Live from IBM Edge2015: IBM z Systems technical overview
IBM Edge 2015 kicked off this morning with an overview of the technology behind the new IBM z13 mainframe courtesy of Charles Webb, an IBM Fellow in the z Systems development group. Before starting, he gave the audience some insight into the naming behind the platform.
According to Charles, many people have asked why IBM would give the “unlucky” 13 designation to the most advanced mainframe technology ever released. He explained that the z13 marks the 13th generation of the CMOS mainframe, and given all of the predictions of the mainframes death over the past 30 years, they decided not to worry about superstition.
The IBM z13 8-core processor chip includes a staggering 3.99 billion transistors and 13.7 miles of copper wire. But despite that, Charles explained that as Moore’s Law slows down, simply trying to develop faster chips doesn’t provide the amount of increased performance that the IBM team wanted to achieve. Going faster than 5.5GHz also poses problems from power density standpoint; it’s simply too hard to provide enough power to the processor to support their needs.
While much of Charles’ presentation was focused on hardware, he remarked that when it comes to selling mainframes “…we don’t just deliver a bucket of silicon and wires and sheet metal. We have to deliver software applications to go along with it…our R&D groups haven’t given me wires that transmit data faster than the speed of light. They haven’t given me scalable atoms.”
According to Charles, the team’s challenge was to slow down clock rate and speed up execution time. They did this in a number of ways, including an incredible amount of work focused on data caching. Here are a few of the things they did:
By adding front-end bandwidth, the z13 can process twice as many instructions per cycle compared to previous generations.
The z13 includes improved branch prediction technology, leveraging the most sophisticated prediction engine in the industry. Charles quipped “we have predictors on our predictors.”
An intra-node interface enables direct L3 cache access between processors. This improved performance by eliminating the need for data to travel through multiple L4 caches.
Charles told the crowd that these days, IBM mainframe development spans the world; as he put it “the sun never sets on z Systems development.” This process allows for constant development and testing. For example, when the US-based team leaves for the day, teams in Asia can test their work and make recommendations so the original team can get to work immediately when they return the next morning. It helps the team get more done, and has sped up the overall development process.
We at Rocket were one of the first IBM partners to get a new z13, and our technical teams are still digging into all of the capabilities. If you haven’t already, you can check out all of the new capabilities on the IBM z13 website.
Fun fact: did you know that IBM invented concept of caching technology for the mainframe in the 1960s?