Axelera AI unveils Titania chiplet architecture
Axelera AI has unveiled Titania, its third-generation silicon architecture, using a programmable, energy-efficient chiplet design for AI acceleration. The company says Titania will deliver high performance at interactive speeds within existing data center power and cooling limits, rather than requiring the 100-kilowatt-plus liquid-cooled racks used by today’s highest-density GPU systems. Its projected performance includes up to 7x the throughput per watt compared with the best data center grade servers available, alongside up to a 20x improvement in total cost of ownership.
Each Titania-based system-in-package will contain two Titania chiplets. Planned formats include standard PCIe cards and OCP-compliant UBB baseboards carrying eight Titania OAM modules connected through a scale-up network. OEM partners will offer these formats in inference server configurations supporting both air and liquid cooling. Axelera AI also reported confirmed Titania commitments.