New to IBM Z

New to IBM Z

Deepen your technical skills, expand your global network, and connect with mentors and other early tenure professionals on the mainframe platform.

 View Only

🐣 Easter Eggs in AI: Hidden Features of IBM Z's AI Tools - Part 2

By Jade Walker posted Mon May 20, 2024 10:00 AM

  

Unveiling the Power of Hardware and Software in AI on IBM Z

Welcome back to our series exploring the hidden, yet potent, features of IBM Z's AI tools. In this second installment, we continue our exciting journey, first embarked upon with Part 1, where we unearthed the software capabilities within IBM Z's AI arsenal. Robert Catterall and I extend our previous talk, "Infusing AI into z/OS-based applications - the why and the how" from IBM Z Day 2023, diving deeper into the advanced hardware and software that bolster AI applications on IBM Z. This post is designed to enlighten those new to IBM Z, with a focus on hardware enhancements and software enablers.

Watch the Replay - (Link at End of Article) ⬇

Hardware Acceleration: Boosting AI Performance

SIMD – Single Instruction Multiple Data: A feature that has evolved through the IBM z13, z14, z15, and z16. SIMD allows a single instruction to process multiple data elements simultaneously – essential for vector processing prevalent in machine learning tasks like model inferencing. This capability provides a substantial performance lift.

z16 Breakthrough – the AIU (AI Accelerator Unit): Integrated directly into the chip, the AIU complements the Z-core general processor, ensuring every processor chip in a z16 system – which can house up to 32 chips – includes AI acceleration. Each AIU boasts over 6 TFLOPS (trillion floating-point operations per second), enabling full-scale model inferencing at immense transaction volumes, maintaining response times around 1 millisecond.

Software Enabling Technologies for z/OS Applications

IBM Deep Learning Compiler: This tool is a bridge between development and deployment, allowing you to "Develop model anywhere, deploy on IBM Z." Build and train your model using any major framework, convert it to ONNX format, and then optimize it for execution on a z16 through the Deep Learning Compiler. The result is an inference program ready for z/OS or Linux deployment, infusing operational applications with AI power.

z/OS Container Extensions (zCX): zCX is a transformative feature of z/OS, allowing the deployment of Linux applications as Docker containers directly within a z/OS environment. This includes machine learning models developed with popular frameworks like TensorFlow and scikit-learn, available through the IBM Z and LinuxONE Container Registry.

Watson Machine Learning for z/OS (WMLz): WMLz allows for the native deployment of ONNX models on z/OS, providing alternative invocation methods tailored for CICS and IMS COBOL applications. It harnesses the IBM Deep Learning Compiler, micro-batching for enhanced parallel processing, and provides comprehensive model lifecycle management.


Db2 for z/OS SQL Data Insights: Introduced in Db2 13 for z/OS, this feature utilizes neural networks to generate "meaning vectors" from table data, empowering users and applications to engage with built-in Db2 functions to discern data patterns.

IBM watsonx Technology: The watsonx suite encompasses:

  • watsonx.data: A data lakehouse architecture optimized for AI and machine learning, prioritizing data usability.
  • watsonx.ai: An advanced studio for data scientists and developers to create, train, and deploy various AI models, both generative and predictive.
  • watsonx.governance: A governance framework ensuring explainability and transparency, which is crucial for deploying trustworthy AI within enterprises.

Conclusion: Harnessing AI in Real-World Applications

AI is far from an enigmatic concept; it's a technology that's ready for application. Whether you're integrating intelligence into operational z/OS-based applications for smarter decision-making or seeking to unearth new insights from existing data – a process we can call "data re-discovery" – the technology is ripe for the taking.

With hardware delivering exceptional performance and throughput, and software accelerating model development and deployment, the integration with operational systems is more seamless than ever. It's time to move past merely contemplating AI; it's time to harness it. Stay tuned for more insights in the following parts of our series, where we'll continue to unlock the full potential of AI on IBM Z.

Watch the Replay - Infusing AI into z/OS-Based Applications: The Why and The How w/ Robert Catterall & Jade Melody Walker

0 comments
11 views

Permalink