IBM has unveiled an upgraded computer chip that could boost its customers' AI capabilities and transform the way some industries handle data-intensive tasks.
For large enterprises that rely on IBM's powerful mainframe systems, Telum II processors have the potential to enable businesses to run smarter and faster in an increasingly AI-driven world: banks can spot fraud faster, insurers can process claims more accurately and retailers can deliver more personalized shopping experiences.
The AI chip market is booming, with demand outstripping supply as companies around the world desperately need more computing power. Revenues at Nvidia, AMD and Intel are soaring, and tech giants like Google, Amazon and Microsoft are developing custom AI processors. This chip rush is driven by the growing popularity of AI applications, from chatbots to self-driving cars. The resulting race is spurring innovation and driving down costs while pushing the boundaries of what's possible in machine learning and data processing.
Experts say the AI chip revolution could have a profound impact on commerce. Retailers with powerful AI hardware can analyze customer behavior in real time and adjust promotions and inventory to maximize sales. Financial institutions are deploying advanced fraud detection systems that can process millions of transactions per second, potentially avoiding billions of dollars in losses. Supply chain managers are using AI to predict disruptions and optimize routes, reducing costs and improving efficiency.
New chip details
At its Hot Chips 2024 conference in Palo Alto, California, IBM showed off its next-generation Telum II processor and Spyre Accelerator, both of which are designed to boost AI capabilities within IBM's Z mainframe systems and are due to launch in 2025.
“Developed using Samsung's 5nm technology, the new IBM Telum II processor features eight high-performance cores running at 5.5GHz,” IBM said in the announcement. This marks a big leap from the previous generation, with the company adding that the processor's “on-chip cache capacity increases by 40 percent, with virtual L3 and virtual L4 increasing to 360MB and 2.88GB, respectively.”
Perhaps most notable for AI applications, each Telum II accelerator is expected to deliver “a four-fold improvement, reaching 24 trillion operations per second (TOPS).” The increased processing power is designed to handle more complex AI models and larger datasets, critical for businesses keeping up with ever-growing volumes of information.
Meanwhile, the Spyre Accelerator is a new addition to IBM's offering, which the company says “contains 32 AI accelerator cores that share a similar architecture to the AI accelerator integrated in the Telum II chip.” The accelerator can be connected to an IBM Z system via PCIe and can be paired with the Telum II processor to deliver even more AI processing power.
Niche efforts in the world of AI
But QueryPal CEO Dev Nag cautioned against over-hyping the announcement: “These are not general-purpose AI chips that compete directly with NVIDIA's GPUs or similar products from AMD,” Nag told PYMNTS. “They're unlikely to be adopted by cloud vendors or used broadly outside of IBM's existing mainframe customer base.”
While great for mainframe users, this development won't change the broader AI landscape, as Nag points out that the mainframe market “peaked in revenues about 20 years ago” and represents a specific enterprise computing segment.
To those familiar with cloud computing trends, IBM's focus on mainframe systems may seem puzzling, but mainframes still play a vital role in certain areas: Many of the world's largest banks, insurers, and retailers continue to rely on these systems for their core operations, focusing on their reliability, security, and ability to handle huge volumes of transactions.
The chip's innovation is less about revolutionizing AI computing and more about modernizing existing systems. The company's proposed “ensemble approach to AI” blends traditional and new AI models, which may bring unique benefits to certain industries. As IBM explains, “Ensemble AI leverages the strengths of multiple AI models to improve overall performance and prediction accuracy compared to individual models.”
The new chip's approach could prove valuable in scenarios such as fraud detection, with IBM saying that “traditional neural networks are designed to provide an initial risk assessment, and when combined with large-scale language models (LLMs), they are tuned to improve performance and accuracy.”
Keeping the mainframe relevant
With its latest announcement, IBM is giving mainframe customers a way to incorporate advanced AI without overhauling their systems — a strategic move to keep these powerful but aging systems relevant in an increasingly cloud-driven world.
“Incorporating AI into enterprise transactions has become essential for many of our clients' workloads,” IBM said in a statement. The company emphasized that its “AI-driven fraud detection solutions are designed to save clients millions of dollars annually,” highlighting the real-world impact of these technological improvements.
But Nag emphasizes that this innovation is limited in scope: “While still important in certain industries like finance and insurance, the mainframe market is a niche segment of enterprise computing, albeit with widespread support across the Fortune 500,” he says, noting that “Nvidia GPUs boast higher revenues than IBM's entire System Z mainframe sales.”
