26 TOPSHailo-8 · 2.5W · PCIe M.2
01Overview
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.
02What Spikedge Optimizes
- 01HailoRT SDK integration with GStreamer zero-copy pipeline
- 02ONNX/TFLite model compilation via Hailo Model Zoo + Dataflow Compiler
- 03INT8 quantization with sensitivity analysis — minimal accuracy loss
- 04Multi-network scheduling on a single Hailo-8 device
- 05PCIe bandwidth optimization for high-throughput multi-camera scenarios
03Specifications
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)
To evaluate your Hailo AI Processors needs on your own platform, schedule an embedded architecture audit or scope your platform class with the system requirements calculator.
Schedule Architecture Audit
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 AuditRelated capabilities