From Transactions to Transformers: Hardware Built for...
summary
Claims enterprise mainframes run 33 billion encrypted transactions per day with eight nines availability and handle 70% of the world’s financial transactions. Telum 2 and Power11 CPUs include on-processor AI acceleration delivering in-transaction fraud detection at millisecond latency and hundreds of thousands TPS. Storage systems are automatically vectorized for RAG pipelines while preserving replication and access controls. A Spyre Accelerator PCIe card ships next week and is integrated with PyTorch; a live mainframe demo is running at the booth.
AI-written summary · about
related talks
17mZero to 50 ExaFLOPS in Under a Year: Lessons from the Trenches
04m4 real-world AI deployments at enterprise scale
23mThe Race to the Bottom: Low Latency in the age of the Transformer
17mScott Baker, IBM | Nvidia GTC 2024
20mKeynote: Future-Proofing Confidential AI
52mSimplifying the adoption of generative AI for enterprises
24mStrictly Confidential: Use Cases & Demos of AI Workflows in TEEs
10mApproach to AI at a large scale