Our solution
Agentic AI systems are autonomous agents that plan, reason, use tools and act across multi-step workflows, place fundamentally different demands on compute infrastructure than batch training or static inference.
They require low-latency CPU processing for orchestration logic, sustained memory bandwidth for context management, and tight integration with accelerators for the reasoning steps in between. As agentic AI moves from research to production, infrastructure designed for sovereignty and predictable performance becomes a strategic asset.
Typical workloads
-
Multi-agent orchestration and coordination
-
Long-context reasoning and planning
-
Tool-calling and API-driven automation
-
Retrieval-augmented generation (RAG) at scale
-
Autonomous scientific and industrial workflows
Why SiPearl
→ High core count and HBM bandwidth optimised for the memory-intensive demand of large context windows.
→ Low-latency CPU orchestration layer that coordinates agent steps and accelerator handoffs efficiently.
→ European silicon enabling sovereign agentic AI in sensitive sectors: defence, healthcare, public sector.
Technical features of Rhea1
Discover all the details of our CPU
Seine Reference Server, a modular solution dedicated to Rhea1
Multifunctional, flexible and versatile, the Seine Reference Server can be used for validation and testing, as a reference design, for software porting and for demonstrations and customer testing.
It is available in two configurations:
- a single Rhea1 processor connecting to up to two GPUs in a single chassis,
- or as a dual socket Rhea1.
Each configuration supports up to two SATA (Serial Advanced Technology Attachment) disks and up to two PCIe NICs.