Nvidia DGX Cloud, built on Oracle Cloud Infrastructure and Nvidia, is generally available with supercomputer clusters starting at $36,999 a month.
The GPU maker is among the early generative AI winners since Nvidia infrastructure is being used to train generative AI and large language models.
In March, Nvidia launched DGX Cloud, an AI supercomputing service, at its GTC conference. Nvidia GTX Cloud includes clusters of supercomputing as well as software. Oracle Cloud Infrastructure, a key partner of Nvidia, was out of the gate first with a supercluster that includes a purpose-built RDMA network, bare-metal compute and high performance local and block storage with scale to over 32,000 GPUs.
Constellation Research analyst Holger Mueller was upbeat about the launch since it will make training generative AI models more cost effective.
"Nvidia has taken a key step to secure its future relevance with making available Nvidia DGX Cloud. In the past it would deliver its GPUs and machines, practically being an arms dealer, and now it is providing the whole platform. And CxOs care for it, as Nvidia offers the portability of their models between on premises and the cloud.
Interesting – but not surprising – choice that Nvidia choose Oracle Cloud – which is uniquely well setup to run AI workloads – with being able to efficiently connect compute nodes with AI. Turns out that the ideal cloud infrastructure design point for database and SaaS apps – is also ideal for AI platforms."
Constellation Research's Andy Thurai added that the Nvidia DGX Cloud option can scale up generative AI training. "Given the market craze and appetite for training bigger and better LLM/generative AI models, single GPU or GPU clusters are not going to be enough to train such massive models like GPT-4," said Thurai. "These models require a supercomputer to train these models so it makes sense that NVIDIA is taking this route. It is not surprising as HPE recently made similar announcements to use their Cray supercomputers for AI model training services as well."
Microsoft Azure and Google Cloud are expected to host DGX Cloud in the future.
Each instance of DGX Cloud features eight NVIDIA 80GB Tensor Core GPUs for 640GB of GPU memory per node as well as Nvidia Base Command Platform software and Nvidia AI Enterprise for pretrained models and AI frameworks.
Nvidia DGX Cloud use cases span multiple industries including healthcare, financial services, insurance and software.
