IBM Z continues to lead in mission-critical computing. The introduction of Spyre Accelerator II delivers significant performance gains for AI workloads, enabling faster inferencing and scaling multi-model deployments on-platform.

Performance and Integration

Spyre II enhances IBM Machine Learning for z/OS (MLz) by delivering low-latency inferencing and high throughput for real-time scoring. This capability is critical for workloads such as fraud detection, anti-money laundering, and claims processing, where milliseconds matter. With PCIe-based clustering, IBM Z systems can support up to 48 Spyre cards, enabling massive parallelism without compromising security or resilience.

Key Technical Advancements

• 32-core AI accelerator chip with 25.6B transistors, built on 5nm technology
• Tight integration with Telum II processors for secure, high-speed inferencing
• Support for generative and agentic AI directly on IBM Z, enabling advanced use cases like LLM-based anomaly detection

What’s new in Spyre 1 vs. Spyre II

The following table lists the features of Spyre I and Spyre II:

Spyre II not only increases core count and clustering capacity but also introduces support for generative and agentic AI workloads, making it ideal for multi-model deployments and advanced analytics.

Impact on MLz Workflows

MLz users can now deploy and score models at scale with minimal changes to existing workflows. The standard process—train, package, deploy—remains intact, but hardware acceleration improves throughput and responsiveness. This means:

• Faster scoring for real-time applications
• Ability to run multiple models concurrently
• On-premises generative AI with full data privacy and compliance

Spyre II positions IBM Z as a platform for enterprise AI that combines performance, security, and operational efficiency—all without moving sensitive data off-platform.