Oracle Launches Zettascale AI Superclusters with 131,072 NVIDIA Blackwell GPUs on OCI

The first zettascale cloud computing clusters powered by the NVIDIA Blackwell platform were revealed by Oracle. The largest AI supercomputer in the cloud, Oracle Cloud Infrastructure (OCI), is now accepting orders. It comes with up to 131,072 NVIDIA Blackwell GPUs.

Executive Vice President of Oracle Cloud Infrastructure Mahesh Thiagarajan stated, “We have one of the broadest offerings in AI infrastructure and are supporting customers that are running some of the most demanding AI workloads in the cloud. “

Also Read: Oracle NetSuite Unveils AI Advancements to Boost Efficiency, Growth, and Business Insights

The first Zettascale computing cluster in the world

More than three times as many GPUs are available at OCI Supercluster’s highest scale as compared to the Frontier supercomputer and more than six times more than other hyperscalers. A selection of HPC storage options, ultra-low latency RoCEv2 with ConnectX-7 and ConnectX-8 SuperNICs or NVIDIA Quantum-2 InfiniBand-based networks, and OCI Compute Bare Metal are all included in the OCI Supercluster.

OCI Superclusters can be ordered with NVIDIA Blackwell or H100 or H200 Tensor Core GPUs driving OCI Compute. Upgrading to 16,384 GPUs, OCI Superclusters equipped with H100 GPUs may achieve 65 ExaFLOPS of performance and 13Pb/s of aggregated network throughput. 

Later this year, OCI Superclusters with H200 GPUs will be available. These clusters will scale to 65,536 GPUs, offering up to 260 ExaFLOPS of performance and 52Pb/s of aggregated network throughput. Using NVLink and NVLink Switch, OCI Superclusters equipped with NVIDIA GB200 NVL72 liquid-cooled bare-metal instances will be able to connect with up to 72 Blackwell GPUs at a total bandwidth of 129.6 TB/s under a single NVLink domain.

Also Read: Oracle’s Initiative to Train 200,000 Indians in AI, Data Science, and Cloud

GPU-to-GPU communication within a single cluster will be smooth because of NVIDIA Blackwell GPUs, which will be available in the first half of 2025 and feature fifth-generation NVLink, NVLink Switch, and cluster networking.

According to Ian Buck, vice president of NVIDIA’s Hyperscale and High-Performance Computing division, “access to powerful computing clusters and AI software is critical as businesses, researchers, and nations race to innovate using AI.”

OCI’s robust security and sovereignty controls are enabling high-performing AI infrastructure, which is being utilized by clients including Zoom and WideLabs.

One of the biggest Portuguese LLMs is trained on OCI by WideLabs

Amazonia IA, one of Brazil’s biggest LLMs, is being trained on OCI by WideLabs, an applied AI business in Brazil. They created bAIgrapher, an application that helps Alzheimer’s patients preserve significant memories by using its LLM to create biographical information based on data collected from patients.

WideLabs runs its AI workloads on the Oracle Cloud São Paulo Region, guaranteeing that private information stays inside national boundaries. 

Also Read: MakeX World Robotics Championship 2024 Nationals to Be Held in Mumbai This October

Nelson Leoni, CEO of WideLabs, stated, “OCI AI infrastructure offers us the most efficiency for training and running our LLMs.” “OCI’s size and adaptability are invaluable as we push forward with innovation in the healthcare industry and other important areas.”

Zoom’s generative AI assistant makes use of OCI’s sovereignty capabilities.

OCI is being used by Zoom, a prominent AI-first collaboration platform, to offer inference for Zoom AI Companion, the business’s free AI personal assistant. Users can produce ideas during brainstorming sessions with colleagues, compose emails and chat messages, summarize meetings and chat threads, and more with the aid of Zoom AI Companion. Zoom will be able to maintain client data locally in the region and satisfy AI sovereignty needs in Saudi Arabia, where OCI’s solution is being initially implemented, thanks to OCI’s data and AI sovereignty capabilities.