NVIDIA's Latest AI Supercomputer: GH200
NVIDIA is introducing its next-gen GPU, the GH200, following the success of the H100. The company is providing GH200 supercomputers to selected companies for testing and deploying AI models.
Key Points: - NVIDIA is rolling out the GH200 GPU to companies interested in testing and deploying models. - Abacus AI, led by CEO Bindu Reddy, is among the first companies to receive GH200 supercomputers. - Abacus AI plans to focus on open-source AI projects using the GH200, starting in January.
Release and Availability: - The GH200, succeeding the H100, is expected to be available by the end of the year, with systems running on it anticipated to start in the second quarter of 2024.
Cloud Service Providers: - AWS announced it would be the first customer to use NVIDIA's GH200 on its cloud platform. Google Cloud, Meta, and Microsoft are also expected to be early adopters, exploring the GH200's potential for handling generative AI workloads. - Oracle Cloud Infrastructure (OCI) has plans to integrate GH200s into its cloud offerings.
Collaboration and Customization: - NVIDIA intends to share the DGX GH200 design with cloud service providers and hyperscalers, allowing them to customize it to better suit their infrastructure.
Industry Trends: - Microsoft and Meta have plans to integrate AMD's Instinct MI300X accelerators for AI workloads, with Meta building a new data center using MI300X. - Intel is set to announce its Gaudi3 AI accelerator on December 14th, positioning it in comparison to NVIDIA's H100.
In summary, NVIDIA's GH200 marks the company's continued commitment to advancing AI capabilities, with prominent companies already gearing up to harness its power for various applications.