IBM recently showcased its new Telum processor, which will be incorporated in the next-gen IBM Z systems. Along with 8 cores and a huge amount of L2 cache, the processor packs a dedicated AI accelerator that is capable of detecting fraud in real-time.
The Federal Trade Commission (FTC) received 4.7 million reports of fraud last year, with $3.3 billion in total losses. Telum, as per IBM, addresses this issue in real-time. IBM used credit card transactions as an example, saying that the processor can detect a fraudulent transaction before it even completes.
Telum good for other tasks as well
IBM says this could pave the way for “a potentially new era of prevention of fraud at scale.” Besides, Telum’s AI accelerator is also capable of handling other workloads. With the aid of machine learning, the processor can conduct risk analysis, detect money laundering, and handle loan processing, along with other things.
The processor has the appearance of modern chips from AMD or Intel. It packs 8 cores with simultaneous multithreading running at 5GHz. A ring that IBM refers to as a “data highway” connects the private cache pools, providing the processor a total of 256MB cache. The AI accelerator passes information to and from the cores and dies it with minimal latency.
IBM says applications can use up to 32 Telum chips, detecting from frauds to large-scale risk analysis for banks. Samsung is working with IBM to build the processor, using its 7nm extreme ultraviolet process technology.
AI becoming stronger and efficient
Accelerated computing is on the rise, and the new processor is another example of the same. Since architectural improvements aren’t fast enough, leading companies are shifting towards dedicated “accelerators” that are specialized for certain types of tasks.
Similarly, a team of researchers used machine learning to predict and explain terrorist attacks. Their tests show the models can accurately predict attacks in areas that are already suffering from terrorism. However, they learned that “black swan events” that happen sporadically aren’t possible to predict.
