INTRODUCING RAIDEN:

The Physical AI compute platform that adapts as models change and scales as systems grow

Physical AI is more than one fast model. It is a continuous loop of sensing, reasoning, planning, and acting. RAIDEN-X4 is a scalable, energy-efficient chiplet platform designed to run that entire workload. It combines our new DNA-X architecture with on-package 64-bit RISC-V cores and scales to 3,360 FP4 TFLOPS and 256 GB of LPDDR5X memory. Talk to our team about how RAIDEN can power robotics, Industry 4.0, smart infrastructure, and edge data centers. You can also inquire about our Early Access Program ahead of customer sampling in early 2027.

LEARN ABOUT THE STACK:

One Architecture, One Software Stack

Models you build on SAKURA-II today move forward with you. MERA supports Hugging Face, ONNX, ExecuTorch, and TensorFlow Lite on x86, Arm, and RISC-V hosts, and extends to RAIDEN. It's new silicon with a proven toolchain.

EXPERIENCE SAKURA-II LIVE:

Generative AI at the edge, shipping today

Visit our Venetian Hospitality Suite to see live demonstrations of our latest models, including LLMs, VLMs, ViTs, and other multimodal workloads, running locally on the SAKURA-II Edge AI Platform with no reliance on cloud infrastructure. Built for demanding edge workloads, SAKURA-II delivers up to 60 TOPS (INT8) in a typical 8 W power envelope, with low-latency Batch=1 inference, up to 32 GB of DRAM, and best-in-class memory bandwidth. Compact M.2 and PCIe form factors bring advanced AI closer to where data is generated, from robotics and manufacturing to aerospace, defense, and embedded systems. Every SAKURA-II is supported by the MERA software stack and the MERA Studio desktop app.

INTRODUCING MERA STUDIO:

From model to live demo in minutes, with no command line

MERA Studio is EdgeCortix’s desktop app for compiling, running, and benchmarking AI models on SAKURA-II without a command line. Choose a ready-to-run model or bring your own weights, then compile it in a few clicks and test it on bundled datasets or your own images. Try live object detection with your webcam, chat with text and vision-language models, or transcribe audio from your microphone in real time. Compare on-device latency side by side at each stage to see which model and settings perform best for your workload.