Oracle has announced what it calls the largest AI supercomputer in the cloud, boasting 16 zettaflops of peak performance powered by 800,000 Nvidia GPUs. The system uses RoCE networking to link GPUs through isolated planes, aiming to reduce latency. This move comes as rivals develop their own advanced GPU-driven cloud infrastructures.
Oracle's new AI supercomputer represents a significant push in cloud computing capabilities. According to the announcement, the system achieves 16 zettaflops of peak performance, making it the largest of its kind in the cloud. It incorporates 800,000 Nvidia GPUs, connected via RoCE networking that links them through isolated planes to minimize latency.
The title of the report highlights Oracle's bet on this supercomputer power, while noting that competitors are quietly preparing next-generation GPU-driven cloud monsters. This development underscores the intensifying competition in AI infrastructure, where high-performance computing is crucial for advanced applications.
No specific timeline for deployment or further details on rival systems were provided in the source, but the focus on RoCE technology suggests optimizations for efficient data transfer in large-scale AI workloads.