Samsung Electronics was chosen as a contract foundry chip maker to fabricate IBM’s newly developed on-chip AI accelerator processor.
Dubbed as Telum processor, the chip is the sort of on-chip AI hardware accelerator processor that hardwired AI algorithm on a chip for real-time AI inferencing.
Expected to be widely used in a wide range of applications across banking, finance, trading, insurance and customer interactions, the chip is designed to address a bunch of enterprise-level workloads with near-zero data latency. For example, it allows clients to leverage full power of AI processor to perform AI-specific workloads, making it an ideal choice to work out real-time financial services workload like fraud detection, loan processing, clearing and settlement of trades, anti-money laundering and risk analysis.
The chip comes built with 8 processor cores that implements a deep super-scalar out-of-order instruction pipeline, operating at more than 5GHz clock frequency.
It also contains 32MB cache per core, implementing 22 billion transistors and 19 miles of wire on 17 metal layers.
It is built with IBM’s long heritage of innovative design and engineering that includes hardware and software co-creation and integration, coming built with a full stack of silicon system, firmware, operating system and software frameworks.
Samsung Electronics, IBM’s long-time foundry partner, will fabricate it with its EUV-based 7nm node process technology.
IBM said that the Telum chip-based system is scheduled to roll off sometime in the first half of 2022