AI reasoning models and agents are ready to transform the industries, but the delivery of their full -scale potential requires massive calculation and optimized software. The “reasoning” process involves several models, generating many additional tokens and requires an infrastructure with a combination of communication, memory and high -speed calculation to guarantee real -time and high quality results.
To meet this request, Coreweave launched NVIDIA GB200 NVL72– Based bodies, becoming the first cloud service provider to make NVIDIA Blackwell platform generally available.
On a rack scale NVIDIA NVLink through 72 NVIDIA Blackwell GPU and 36 CPU NVIDIA Gracescaling up to 110,000 GPU with Nvidia Quantum-2 Infiniband Networking, these instances provide the scale and the performance necessary to create and deploy the next generation of reasoning models and AI agents.
NVIDIA GB200 NVL72 on Coreweave
NVIDIA GB200 NVL72 is a solution to the rack scale cooled by liquid with a NVLink area of 72 GPU, which allows the six dozen GPUs to act as a single massive GPU.
Nvidia Blackwell offers many technological breakthroughs that accelerate the generation of inference tokens, increasing performance while reducing service costs. For example, fifth generation NVLink allows 130 TB / S GPU bandwidth in a 72 GPU NVLink field, and the second generation transformer engine allows FP4 for faster AI performance while maintaining high precision.
The Corewave Managed Cloud Service Portfolio is specially designed for Blackwell. Coreweave Kubernetes service Optimizes the workload orchestration by exposing NVLink domain IDs, guaranteeing effective planning in the same rack. Slurm on Kubernetes (SUNK) supports the plug-in of topology blocks, allowing an intelligent workload distribution on the GB200 NVL72 racks. In addition, the Coreweave observability platform provides real-time information on NVLink performance, the use of GPUs and temperatures.
Corewave GB200 NVL72 instances have Infiniband NVIDIA Quantum-2 networking which offers a 400 GB / S bandwidth per GPU for clusters up to 110,000 GPU. NVIDIA Bluefield-3 DPUS Also supplies the accelerated multi-local cloud network, access to high performance data and GPU calculation elasticity for these instances.
Complete accelerated IT platform for corporate AI
NVIDIA’s AI AI Platform combines cutting-edge software with Blackwell infrastructure to help companies create fast, precise and scalable AI agents.
Nvidia blueprints Provides predefined, customizable and ready to deploy predefined reference workflows to help developers create real applications. NVIDIA NIM is a set of easy -to -use microservices designed for secure and reliable deployment of high performance AI models for inference. Nvidia Nemo Includes tools for training, personalization and continuous improvement of AI models for use cases of modern companies. Companies can use Nvidia Blueprints, Nim and Nemo to build and refine models for their specialized AI agents.
These software components, all part of the NVIDIA AI ENTERPRISE The software platform, are key catalysts to deliver Agentic on a large scale and can be easily deployed on Coreweave.
Bring a new generation AI to the cloud
The general availability of NVIDIA GB200 NVL72 bodies on Coreweave highlights the last people in business collaboration, focused on the supply of the latest IT solutions accelerated to the Cloud. With the launch of these cases, companies now have access to the scale and the performance necessary to feed the next wave of models of reasoning and AI agents.
Customers can start providing instances based on GB200 NVL72 via the Coreweave Kubernetes service in the US-West-01 region using the GB200-4X instance ID. To start, Contact Coreweave.