GPUs, anticipating even greater performance with further optimizations.
AI Models for Enterprise Applications
IBM is leveraging one of the initial deployments of NVIDIA GB200 NVL72 systems, scaling up to thousands of Blackwell GPUs on CoreWeave, to train its next-generation Granite models. These open-source, enterprise-ready AI models deliver state-of-the-art performance while optimizing safety, speed, and cost efficiency. Supported by a robust partner network, Granite models provide the foundation for solutions like IBM watsonx Orchestrate, which enables enterprises to build powerful AI agents that automate and accelerate workflows across the organization.
CoreWeave’s deployment for IBM also incorporates the IBM Storage Scale System, delivering high-performance storage solutions for AI applications. This allows CoreWeave customers to access IBM’s storage platform within dedicated environments and AI cloud platforms.
Sriram Raghavan, Vice President of AI at IBM Research, expressed enthusiasm about the acceleration that NVIDIA GB200 NVL72 brings to training the Granite models. He remarked, "This collaboration with CoreWeave will enhance IBM’s capabilities to build advanced, high-performance, and cost-efficient models for powering enterprise and agentic AI applications with IBM watsonx."
Scaling Compute Resources for AI Development
Mistral AI, a Paris-based leader in open-source AI, is also benefiting from CoreWeave’s infrastructure, which is now equipped with GB200 NVL72 systems. These systems are accelerating the development of Mistral AI’s language models, such as Mistral Large, which are known for their strong reasoning capabilities. To effectively train and deploy these models, Mistral AI requires a cloud provider offering large, high-performance GPU clusters with NVIDIA Quantum InfiniBand networking and a reliable infrastructure management system—requirements that CoreWeave meets with industry-leading reliability and resiliency through tools like CoreWeave Mission Control.
Thimothee Lacroix, Co-founder and Chief Technology Officer at Mistral AI, noted a 2x improvement in performance for dense model training right out of the box with NVIDIA GB200 NVL72, without any further optimizations. He highlighted the new possibilities that this technology opens up for model development and inference.
Expanding NVIDIA Blackwell Instances
Beyond these long-term customer solutions, CoreWeave is offering instances with rack-scale NVIDIA NVLink across 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs, scaling up to 110,000 GPUs with NVIDIA Quantum-2 InfiniBand networking. These instances, powered by the NVIDIA GB200 NVL72 rack-scale accelerated computing platform, provide the scale and performance necessary to build and deploy the next generation of AI reasoning models and agents.
In summary, the introduction of NVIDIA GB200 NVL72 systems by CoreWeave marks a pivotal advancement in cloud computing, enabling leading AI companies to push the boundaries of AI model training and deployment. This development not only enhances performance and efficiency but also opens up new opportunities for innovation in AI applications. As more companies adopt this technology, the impact on AI development and deployment is expected to be profound, leading to faster, more efficient, and cost-effective AI solutions across various industries.
For more Information, Refer to this article.
