IBM announces z16 mainframe with Telum processors

IBM announces the z16, a mainframe for enterprise applications. According to the company, the system is suitable for handling large volumes of transactions, among other things, where artificial intelligence can detect fraud.

IBM’s z16 series includes IBM’s Telum processors announced last year. They have a built-in accelerator for AI calculations. Telum is a 7nm produced dual-chip module with a total of sixteen cores. IBM offers several configurations of the z16, with up to 200 configurable cores.

For example, the IBM z16 A01 is available in five variants, with 39 to 200 cores: Max39, Max82, Max125, Max168 and Max200. This system offers up to 40TB of redundant array of independent memory. This is a similar technology to raid, but for ram, in which extra modules and algorithms protect the memory system against failure.

As an example, IBM highlights that the z16 can be used for the deployment of deep learning models to detect fraud in real time when processing large numbers of transactions. With previous systems, this would not yet be possible due to latency, which means that only ten percent of the transaction traffic would be monitored with fraud detection models.

The IBM z16 could handle 300 billion inference requests per day at a millisecond of latency. IBM extrapolated this result after internal testing with a z16 with 128GB and Ubuntu 20.04, when running a synthetic test to detect credit card fraud. Inference is the step for trained neural networks to make predictions.