AMD's Instinct MI355X accelerators for AI workloads are gaining popularity, with Oracle recently becoming a major customer. In its latest financial report, Oracle revealed that it had purchased 30,000 AMD Instinct MI355X accelerators. Larry Ellison stated, "In Q3, we entered into a multi-billion dollar agreement with AMD to deploy a cluster of 30,000 of their newest MI355X GPUs." He also mentioned that "All four of the top cloud security firms - CrowdStrike, Cyber Reason, Newfold Digital, and Palo Alto - have chosen to migrate to the Oracle Cloud. Additionally, Oracle has developed a new product called the AI data platform, which allows its extensive database customer base to utilize the latest AI models from OpenAI, XAI, and Meta to analyze the data stored in their millions of existing Oracle databases. By leveraging Oracle version 23 AI's vector capabilities, customers can automatically convert their data into a format understood by AI models, enabling these models to learn, comprehend, and analyze all aspects of their organization or government agency's data, thereby unlocking its value while maintaining data privacy and security."
The AMD Instinct MI355X accelerator introduces the CDNA4 architecture on TSMC's N3 process node, focusing on accelerating AI workloads. This chiplet-based GPU provides 2.3 petaflops of FP16 compute and 4.6 petaflops of FP8 compute, representing a 77% performance improvement over the MI300X series. A significant advancement of the MI355X is its support for reduced-precision FP4 and FP6 numerical formats, enabling up to 9.2 petaflops of FP4 compute. It features 288 GB of HBM3E memory across eight stacks, delivering a total bandwidth of 8 TB/s. The MI355X is expected to enter the market in the second half of 2025, aligning with AMD's annual data center GPU launch cadence. Oracle is likely to prepare its data center infrastructure for these GPUs ahead of their shipment by AMD.
