Covering Disruptive Technology Powering Business in The Digital Age

Written by: Martin Dale Bolima, Journalist, AOPG

IBM has unveiled its latest innovation: The next-generation IBM Z processor chip called Telum. The processor chip features industry-first, real-time Artificial Intelligence (AI) capabilities and is the culmination of years of research by IBM.

“It’s (Telum) really an evolution and continuation of our approach to hardware and software integration, and that’s the model of how we are able to deliver innovation that really matters to our clients,” said Barry Baker, Vice President at IBM Z Product Management in a recently held media briefing. “So, in many ways, this is an evolution of our processor design but it is also the first processor that contains on-chip acceleration for AI inferencing.”

Inside Telum are 8 processor cores, each with 32MB of cache, that were first presented by IBM Research at the 2021 ICCC Conference earlier this year. Aside from those 8 cores, Telum also contains 22 billion transistors and 19 miles of wire on 17 metal layers, and these specifications are set to make this next-generation chip the first processor in the industry capable of bringing real-time, deep learning inference to transactional workloads that are response-time sensitive.

Telum is the first chip ever created by the pioneering IBM Research AI Hardware Centre, which is a collaborative ecosystem of partners that include leading industry, academic and government partners based in Albany, New York.  It has, to date, 16 members specifically working on AI hardware, including the recently launched Telum.

“Telum is the first IBM product with technology developed under the umbrella of the AI Hardware Centre,” said Kailash Gopalakrishnan, IBM Fellow and Senior Manager, Accelerator Architectures and Machine Learning at IBM Research. “The AI core that is integrated within Telum is the third-generation AI core that IBM has designed from the ground up. It’s optimised for latency and performance, (which is) obviously critical for low-latency transactions.”

Telum also boasts a unique design that will enable thread AI workloads to be accelerated by as much as 10x as compared to leading server processors available in the market. This increased and transformative AI capability, in turn, will allow end-users to switch from fraud detection to fraud prevention in which 100% of fraud will be detected and caught even before the transaction is completed.

Gopalakrishnan added that Telum can potentially open more expansive AI utilisation across industries, noting how the future of AI is integration with general-purpose cores, stating, “The critical innovation here is to really integrate the AI cores tightly with the CPU cores in a single chip so that we can have a very low-latency AI inference that’s critical to business use cases.”

“What we’re seeing in this particular chip and moving forward is how do we bring in all the features, like security, reliability, downtime and so on and so forth and integrate that together with AI capability,” Gopalakrishnan added. “The future is really about integrating AI capabilities right next to specific applications, specific general-purpose cores. This is a huge step in that direction that will make a dramatic impact on the kinds of use cases that we see really often.”

“A lot of the work we have done since 2017 focused on making sure that we develop AI cores incorporating the different low-precision formats but, at the same time, we want to make sure that the sustained performance of AI workloads and specifically our inference running on these cores is high,” said Gopalakrishnan in explaining some of the thought process and challenges attendant to researching, designing and making the Telum processor.

Anthony Saporito, Senior Technical Staff Member at IBM Z Hardware Development later on noted that Telum will be built in on IBM Z systems, potentially enabling end-users to save on costs in the future.

“The AI capabilities that we built into Telum are embedded in every Telum processor chip. We’re announcing the Telum chip right now but they will be built into IBM Z systems in the near future,” said Saporito.  “If somebody purchases one of those, they get this embedded AI capability as part of that system. So, in general, this absolutely supports what we call TCO (Total Cost of Ownership) in terms of having that capability built into the system with the chip.”

Telum is named after the javelin of the Greek goddess Artemis, which incidentally is the codename for the next IBM Z system in development. The name, according to Baker, is “meant to evoke a thought around forward movement, speed and precision,”—exactly the things Telum promises to deliver.

(0)(0)