It’s been my pleasure to collaborate with Ishwar Fernandes and the talented CSI and IBM teams on this exciting project that demonstrates the practical applications of AI on IBM Power systems. In their recent technical demonstration, CSI showcases how they’re leveraging IBM Power10 technology to run sophisticated AI workloads efficiently and securely on-premises.
Great job, Ishwar!
Infrastructure Setup
The demonstration runs on a robust infrastructure consisting of:
– IBM Power, with Power10, MMA inferences technology
– NO GPU’s
– OpenShift Container Platform
– Three master nodes
– Five worker nodes – with AI running on just two of the Power Worker Nodes
Highlighted Use Cases
1. Custom Question & Answer System
– Processes uploaded PDF documents
– Demonstrates real-time analysis of technical documentation
– Showcased using IBM Power S1012 documentation
– Provides accurate, context-aware responses
2. PII Data Processing
– Identifies and manages sensitive information
– Handles numerical data extraction and masking
– Essential for compliance and data security
3. Marketing Content Generation
– Creates comprehensive marketing briefs
– Generates structured content with clear objectives
– Streamlines content creation workflows
4. Email Summarization
– Quickly distills key information from email communications
– Improves productivity through automated comprehension
– Provides concise, actionable summaries
5. Sales Insight Analysis
– Analyzes sales conversations
– Extracts main topics and key points
– Helps identify customer needs and priorities
Technical Highlights
– Flexible LLM Integration: Currently utilizing Google FLAN with capability to integrate other models from HuggingFace or NVIDIA
– On-Premises Processing: All AI inferencing runs locally on Power10 hardware
– Multi-Architecture Support: Leverages OpenShift’s Multi-Architecture Compute clustering, which provides support for both x86 and IBM Power Worker Nodes within the same OpenShift cluster (v4.14 and later).
– Seamless Data Movement: Enables efficient data transfer between x86 and Power architectures
Looking Forward
This demonstration showcases the practical reality of enterprise AI deployment on IBM Power systems. CSI’s implementation proves that organizations can run sophisticated AI workloads efficiently while maintaining data sovereignty and security through on-premises processing.
The combination of IBM Power10’s processing capabilities with OpenShift’s container orchestration creates a robust foundation for Enterprise AI applications. This solution opens new possibilities for organizations looking to implement AI while maintaining control over their data and processing infrastructure.
Next Steps
- Customers, get in touch with CSI and or IBM to identify and or progress AI on Power projects.
- Business Partners, get in touch with IBM to assist with your own AI on Power development.
- ISV’s get in touch for assistance integrating AI on Power with your own solutions.
Credit
(CSI)
(IBM)
Ishwar Fernandes, Head of Technical Architects, CSI
Dr Sebastian Lehrig, WW AI on IBM Power Leader
Andrew Laidlaw, Supporting IBM Power Business Partners, IBM UK
Sylvain Delabarre, IBM Client Engineering EMEA
Many thanks to:
Ishwar Fernandes for taking the time and effort to learn and share the art of the possible with AI on IBM Power.
Sebastian Lehrig for sharing lessons learned from Global AI on Power projects.
Andrew Laidlaw for supporting CSI through this project.
IBM Client Engineering from Montpellier for their support in hosting and supporting the project, while CSI provision local resources for longer term use.
Sylvain Delabarre from IBM Client Engineering, who is always very helpful running and supporting customer proof of concepts, proof of technology and delivering minimal viable products.
Leave a Reply