Find edge AI hardware
Task oriented object detection · model-size class nano (YOLO-n) · batch 1 · 13 platforms
Not like-for-like across groups: same model-size class and batch, but 4 different model/precision combinations. Compare within a group; each group is sorted by result.
YOLO26n-obb · FP16 (1)
| Platform | Result | Model · precision · stack | Grade · date | Power | Price | Source |
|---|---|---|---|---|---|---|
| Apple M4 (10-core GPU, 120 GB/s) | 175 FPS latency_e2e b=1 | YOLO26n-obb FP16 · Ultralytics 8.4.168 Core AI (coreai-torch 0.4.3), macOS 27.0 | B 2026-10-01 (page last edited) | unknown no power published | unknown | Ultralytics documentation: Apple Core AI integration benchmark |
yolov11n_obb · INT8 (Hailo quantized) (3)
| Platform | Result | Model · precision · stack | Grade · date | Power | Price | Source |
|---|---|---|---|---|---|---|
| Hailo-10H (M.2/PCIe, x86 host) | 166 FPS throughput b=1 | yolov11n_obb INT8 (Hailo quantized) · Hailo DFC v5.4.0 / HailoRT (model zoo) | D 2026-09-03 (page last edited) | 2.50 W reported in result source — spec | unknown | Hailo Model Zoo public models (Hailo-10H) |
| Hailo-8 (M.2/PCIe, x86 host) | 100 FPS throughput b=1 | yolov11n_obb INT8 (Hailo quantized) · Hailo DFC v2.20.0 / HailoRT (model zoo) | D 2026-10-15 (page last edited) | 2.50 W reported in result source — spec | unknown | Hailo Model Zoo public models (Hailo-8) |
| Hailo-8L (M.2/PCIe, x86 host) | 78.50 FPS throughput b=1 | yolov11n_obb INT8 (Hailo quantized) · Hailo DFC v2.20.0 / HailoRT (model zoo) | D 2026-10-15 (page last edited) | 1.50 W reported in result source — spec | unknown | Hailo Model Zoo public models (Hailo-8L) |
yolov8n-obb · INT8 (8)
| Platform | Result | Model · precision · stack | Grade · date | Power | Price | Source |
|---|---|---|---|---|---|---|
| Rockchip RK3576 (NPU, single core of 2) | 101 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RV1126B (NPU) | 85.00 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RK3588 (NPU, single core of 3) | 77.10 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RK1808 (NPU) | 42.80 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RK3562 (NPU) | 42.10 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RV1126 (NPU) | 37.30 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RK3566/RK3568 (NPU) | 35.50 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
| Rockchip RV1109 (NPU) | 25.10 FPS throughput b=1 | yolov8n-obb INT8 · RKNN-Toolkit2 / RKNPU (rknn_model_zoo) | C 2026-10-13 (page last edited) | unknown no power published | unknown | Rockchip rknn_model_zoo README benchmark table |
yolov8n-obb · INT8 (VNNX) (1)
| Platform | Result | Model · precision · stack | Grade · date | Power | Price | Source |
|---|---|---|---|---|---|---|
| Microchip PolarFire SoC Video Kit (VectorBlox CoreVectorBlox) | 7.03 FPS inference_time b=1 | yolov8n-obb INT8 (VNNX) · VectorBlox SDK (VNNX), no compression | D 2026-09-11 (page last edited) | unknown no power published | unknown | Microchip VectorBlox SDK tutorial metrics |
Confidence: A audited MLPerf closed, B documented third-party, C community, D vendor-published, E estimate. Unknown is never assumed. Other like-for-like groups: nano (YOLO-n), batch 1 (13); medium (YOLO-m), batch 1 (3); medium (YOLO-m), batch 8 (3); nano (YOLO-n), batch 8 (3); small (YOLO-s), batch 1 (3); small (YOLO-s), batch 8 (3)