IDUG Content Committee

 View Only

Cool factor of z16

By Emil Kotrc posted Jun 07, 2022 07:03 AM


Cool Factor of IBM z16

Recently, I have been visiting a technical university with a short speech at a seminar and my topic was IBM mainframes. Fortunately, this was after IBM's announcement of the new machine IBM z16, so I had some great content to cover. One of the discussion point of course was the new processor - IBM Telum, and because the audience was geeky enough, a question was how does this chip compare with other processors, especially with Apple M1 Ultra released a bit sooner. Apparently, this is not a surprising question, I have heard this one even on IBM's session by Christian Jacobi (you can also listen to this podcast or terminal talk), chief architect of Telum, and I have also seen this great blog, which goes into more details.

So, how do you really compare IBM Telum and Apple M1 Ultra? Both are great high end processors, but for different purposes. Telum is a business oriented server processor, while M1 is a high end consumer chip. You would not find Telum in a stock or even in a high end PC and you would not find M1 in a server (although there are already some ARM based chips used in servers, see for example Amazon Graviton).

Key Highlights

Key highligts of IBM Telum (and z16)/differences from my perspective:

  • Telum implements the zArchitecture, which has its roots in IBM S/360. On the other hand M1 is ARM based processor.
  • Different cache design. While M1 uses more traditional cache hierarchy (private L1, shared L2), Telum processor uses physical and virtual caches - Level 1 cache is private for a core, Level 2 cache is dedicated to a processor core, but part of it can be shared with other cores as a virtual Level 3 and Level 4 caches. This means that there are no dedicated physical Level 3 and Level 4 caches. A lot more on that topic here or here. This sounds really as an interesting design, I have no idea, what could be the cache coherency impact, but I assume IBM has a great experience with similar implementations. To me it seems a bit similar as in some higher abstraction layers, such as Coupling Facility and Db2 group buffer pool p-locks. I am really looking forward how this will perform in real life scenarios and whether this can influence the future of cache design.
  • There is no GPU (graphical processing unit) in Telum compared to M1, not needed for server class machines
  • On chip accelerators - both have "AI" accelerators, which help with tensor calculations, although it does not sound that cool as Artificial Intelligence accelerator. Especially, for Telum this accelerator is important for data gravity as it provides low latency real time AI processing. If you are interested into details, refer to the latest principles of operations (which is a difficult read), or perhaps take a look at the IBM Z Deep Neural Network library. Most likely, you will not use the deep learning using the hardware instructions.

    Telum has even more accelerators in place:

    • sort - this type of acceleration was introduced with z15 and especially for Db2 can be very beneficial
    • crypto - again present already, now IBM is pushing even further with quantum-safe cryptography.
    • compression - before z15, zEDC was performed using an optional card, while in z15 and z16 this is done by an onboard chip for no additional cost.
  • Memory. Maximum supported memory by Telum is 10 TB, by M1 is 128 GB. Both are huge for its purpose. M1 looks much smaller here, but keep in mind that this chip is to be used in a personal workstation not in a server serving thousands of users. Amazing numbers (my personal PC I am using right now has "just" 24 GB)
  • IBM z16 uses enhanced RAIM (Redundant Array of Independent Memory originally introduced with z196) and comes with full transparent memory encryption.
  • Higher Telum clock speed (5.2 GHz) compared to M1 (3.2 GHz). From the past we have learned that the clock speed is no longer the major factor of the CPU performance. There are new instructions, new CPU optimizations, and more. However, what if you have a traditional COBOL application that was not compiled for years and not binary optimized? In such a case, the faster CPU can still bring the performance benefits even using older instructions.

There are many other things I did not cover, but please refer to the links section for more info.

z16 and Db2

Now, how would Db2 benefit from z16? In the Db2 13 announcement, IBM highlights the following topics. You can also learn more in the already published Db2 13 Redbook!

  • New SQL Data Insights feature - This feature provides AI capability (I would rather say machine learning or deep learning) directly to the platform without the need of ETL to another platform for data insights. There are four semantic type queries available at the moment

    • Similarity query
    • Dissimilarity query
    • Inductive reasoning semantic clustering query
    • Inductive reasoning analogy query

    These types of queries can help you to find groups of similar entities, detect outliers, determine relationships between pair of entities and so on.

    SQL Data Insights uses IBM Zacceleration for both - training the model and actual computation. The IBM z Deep Neural Network library (zDNN), see above, is used here and can benefit from the latest acceleration. However, this feature is available even for older mainframe models.

  • New System Recovery Boost (SRB) in IBM z16 can help improve Db2 for z/OS restart performance, thereby minimizing downtime - recovery boost was introduced with z15, but with z16 it provides more features including middleware restart boost - exactly what Db2 can take benefit of.
  • Better insights to tune Db2 performance in data sharing environments with the exploitation of IBM z16 - z16 introduces the ability to track data and directory residency statistics for Group Buffer Pools, which can help Db2 to adjust the workload to improve the performance. Group Buffer Pool residency time reporting requires Coupling Facility Level (CFLEVEL) 25 with IBM z16.
  • Starting with z15, Db2 can utilize the expanded SORTL instruction to improve the sort performance.


So what is really a cool factor of z16 at the end? It is brand new, there is some excitement, and there are some high expectations. There is also a huge potential for Db2 to exploit z16 features. Time will show. From my little world's perspective, this is a cool stuff I showed to the students at the university I mentioned at the beginning of my blog. Hopefully, some of the students enjoyed it and hopefully some of you will soon enjoy Db2 and the synergy with IBM z16.

In the mean time you can enjoy this video where a famous YouTuber gets really excited about the new z16 ;-)


Author: Emil Kotrc

Created: 2022-06-06 Mon 16:00