{ "query": { "task": "image classification" }, "task": "image classification", "size_class": "mobile-class CNN", "batch": "1", "mixed": false, "mixed_warning": null, "groups": [ { "size": "mobile-class CNN", "batch": "1", "platforms": 16, "rows": 58 }, { "size": "mid-size CNN", "batch": "1", "platforms": 13, "rows": 46 }, { "size": "nano (YOLO-n)", "batch": "1", "platforms": 4, "rows": 7 }, { "size": "mobile-class CNN", "batch": "8", "platforms": 3, "rows": 6 }, { "size": "mid-size CNN", "batch": "8", "platforms": 3, "rows": 3 }, { "size": "large (YOLO-l)", "batch": "1", "platforms": 1, "rows": 1 }, { "size": "medium (YOLO-m)", "batch": "1", "platforms": 1, "rows": 1 }, { "size": "small (YOLO-s)", "batch": "1", "platforms": 1, "rows": 1 }, { "size": "xlarge (YOLO-x)", "batch": "1", "platforms": 1, "rows": 1 }, { "size": "mid-size CNN", "batch": "unknown", "platforms": 41, "rows": 416 }, { "size": "mobile-class CNN", "batch": "unknown", "platforms": 31, "rows": 307 }, { "size": "unknown", "batch": "1", "platforms": 5, "rows": 53 }, { "size": "unknown", "batch": "unknown", "platforms": 6, "rows": 344 } ], "notes": [], "count": 16, "results": [ { "platform": "Qualcomm Snapdragon X2 Elite (CRD)", "platform_url": "/platform/qualcomm-snapdragon-x2-elite-crd.html", "perf": 3952.57, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v2", "precision": "w8a16", "stack": "Qualcomm AI Hub onnx (QAIRT 2.50.0.260828221209)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026 (year only)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "AI PC", "vendor": "Qualcomm", "source": { "name": "Qualcomm AI Hub Models perf.yaml (BSD-3)", "url": "https://github.com/qualcomm/ai-hub-models/tree/main/src/qai_hub_models/models/mobilenet_v2", "date": "2026 (repo main)" }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 1, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Hailo-10H (M.2/PCIe, x86 host)", "platform_url": "/platform/hailo-10h-m-2-pcie-x86-host.html", "perf": 2798, "perf_unit": "FPS", "metric": "throughput", "model": "mobilenet_v2_1.0", "precision": "INT8 (Hailo quantized)", "stack": "Hailo DFC v5.4.0 / HailoRT (model zoo)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": 2.5, "power_label": "reported in result source — spec", "power_kind": "spec", "perf_per_w": 1119, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "module typical (vendor spec)", "price_usd": null, "price_label": "unknown", "form_factor": "AI accelerator (M.2)", "vendor": "Hailo", "source": { "name": "Hailo Model Zoo public models (Hailo-10H)", "url": "https://github.com/hailo-ai/hailo_model_zoo/tree/v5.4.0/docs/public_models/HAILO10H", "date": null }, "power_source": "https://github.com/hailo-ai/hailo_model_zoo/tree/v5.4.0/docs/public_models/HAILO10H", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.8367, "unknown_badges": [ "price unknown" ] }, { "platform": "Hailo-8 (M.2/PCIe, x86 host)", "platform_url": "/platform/hailo-8-m-2-pcie-x86-host.html", "perf": 2597, "perf_unit": "FPS", "metric": "throughput", "model": "mobilenet_v2_1.0", "precision": "INT8 (Hailo quantized)", "stack": "Hailo DFC v2.20.0 / HailoRT (model zoo)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": 2.5, "power_label": "reported in result source — spec", "power_kind": "spec", "perf_per_w": 1039, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "chip typical (vendor spec)", "price_usd": null, "price_label": "unknown", "form_factor": "AI accelerator (M.2)", "vendor": "Hailo", "source": { "name": "Hailo Model Zoo public models (Hailo-8)", "url": "https://github.com/hailo-ai/hailo_model_zoo/tree/v2.20.0/docs/public_models/HAILO8", "date": null }, "power_source": "https://github.com/hailo-ai/hailo_model_zoo/tree/v2.20.0/docs/public_models/HAILO8", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.7768, "unknown_badges": [ "price unknown" ] }, { "platform": "Hailo-8L (M.2/PCIe, x86 host)", "platform_url": "/platform/hailo-8l-m-2-pcie-x86-host.html", "perf": 1738, "perf_unit": "FPS", "metric": "throughput", "model": "mobilenet_v2_1.0", "precision": "INT8 (Hailo quantized)", "stack": "Hailo DFC v2.20.0 / HailoRT (model zoo)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": 1.5, "power_label": "reported in result source — spec", "power_kind": "spec", "perf_per_w": 1159, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "chip typical (vendor spec)", "price_usd": null, "price_label": "unknown", "form_factor": "AI accelerator (M.2)", "vendor": "Hailo", "source": { "name": "Hailo Model Zoo public models (Hailo-8L)", "url": "https://github.com/hailo-ai/hailo_model_zoo/tree/v2.20.0/docs/public_models/HAILO8L", "date": null }, "power_source": "https://github.com/hailo-ai/hailo_model_zoo/tree/v2.20.0/docs/public_models/HAILO8L", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.7199, "unknown_badges": [ "price unknown" ] }, { "platform": "Qualcomm Snapdragon X Elite (CRD)", "platform_url": "/platform/qualcomm-snapdragon-x-elite-crd.html", "perf": 2202.64, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v2", "precision": "w8a8", "stack": "Qualcomm AI Hub onnx (QAIRT 2.50.0.260828221209)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026 (year only)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "AI PC", "vendor": "Qualcomm", "source": { "name": "Qualcomm AI Hub Models perf.yaml (BSD-3)", "url": "https://github.com/qualcomm/ai-hub-models/tree/main/src/qai_hub_models/models/mobilenet_v2", "date": "2026 (repo main)" }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.5573, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Qualcomm Dragonwing IQ-9075 EVK", "platform_url": "/platform/qualcomm-dragonwing-iq-9075-evk.html", "perf": 1964.64, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v2", "precision": "w8a8", "stack": "Qualcomm AI Hub qnn_dlc (QAIRT 2.50.0.260828221209)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026 (year only)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC dev kit", "vendor": "Qualcomm", "source": { "name": "Qualcomm AI Hub Models perf.yaml (BSD-3)", "url": "https://github.com/qualcomm/ai-hub-models/tree/main/src/qai_hub_models/models/mobilenet_v2", "date": "2026 (repo main)" }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.4971, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Qualcomm Dragonwing RB3 Gen 2 (QCS6490)", "platform_url": "/platform/qualcomm-dragonwing-rb3-gen-2-qcs6490.html", "perf": 939.85, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v2", "precision": "w8a8", "stack": "Qualcomm AI Hub onnx (QAIRT 2.50.0.260828221209)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026 (year only)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC dev kit", "vendor": "Qualcomm", "source": { "name": "Qualcomm AI Hub Models perf.yaml (BSD-3)", "url": "https://github.com/qualcomm/ai-hub-models/tree/main/src/qai_hub_models/models/mobilenet_v2", "date": "2026 (repo main)" }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.2378, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Microchip PolarFire SoC Video Kit (VectorBlox CoreVectorBlox)", "platform_url": "/platform/microchip-polarfire-soc-video-kit-vectorblox-corevectorblox.html", "perf": 671.59, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet-v1-0.25-128", "precision": "INT8 (VNNX)", "stack": "VectorBlox SDK (VNNX), no compression", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "FPGA SoC (edge)", "vendor": "Microchip", "source": { "name": "Microchip VectorBlox SDK tutorial metrics", "url": "https://github.com/Microchip-Vectorblox/VectorBlox-SDK/blob/master/docs/tutorial_metrics_appendix.md", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1699, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "NXP i.MX 9 with eIQ Neutron-S NPU (i.MX 95/952/943; doc does not specify)", "platform_url": "/platform/nxp-i-mx-9-with-eiq-neutron-s-npu-i-mx-95-952-943-doc-does-not-specify.html", "perf": 619.04, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenetv2-12 (neutron-converted ONNX)", "precision": "INT8 (quantized)", "stack": "ONNX Runtime 1.24.3 onnxruntime_perf_test; delegate: Neutron execution provider", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC", "vendor": "NXP", "source": { "name": "NXP UG10166 (vendor document)", "url": "https://www.nxp.com/docs/en/user-guide/UG10166.pdf", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1566, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "NXP i.MX 95", "platform_url": "/platform/nxp-i-mx-95.html", "perf": 588.24, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v1_1.0_224_quant", "precision": "INT8 (quantized)", "stack": "TensorFlow Lite 2.19.0 label_image.py (Python); delegate: Neutron Delegate", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC", "vendor": "NXP", "source": { "name": "NXP UG10166 (vendor document)", "url": "https://www.nxp.com/docs/en/user-guide/UG10166.pdf", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1488, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Google Coral USB Accelerator (Edge TPU) on Xeon Gold 6154 host, USB 3.0", "platform_url": "/platform/google-coral-usb-accelerator-edge-tpu-on-xeon-gold-6154-host-usb-3-0.html", "perf": 416.67, "perf_unit": "FPS", "metric": "inference_time", "model": "MobileNet v1 (224x224)", "precision": "INT8", "stack": "TensorFlow Lite C++ benchmark (Edge TPU runtime)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": 2, "power_label": "platform spec envelope (not measured with this result)", "power_kind": "spec", "perf_per_w": 208.3, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "spec: Edge TPU 4 TOPS at 0.5 W per TOPS (2 W), Edge TPU only; host/board power excluded. Dev Board datasheet: \"4 TOPS @ 2 W\". Price not shown on coral.ai (store is JS buy-flow); left empty", "price_usd": null, "price_label": "unknown", "form_factor": "AI accelerator (USB)", "vendor": "Google", "source": { "name": "Google Coral Edge TPU benchmarks page (Table 1)", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "date": null }, "power_source": "https://coral.ai/docs/edgetpu/benchmarks/", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1426, "unknown_badges": [ "price unknown" ] }, { "platform": "Google Coral Dev Board (Edge TPU)", "platform_url": "/platform/google-coral-dev-board-edge-tpu.html", "perf": 416.67, "perf_unit": "FPS", "metric": "inference_time", "model": "MobileNet v1 (224x224)", "precision": "INT8", "stack": "TensorFlow Lite C++ benchmark (Edge TPU runtime)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": 2, "power_label": "platform spec envelope (not measured with this result)", "power_kind": "spec", "perf_per_w": 208.3, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "spec: Edge TPU 4 TOPS at 0.5 W per TOPS (2 W), Edge TPU only; host/board power excluded. Dev Board datasheet: \"4 TOPS @ 2 W\". Price not shown on coral.ai (store is JS buy-flow); left empty", "price_usd": null, "price_label": "unknown", "form_factor": "edge SBC", "vendor": "Google", "source": { "name": "Google Coral Edge TPU benchmarks page (Table 1)", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "date": null }, "power_source": "https://coral.ai/docs/edgetpu/benchmarks/", "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1426, "unknown_badges": [ "price unknown" ] }, { "platform": "NXP i.MX 8M Plus", "platform_url": "/platform/nxp-i-mx-8m-plus.html", "perf": 389.56, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v1_1.0_224_quant", "precision": "INT8 (quantized)", "stack": "TensorFlow Lite 2.19.0 label_image (C++); delegate: VX Delegate (USE_GPU_INFERENCE=0)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC", "vendor": "NXP", "source": { "name": "NXP UG10166 (vendor document)", "url": "https://www.nxp.com/docs/en/user-guide/UG10166.pdf", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0986, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "NXP i.MX 93", "platform_url": "/platform/nxp-i-mx-93.html", "perf": 256.21, "perf_unit": "FPS", "metric": "inference_time", "model": "mobilenet_v1_1.0_224_quant (vela)", "precision": "INT8 (quantized)", "stack": "TensorFlow Lite 2.19.0 label_image (C++); delegate: Ethos-U Delegate (Vela-compiled model)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SoC", "vendor": "NXP", "source": { "name": "NXP UG10166 (vendor document)", "url": "https://www.nxp.com/docs/en/user-guide/UG10166.pdf", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0648, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Intel Xeon Gold 6154 desktop (CPU only, TFLite)", "platform_url": "/platform/intel-xeon-gold-6154-desktop-cpu-only-tflite.html", "perf": 19.61, "perf_unit": "FPS", "metric": "inference_time", "model": "MobileNet v2 (224x224)", "precision": "FP32/INT8 TFLite on CPU", "stack": "TensorFlow Lite C++ benchmark (Edge TPU runtime)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "server/desktop CPU", "vendor": "Google", "source": { "name": "Google Coral Edge TPU benchmarks page (Table 1)", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.005, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] }, { "platform": "Coral Dev Board CPU (quad Cortex-A53 @1.5 GHz, CPU only)", "platform_url": "/platform/coral-dev-board-cpu-quad-cortex-a53-1-5-ghz-cpu-only.html", "perf": 8.2, "perf_unit": "FPS", "metric": "inference_time", "model": "MobileNet v2 (224x224)", "precision": "TFLite on CPU", "stack": "TensorFlow Lite C++ benchmark (Edge TPU runtime)", "size_class": "mobile-class CNN", "batch": "1", "like_for_like": true, "differs": [], "confidence": "D", "confidence_label": "D – vendor-published", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "undated", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "form_factor": "edge SBC", "vendor": "Google", "source": { "name": "Google Coral Edge TPU benchmarks page (Table 1)", "url": "https://coral.ai/docs/edgetpu/benchmarks/", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0021, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ] } ], "no_data": null, "legend": "Confidence: A audited MLPerf closed, B documented third-party, C community, D vendor-published, E estimate. Unknown is never assumed." }