Python Ecosystem
Solution report card
| Runs on IBM i? | ❌ | |
| On-prem | ✅ | |
| IBM Cloud | ✅ | |
| AI capabilities | Machine Learning Deep Learning Generative AI Agentic AI more… | |
| Commercial support | ✅ | |
| Free to try? | ✅ | |
| Requirements | IBM Power |
What is the Python Ecosystem for IBM Power?
Section titled “What is the Python Ecosystem for IBM Power?”The Python Ecosystem for IBM Power (PyEco) is IBM’s supported distribution of Python and the scientific/AI Python stack — optimized and validated for IBM Power architecture. It provides the key ML and AI libraries that data scientists and developers need, packaged and tested to take advantage of IBM Power’s hardware capabilities including the MMA (Matrix Math Accelerator) on Power10.
PyEco runs on Linux on IBM Power (not directly on IBM i), making it the natural complement to IBM i in a hybrid on-premises AI architecture: IBM i holds the data in Db2 for i, while a Power Linux partition or server running PyEco handles the compute-intensive AI workloads.
What’s included
Section titled “What’s included”PyEco provides IBM-optimized builds of the core Python AI stack:
- PyTorch and TensorFlow — Deep learning frameworks with Power-optimized kernels
- scikit-learn — Classical ML algorithms
- XGBoost and LightGBM — Gradient boosting frameworks
- NumPy, SciPy, pandas — Scientific computing foundations
- ONNX Runtime — Cross-platform inference engine
- Hugging Face Transformers — Access to thousands of pre-trained models
All packages are built for the ppc64le (Power little-endian) architecture and tested for compatibility with each other.
Why use PyEco instead of upstream pip packages?
Section titled “Why use PyEco instead of upstream pip packages?”Standard pip packages from PyPI may not include Power-optimized builds, or may not be available for ppc64le at all. IBM’s PyEco provides:
- Validated compatibility — All packages are tested together
- Hardware optimization — Kernels tuned for Power’s vector/SIMD capabilities and MMA
- Commercial support — IBM support available for production deployments
- Regular updates — Packages kept current with upstream releases
Use with IBM i data
Section titled “Use with IBM i data”PyEco on a Power Linux partition can connect to Db2 for i on an adjacent IBM i partition via JDBC (JT400), ODBC, or the Mapepire protocol, making it straightforward to use IBM i data as the input to ML training and inference workloads.
See the Machine Learning and Deep Learning journey pages for examples.