IBM z16 processes 300 billion AI operations a day

Anyone who thinks mainframes don’t evolve anymore is wrong. This (15), IBM made the official announcement of the IBM z16 , its new generation of equipment of the type. With the help of an artificial intelligence accelerator , the novelty is able to analyze millions of transactions in real time to prevent credit card fraud and deal with various other tasks that require immediate response.

IBM has been designing mainframes since the 1950s, so it is one of the pioneers in this area. The IBM z16 arrives as the successor to the IBM z15, the mainframe that the company announced in October 2019.

One of the attractions of the new version is IBM Telum . This is the name of an eight-core, 22 billion-transistor processor presented by the company in 2021 that, in addition to these features, has an artificial intelligence accelerator that adds more than 6 teraflops of processing power to the chip.

What is the IBM z16 capable of?

These and other characteristics make the IBM z16 specially prepared to prevent fraud in the financial system but, yes, it can be used in other applications. According to the company, this is possible because the mainframe “can process 300 billion inference operations per day with just a millisecond of latency ”.

What does that mean? That the IBM z16 performs monstrously! But let’s go into a little more detailed explanation.

An inference can be understood as the operation that is performed by a system based on neural networks a concept closely linked to artificial intelligence properly trained. The result of each operation consists of a forecast that can be used for immediate decision making, for example.

And the latency? Think of it as the response time it takes the computer to process the task and deliver the result. The lower this number, the better.

In its tests, IBM claims that it tested the IBM z16 on a version with 48 IFLs (from Integrated Facility for Linux , a type of processor dedicated to workloads on Linux) and 128GB of RAM . Based on Ubuntu 20.04, the mainframe ran a credit card fraud detection system that could process 300 billion inference operations in a single day, each with a latency of just a millisecond.

Leave a Reply