HWBench

Which hardware is fastest for small YOLO models at batch 8?

The fastest result is the Hailo-8 (M.2/PCIe, x86 host) at 633 FPS, running yolov5s_wo_spp in INT8 (Hailo quantized). That is the vendor's own figure, from a page last edited on 2026-10-15. Hailo-10H (M.2/PCIe, x86 host) comes next at 370 FPS, then Hailo-8L (M.2/PCIe, x86 host) at 275 FPS. All of these are small models at batch 8, so they compare directly.

Top results
#PlatformResultPowerPerf/WPriceGrade · dateSource
1Hailo-8 (M.2/PCIe, x86 host)633 FPS
yolov5s_wo_spp · INT8 (Hailo quantized)
2.5 W (spec)253 FPS/W (spec power)unknownD 2026-10-15 (page last edited)Hailo Model Zoo public models (Hailo-8)
2Hailo-10H (M.2/PCIe, x86 host)370 FPS
yolov5s_wo_spp · INT8 (Hailo quantized)
2.5 W (spec)148 FPS/W (spec power)unknownD 2026-09-03 (page last edited)Hailo Model Zoo public models (Hailo-10H)
3Hailo-8L (M.2/PCIe, x86 host)275 FPS
yolov5s_wo_spp · INT8 (Hailo quantized)
1.5 W (spec)183 FPS/W (spec power)unknownD 2026-10-15 (page last edited)Hailo Model Zoo public models (Hailo-8L)

Filters: task=object detection, size=small (YOLO-s), batch=8. Power marked “spec” is the vendor's rated figure, not a measurement. Ranking: grade A–C dated results first where available, then headroom and perf/W. Open in /find · JSON

Sources

Generated 2026-10-16 from HWBench data only. All answers · Submit a result

Machine-readable: /find.json (see llms.txt); full tables are in “Show data”. Confidence grades: A audited MLPerf closed · B documented third-party · C community · D vendor-published · E estimate. Unknown values are shown as unknown, never estimated. Generated 2026-10-16 from public sources.