What is an AIPU?
The Axelera AIPUs are purpose-built AI inference processors. Designed intentionally to bring the highest performance possible within the smallest of power envelopes.
Axelera has announced three distinct versions of our AIPU to support different use cases, different form factors, and different performance requirements.
- MetisTM: a 214 TOP powerhouse packed into a 4-8watt budget. Metis brings inference to the edge in a compact M.2 module or PCIe card, delivering hundreds to thousands frames per second of real time analysis.
- EuropaTM: a 629 TOP inference accelerator that incorporates on-die pre/post processing, doubles the AIPU core-count and adds 16 vector processing units and scales from "heavy edge" to enterprise data center.
- TitaniaTM: a chiplet that scales the Digital In-Memory compute architecture into high performance compute and AI Gigafactories.
Metis AIPU Architecture
We created the future of efficient inference with industry-leading performance and usability, at a fraction of the cost of existing solutions.
Meet our Metis AI Processing Unit (AIPU). Optimized for vision use cases, with the flexibility to run local language models, and scalability to meet the future demands of inference workloads.
214 TOPs
Europa AIPU Architecture
Europa extends AI inference capabilities and unlocks AI applications that more compact devices can’t handle. Consolidate workloads across facilities, deploy robotics and autonomous systems, and run complex models locally without datacenter infrastructure. Process in real-time and scale your operations from a single deployment, all in a power-envelope small enough to fit your existing footprint.
629 TOPs
Axelera Embedded 110m
Powered by the Metis architecture
Providing best-in-class performance, the Axelera Embedded 110m is the best solution for AI acceleration in constrained environments. Powered by a single Metis AIPU and containing 1 GB DRAM dedicated memory, it minimizes power consumption and simplifies integration. It opens the door for every application to benefit from AI inference processing. Providing a significant boost to any system with an NGFF (Next Generation Form Factor) M.2 socket.
1 GB DRAM dedicated memory
Axelera Embedded 113m
Powered by the Metis architecture
Providing breakthrough AI performance, the Axelera Embedded 113m card is the ideal solution to unlock LLM and VLM applications on edge and embedded devices. Powered by the Metis AIPU and available with up to 8 GB of dedicated memory, it maximizes efficiency while minimizing power consumption. It enables multi-camera vision inference with cascaded or parallel models, opening new possibilities for intelligent applications. With secure boot, extended temperature options, and effortless software integration through the Voyager® toolchain, the Embedded 113c delivers uncompromised performance and flexibility in a compact and energy efficient form factor.
up to 8 GB DRAM dedicated memory
Axelera Edge 130p
Powered by the Metis architecture
For companies and developers seeking more performance, Axelera AI has created a range of PCIe accelerators. Powered by one Metis AIPU, this card can deliver up to 214 TOPS, enabling it to handle the most demanding vision and language applications. This performance is enhanced through the exceptional developer experience offered by the Voyager® toolchain to accelerate application deployment.
Up to 214 TOPS
Up to 3200 FPS for ResNet-50
Axelera Server 150p
Powered by 4 quad-core Metis AIPUs
Need best in class performance? Powered by 4 Metis AIPUs, this card can provide peak performance of 856 TOPS, and is capable of tackling the most challenging vision applications. This performance is enhanced through the exceptional developer experience offered by the Voyager toolchain to accelerate application deployment.
Up to 856 TOPS
Up to 12800 FPS for ResNet-50
Up to 38884 FPS for MobileNet V2-1.0
Axelera Embedded 111c
A Single Board Computer powered by Metis and an ARM-based RK3588 CPU
The Axelera Embedded 111c offers everything needed to build accurate and fast inference solutions, all in a compact compute board, and right out of the box.
Equipped with a quad-core Metis AIPU and an RK3588 processor, this single-board computer (SBC) offers exceptional performance for demanding workloads in an energy-efficient design. Its architecture includes a four-lane PCIe connection for seamless inference acceleration, making it ideal for multimodal applications using parallel or cascade model processing.
Connectivity is robust, with four USB ports, HDMI 2.0, dual Gigabit LAN, and versatile GPIO, UART, and I2C interfaces. Storage is equally flexible, featuring 64GB onboard eMMC, two SATA ports, and a microSD slot. Paired with Axelera’s Voyager toolchain, the board simplifies AI model deployment and optimization, ensuring faster prototyping and reduced time-to-market. Whether running PyTorch, ONNX or TensorFlow, developers benefit from a scalable, ready-to-deploy platform that balances efficiency, versatility, and power in edge AI compute.
Want to get started immediately?
One SDK. Every Platform.
Deploy your AI applications across both Metis and Europa with the same Voyager SDK. Develop once and scale across the entire Axelera platform, from compact edge devices to enterprise servers. The Voyager SDK uses simple YAML configuration to automatically compile, optimize, and deploy your complete application pipeline, whether you're running vision models, image processing, or language models. Build on one platform, deploy anywhere your applications need to run.
"The Ultralytics YOLO family of models has set the industry standard for computer vision globally. The performance we've achieved running YOLO models on Axelera AI's Metis platform has been remarkable, delivering the speed and efficiency our users demand for real-world computer vision applications. Our collaboration has proven that advanced AI inference can be both powerful and accessible. With Europa's expanded capabilities coming next year, we're excited to broaden YOLO's deployment across entirely new use cases.
Advantech has successfully integrated Axelera AI's Metis AIPU into our products, bringing efficient AI processing to customers worldwide. The real-world performance and reliability have exceeded expectations, enabling breakthrough applications in manufacturing automation and intelligent infrastructure. We're thrilled to support the Europa launch and look forward to incorporating its 629 TOPS of processing...[read more]
“SECO is a leading provider of Edge AI solutions, empowering customers to bring intelligence and efficiency directly to their devices. We integrated Metis into our Edge AI product line earlier this year and have been impressed by the significant boost in performance and scalability it delivers. We are excited to partner with Axelera as they continue to expand their world-class inference portfolio and enable even broader adoption of AI at the edge across industries.
