IBM to launch Telum chips aimed at AI inferencing workloads such as fraud detection

IBM has unveiled plans to launch its first chip with AI inferencing acceleration, with the ability to conduct tasks such as fraud detection, from early 2022.

The chip, named Telum, is the first one to be produced by the IBM Research Hardware Centre. The company expects it to be used by financial firms to prevent fraud, rather than simply detect it.

Planned for IBM Z and LinuxONE systems, specific features of Telum are increased performance, improved security, a new cache design, and an integrated accelerator designed for real-time embedded artificial intelligence.

The AI accelerator also makes the new chip extensible with future firmware and hardware updates available on the same hardware platform over time.

Telum has been in development for three years and was built on 7nm extreme ultraviolet technology created by Samsung.

“The completely redesigned cache and chip-interconnection infrastructure provides 32MB cache per core and can scale to 32 Telum chips. The dual-chip module design contains 22 billion transistors and 19 miles of wire on 17 metal layers,” said IBM.

“The chip contains 8 processor cores with a deep super-scaler out-of-order instruction pipeline, running with more than 5GHz clock frequency, optimised for the demands of heterogenous enterprise-class workloads.”