IBM Telum – Using Silicon to Address Evolving Business Challenges

By Charles King, Pund-IT®  August 25, 2021

The pervasiveness of technologies that touch virtually every corner of consumer and commercial life tends to obscure how solutions evolve to address changing behaviors, circumstances and markets. While those shifts are often subtle, vendors can alter their own efforts to benefit customers and help them deal with serious existing and emerging threats.

It is important to note that these benefits are not limited to any specific form or class of technology, like silicon hardware, software and services. In fact, new and valuable innovations can arise anywhere, even areas that conventional wisdom considers fundamentally fixed and mature. Some might believe that the new IBM Telum Processor the company announced at Hot Chips this week belongs in this category. So, let’s consider Telum and what it promises to IBM’s global enterprise customers.

IBM Telum Processor – Beneath the hood

What is it that makes Telum so special? In short, it is the first IBM processor that contains on-chip acceleration for AI inferencing so that fraud detection solutions can be effectively employed at scale and in real time during the microseconds it takes for a transaction to take place. That should intrigue IBM customers in banking, finance, trading and insurance applications, as well as other areas where fraud impacts customer interactions. The company is planning to bring Telum-based IBM Z mainframe solutions to market during the first half of 2022.

On the technical side, Telum chips contain 8 newly designed processor cores with a deep super-scalar out-of-order instruction pipeline. The cores run at over 5GHz clock frequency and are optimized for the demands of heterogenous enterprise class workloads. The new chip’s redesigned cache and chip-interconnection infrastructure delivers 4X as much L2 cache per core as IBM’s z15 provides. Counting the virtual L3 and L4 IBM offers, Telum supports 1.5X as much total cache per core as the z15. Telum-based solutions can scale up to 32 chips.

Telum also supports enhanced security features, including improvements to Secure Execution environments that make the new chip a great choice for supporting Hyper Protect Virtual Servers and trusted execution environments for cloud infrastructures. In fact, Telum’s new and improved capabilities can proactively protect systems against advanced attacks that can threaten traditional systems.

Telum is the first IBM chip featuring technologies developed at the company’s Research AI Hardware Center. The built-in AI accelerator resulted from a collaboration between IBM Systems and IBM Research teams. Each Telum processor contain 22 billion transistors and 19 miles of wire on 17 metal layers, and processors are deployed in two chip modules. Telum production is being facilitated by IBM partner Samsung, leveraging the company’s 7nm EUV technology node.

The trouble with transactions

Why did IBM invest three years and tens of millions of dollars developing Telum? Three points come to mind. First and foremost is the company’s longstanding leadership in supporting business transaction processing by means of its homegrown IBM Z mainframe solutions. The robust performance, scalability, reliability, and security that those systems offer is why so many global finance, trading, insurance companies and retailers continue to depend on IBM to support vital transactional applications and workloads.

Second, computerized business transactions are the richest target for cybercriminals and fraudsters. Legendary 20th century robber Willie Sutton reputedly said he and his gang targeted banks, “Because that’s where the money is.” The problem today is that online transactions can be attacked at virtually any time by people operating from virtually anywhere. The Federal Trade Commission’s 2020 Consumer Sentinel Network Databook noted that consumers reported losing more than $3.3 billion to fraud in 2020, nearly double the $1.8 billion they lost in 2019. That is a payday that would have warmed Willie Sutton’s heart and fattened his wallet.

Finally, conventional approaches to fraud detection tend to be reactive, catching incidents (though seldom fraudsters) after an attack occurs. Those solutions can also be compute-intensive, time consuming and impractical. For example, in retail situations analysis and fraud detection is often performed far away from mission-critical transaction systems and business data. As a result, warnings often cannot be provided in time to halt dodgy deals and prevent bad actors from getting away with a company’s or customer’s goods.

Telum’s transactional innovations

In other words, compute-enabled business transactions are an area ripe for new innovative approaches.

How might Telum impact situations such as these? In essence, the new processor’s on-chip inferencing capabilities will enable IBM Z solutions to simultaneously run AI workloads, including fraud analysis and transactional workloads in real time and at enterprise scale.

In contrast, many alternative solutions perform AI inferencing off-platform resulting in significant latency delays which can impede the effectiveness of fraud detection and resolution. Additionally, while some vendors offer on-chip inferencing, they distribute AI support across all the chips in a system which is a less effective methodology for mixed transaction processing workloads than IBM Telum’s centralized approach.

What this means practically is that the speed and scale of Telum-based inference will enable IBM customers to detect frauds while they are in process rather than chasing fraudsters after the fact. Plus, customers will be able to achieve those benefits without impacting essential transaction performance in any way.

Clearly, that should be massively beneficial for the global banks, trading companies and other institutions that have long depended on IBM Z for their most mission-critical workloads. However, the scale and speed of IBM Telum solutions make it practical to apply AI-enabled processes and functions to all transactional workloads

Final analysis

IBM Telum Processors can be examined and analyzed though a variety of lenses—microprocessor advancements, AI innovations, the collaborative value of the company’s new Research AI Hardware Center and its strategic partnership with Samsung, and many others. However, the primary point and takeaway from this announcement should be the way in which Telum reflects IBM’s continuing focus on wresting the greatest value from new compute technologies and using them to benefit its customers and partners.

It is also important to note that Telum is anything but a one-off. Instead, it is merely the latest example of the innovations that enterprises have enjoyed and come to expect from IBM in the form of solutions that enable them to successfully adapt to ongoing marketplace changes and address new and growing challenges. There should be many more to come.

© 2021 Pund-IT®. All rights reserved.