Silicon
Edge AI

Hailo AI Processors

Hailo-8 · Hailo-8L · Hailo-15

26 TOPSHailo-8 · 2.5W · PCIe M.2

Hailo AI processors are purpose-built inference accelerators that run ML models completely independently of the host CPU. At 26 TOPS in 2.5W, Hailo-8 achieves performance/watt ratios impossible with GPU-based solutions. The HailoRT SDK integrates with GStreamer and OpenCV for seamless pipeline integration.

What Spikedge Optimizes

  • HailoRT SDK integration with GStreamer zero-copy pipeline
  • ONNX/TFLite model compilation via Hailo Model Zoo + Dataflow Compiler
  • INT8 quantization with sensitivity analysis — minimal accuracy loss
  • Multi-network scheduling on a single Hailo-8 device
  • PCIe bandwidth optimization for high-throughput multi-camera scenarios

Specifications

AI Performance26 TOPS (Hailo-8) / 13 TOPS (Hailo-8L) / 40 TOPS (Hailo-15)
Power2.5W (Hailo-8) typical
InterfacePCIe 3.0 x4 / M.2 B+M Key
SDKHailoRT, Hailo Dataflow Compiler, Model Zoo
FrameworksONNX, TFLite, PyTorch (via export)

Hailo-8 changes the edge AI calculus

26 TOPS at 2.5W means you can run YOLO on a Raspberry Pi CM4 form-factor. Let's explore what's possible.

Schedule Architecture Audit