Find edge AI hardware
Task image classification · model-size class by model · batch per group · 50 platforms · edge hardware only · model_key "resnet-50"
Results are compared only within the same model. 1 model have results for this task.
How to read power: results are compared only within the same model. By efficiency ranks only results whose source publishes a power figure; By speed ranks every result. Measured = measured for this workload. Vendor-measured (typical/avg) = the vendor’s own measurement at typical or average load, not an independent test and not a maximum, so fps/W looks better than at max TDP. Chip, Module or Board TDP = the vendor’s published maximum, not measured. Typical power, not max TDP = only a typical figure is published, so fps/W looks better than at full load. Scope says what the watts cover: Accelerator only (host not included), Module, or Whole system. Max of range = top of a published range, so fps/W is a floor (≥). “— not published” = the source gives no power; we never estimate one.
ResNet-50 50 platforms · 12 with published power
| Rank | Platform | FPS/W | FPS | Power | Model · precision · batch | Grade · date | Source |
|---|---|---|---|---|---|---|---|
| 1 | Intel Celeron 6305E | 13.26Chip TDP · Accelerator only | 199 FPS | 15.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 2 | Intel Core i7-1185G7 | ≥12.77Chip TDP · Accelerator only · floor | 358 FPS | 28.00 WAccelerator only, host not includedChip TDP · max of range 12-28 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 3 | Intel Atom x7425E | 10.48Chip TDP · Accelerator only | 126 FPS | 12.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 4 | Intel Core Ultra 7 265H | ≥9.75Chip TDP · Accelerator only · floor | 1121 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 20-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 5 | Intel Core Ultra 7 155H | ≥8.52Chip TDP · Accelerator only · floor | 980 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 20-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 6 | Intel Core i7-1185GRE | ≥7.97Chip TDP · Accelerator only · floor | 223 FPS | 28.00 WAccelerator only, host not includedChip TDP · max of range 12-28 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 7 | AMD Kria KV260 Vision AI Starter Kit (K26 SOM) | 6.77Module TDP · Module | 102 FPS | 15.00 WModuleModule TDP | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo |
| 8 | Intel Core i7-1355U | ≥6.10Chip TDP · Accelerator only · floor | 335 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 12-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 9 | Intel Atom X6425E | 4.91Chip TDP · Accelerator only | 58.96 FPS | 12.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 10 | Intel Core i7-12700H | ≥4.54Chip TDP · Accelerator only · floor | 522 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 45-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 11 | Intel Core i5-1335U | ≥4.21Chip TDP · Accelerator only · floor | 231 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 12-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 12 | Intel Core i5-1235U | ≥3.64Chip TDP · Accelerator only · floor | 200 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 15-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| – | Lenovo ThinkEdge SE455i V3 (2x NVIDIA L4-PCIe-24GB) | 25201 FPS | — not published | MLPerf resnetint8 · batch unknown | A 2026-04-01 (published) | MLPerf Inference v6.0, closed division, edge suiteMLPerf Inference v6.0 | |
| – | NVIDIA Jetson AGX Orin 64GB | 6424 FPS | — not published | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) | |
| – | NVIDIA Jetson AGX Orin (MaxQ) | 150● Measured · Whole system | 3526 FPS | 23.57 WWhole system Measured | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| – | AMD Versal AI Core VCK190 (VC1902) | 3027 FPS | — not published | tf2_resnet50_imagenet_224_224_7.76GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| – | NVIDIA Jetson Orin NX 16GB | ≥66.01Module TDP · Module · floor | 2641 FPS | 40.00 WModuleModule TDP · max of range 10-40 W | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| – | Qualcomm Snapdragon X2 Elite (CRD) | 2445 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| – | Axelera Metis M.2 card | 270● Measured · Module | 2050 FPS | 7.60 WModule Measured | ResNet-50 v1.5INT8 (vendor quantized) · batch unknown | C 2026-05-07 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p04-axelera-metis), 2026-05-07 |
| – | Intel Core Ultra 9 288V | ≥53.54Chip TDP · Accelerator only · floor | 1981 FPS | 37.00 WAccelerator only, host not includedChip TDP · max of range 17-37 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| – | Axelera Metis Embedded 113m M.2 card | ≥205Module TDP · Module · floor | 1849 FPS | 9.00 WModuleModule TDP · max of range 3.5-9 W | ResNet-50 v1.5INT8 · batch 1 | C 2026-10-15 (page last edited) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| – | Axelera Metis Edge 130p PCIe card | ≥122Board TDP · Module · floor | 1825 FPS | 15.00 WModuleBoard TDP · max of range 8-15 W | ResNet-50 v1.5INT8 · batch 1 | C 2026-10-15 (page last edited) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| – | NVIDIA Jetson Orin NX 16GB (MaxQ) | 113● Measured · Whole system | 1682 FPS | 14.95 WWhole system Measured | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| – | Hailo-8 (M.2/PCIe, x86 host) | 1372 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-10-15 (page last edited) | Hailo Model Zoo public models (Hailo-8)hailo_model_zoo v2.20.0, DFC v2.20.0 | |
| – | Hailo-8 AI processor (chip, HNC18BI11BH) | 381Vendor-measured (typical/avg) · Accelerator only | 1371 FPS | 3.60 WAccelerator only, host not includedVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch 1 | D undated | Vendor page · Hailo-8 Datasheet Revision 1.9.2 (April 2026)Hailo-8 Datasheet Revision 1.9.2 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| – | Hailo-8 M.2 module | 343● Measured · Module | 1371 FPS | 4.00 WModule Measured | ResNet-50INT8 (vendor quantized) · batch unknown | C 2026-05-05 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p03-ina228), 2026-05-05 |
| – | Qualcomm Dragonwing IQ-9075 EVK | 1192 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| – | Qualcomm Snapdragon X Elite (CRD) | 1186 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| – | MemryX MX3 M.2 module | 115Module TDP · Module | 1148 FPS | 10.00 WModuleModule TDP | ResNet-50not stated (MX3 activations BF16 / weights 4-8-16 bit per vendor) · batch 1 | C 2026-07-15 (published) | MemryX Model Explorer (SDK 2.2)MemryX SDK 2.2 model explorer |
| – | DEEPX DX-M1 M.2 LPDDR5x2 module | 215● Measured · Module | 1009 FPS | 4.70 WModule Measured | ResNet-50INT8 (vendor quantized) · batch unknown | C 2026-06-10 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p05-deepx-m1), 2026-06-10 |
| – | Hailo-8 M.2 Key M ET module (HM218B1C2HAE / HM218B1C2FAE) | 277Vendor-measured (typical/avg) · Module | 915 FPS | 3.30 WModuleVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch unknown | D undated | Vendor page · Datasheet (mirror)Hailo-8 M.2 Key M ET Module Data Sheet Rev. 2.2; Table 3 - Power Consumption for typical configurations (typical at 25 C) |
| – | Intel Core Ultra 7 255H (Lenovo ThinkEdge SE100) | 835 FPS | — not published | MLPerf resnetINT8 · batch unknown | A 2026-04-01 (published) | MLPerf Inference v6.0, closed division, edge suiteMLPerf Inference v6.0 | |
| – | Hailo-8L AI processor (chip, HNC1LBI11BH) | 263Vendor-measured (typical/avg) · Accelerator only | 500 FPS | 1.90 WAccelerator only, host not includedVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch 8 | D undated | Vendor page · Hailo-8L Datasheet Revision 1.9.3 (April 2026)Hailo-8L Datasheet Revision 1.9.3 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| – | Hailo-8L M.2 Key B+M ET module | 228Vendor-measured (typical/avg) · Module | 433 FPS | 1.90 WModuleVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch unknown | D undated | Vendor page · Datasheet (mirror)Hailo-8L M.2 Key B+M ET Module Data Sheet Rev. 4.0; Table 3 - Power consumption for typical configurations (typical at 25 C) |
| – | Qualcomm Dragonwing RB3 Gen 2 (QCS6490) | 379 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| – | Hailo-10H (M.2/PCIe, x86 host) | 331 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-09-03 (page last edited) | Hailo Model Zoo public models (Hailo-10H)hailo_model_zoo v5.4.0, DFC v5.4.0 | |
| – | Terasic DE10-Agilex Development Board (Agilex 7, FPGA AI Suite) | 220 FPS | — not published | resnet-50-tfnot stated in table (set by .arch file) · batch 1 | D 2026-08-10 (published) | Altera FPGA AI Suite Handbook (performance tables)FPGA AI Suite 2026.1.1 (doc 863373) | |
| – | AMD Zynq UltraScale+ ZCU102 (ZU9EG) | 214 FPS | — not published | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| – | AMD Zynq UltraScale+ ZCU104 (ZU7EV) | 183 FPS | — not published | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| – | Hailo-8L (M.2/PCIe, x86 host) | 138 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-10-15 (page last edited) | Hailo Model Zoo public models (Hailo-8L)hailo_model_zoo v2.20.0, DFC v2.20.0 | |
| – | Rockchip RK3588 (NPU, single core of 3) | 113 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Intel N100 | 18.39Chip TDP · Accelerator only | 110 FPS | 6.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| – | Rockchip RK3576 (NPU, single core of 2) | 96.00 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Rockchip RV1126B (NPU) | 78.20 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Altera Agilex 5 FPGA E-Series 065B Modular Development Kit (FPGA AI Suite, OFS PCIe) | 64.00 FPS | — not published | resnet-50-tfnot stated in table (set by .arch file) · batch 1 | D 2026-08-10 (published) | Altera FPGA AI Suite Handbook (performance tables)FPGA AI Suite 2026.1.1 (doc 863373) | |
| – | Rockchip RK3562 (NPU) | 55.00 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Rockchip RK3566/RK3568 (NPU) | 38.60 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Rockchip RK1808 (NPU) | 37.10 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Rockchip RV1126 (NPU) | 36.20 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| – | Rockchip RV1109 (NPU) | 24.40 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) |
| Rank | Platform | FPS/W | FPS | Power | Model · precision · batch | Grade · date | Source |
|---|---|---|---|---|---|---|---|
| 1 | Lenovo ThinkEdge SE455i V3 (2x NVIDIA L4-PCIe-24GB) | 25201 FPS | — not published | MLPerf resnetint8 · batch unknown | A 2026-04-01 (published) | MLPerf Inference v6.0, closed division, edge suiteMLPerf Inference v6.0 | |
| 2 | NVIDIA Jetson AGX Orin 64GB | 6424 FPS | — not published | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) | |
| 3 | NVIDIA Jetson AGX Orin (MaxQ) | 150● Measured · Whole system | 3526 FPS | 23.57 WWhole system Measured | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| 4 | AMD Versal AI Core VCK190 (VC1902) | 3027 FPS | — not published | tf2_resnet50_imagenet_224_224_7.76GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| 5 | NVIDIA Jetson Orin NX 16GB | ≥66.01Module TDP · Module · floor | 2641 FPS | 40.00 WModuleModule TDP · max of range 10-40 W | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| 6 | Qualcomm Snapdragon X2 Elite (CRD) | 2445 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| 7 | Axelera Metis M.2 card | 270● Measured · Module | 2050 FPS | 7.60 WModule Measured | ResNet-50 v1.5INT8 (vendor quantized) · batch unknown | C 2026-05-07 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p04-axelera-metis), 2026-05-07 |
| 8 | Intel Core Ultra 9 288V | ≥53.54Chip TDP · Accelerator only · floor | 1981 FPS | 37.00 WAccelerator only, host not includedChip TDP · max of range 17-37 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 9 | Axelera Metis Embedded 113m M.2 card | ≥205Module TDP · Module · floor | 1849 FPS | 9.00 WModuleModule TDP · max of range 3.5-9 W | ResNet-50 v1.5INT8 · batch 1 | C 2026-10-15 (page last edited) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| 10 | Axelera Metis Edge 130p PCIe card | ≥122Board TDP · Module · floor | 1825 FPS | 15.00 WModuleBoard TDP · max of range 8-15 W | ResNet-50 v1.5INT8 · batch 1 | C 2026-10-15 (page last edited) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| 11 | NVIDIA Jetson Orin NX 16GB (MaxQ) | 113● Measured · Whole system | 1682 FPS | 14.95 WWhole system Measured | MLPerf ResNetINT8/FP16 (MLPerf closed) · batch unknown | D 2023-09-08 (published) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| 12 | Hailo-8 (M.2/PCIe, x86 host) | 1372 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-10-15 (page last edited) | Hailo Model Zoo public models (Hailo-8)hailo_model_zoo v2.20.0, DFC v2.20.0 | |
| 13 | Hailo-8 AI processor (chip, HNC18BI11BH) | 381Vendor-measured (typical/avg) · Accelerator only | 1371 FPS | 3.60 WAccelerator only, host not includedVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch 1 | D undated | Vendor page · Hailo-8 Datasheet Revision 1.9.2 (April 2026)Hailo-8 Datasheet Revision 1.9.2 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| 14 | Hailo-8 M.2 module | 343● Measured · Module | 1371 FPS | 4.00 WModule Measured | ResNet-50INT8 (vendor quantized) · batch unknown | C 2026-05-05 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p03-ina228), 2026-05-05 |
| 15 | Qualcomm Dragonwing IQ-9075 EVK | 1192 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| 16 | Qualcomm Snapdragon X Elite (CRD) | 1186 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| 17 | MemryX MX3 M.2 module | 115Module TDP · Module | 1148 FPS | 10.00 WModuleModule TDP | ResNet-50not stated (MX3 activations BF16 / weights 4-8-16 bit per vendor) · batch 1 | C 2026-07-15 (published) | MemryX Model Explorer (SDK 2.2)MemryX SDK 2.2 model explorer |
| 18 | Intel Core Ultra 7 265H | ≥9.75Chip TDP · Accelerator only · floor | 1121 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 20-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 19 | DEEPX DX-M1 M.2 LPDDR5x2 module | 215● Measured · Module | 1009 FPS | 4.70 WModule Measured | ResNet-50INT8 (vendor quantized) · batch unknown | C 2026-06-10 (published) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p05-deepx-m1), 2026-06-10 |
| 20 | Intel Core Ultra 7 155H | ≥8.52Chip TDP · Accelerator only · floor | 980 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 20-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 21 | Hailo-8 M.2 Key M ET module (HM218B1C2HAE / HM218B1C2FAE) | 277Vendor-measured (typical/avg) · Module | 915 FPS | 3.30 WModuleVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch unknown | D undated | Vendor page · Datasheet (mirror)Hailo-8 M.2 Key M ET Module Data Sheet Rev. 2.2; Table 3 - Power Consumption for typical configurations (typical at 25 C) |
| 22 | Intel Core Ultra 7 255H (Lenovo ThinkEdge SE100) | 835 FPS | — not published | MLPerf resnetINT8 · batch unknown | A 2026-04-01 (published) | MLPerf Inference v6.0, closed division, edge suiteMLPerf Inference v6.0 | |
| 23 | Intel Core i7-12700H | ≥4.54Chip TDP · Accelerator only · floor | 522 FPS | 115 WAccelerator only, host not includedChip TDP · max of range 45-115 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 24 | Hailo-8L AI processor (chip, HNC1LBI11BH) | 263Vendor-measured (typical/avg) · Accelerator only | 500 FPS | 1.90 WAccelerator only, host not includedVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch 8 | D undated | Vendor page · Hailo-8L Datasheet Revision 1.9.3 (April 2026)Hailo-8L Datasheet Revision 1.9.3 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| 25 | Hailo-8L M.2 Key B+M ET module | 228Vendor-measured (typical/avg) · Module | 433 FPS | 1.90 WModuleVendor-measured (typical/avg) | ResNet-50not stated (Hailo INT8 quantized) · batch unknown | D undated | Vendor page · Datasheet (mirror)Hailo-8L M.2 Key B+M ET Module Data Sheet Rev. 4.0; Table 3 - Power consumption for typical configurations (typical at 25 C) |
| 26 | Qualcomm Dragonwing RB3 Gen 2 (QCS6490) | 379 FPS | — not published | resnet50w8a8 · batch 1 | D 2026-10-07 (published) | Qualcomm AI Hub Models perf.yaml (BSD-3)ai-hub-models v0.64.0 | |
| 27 | Intel Core i7-1185G7 | ≥12.77Chip TDP · Accelerator only · floor | 358 FPS | 28.00 WAccelerator only, host not includedChip TDP · max of range 12-28 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 28 | Intel Core i7-1355U | ≥6.10Chip TDP · Accelerator only · floor | 335 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 12-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 29 | Hailo-10H (M.2/PCIe, x86 host) | 331 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-09-03 (page last edited) | Hailo Model Zoo public models (Hailo-10H)hailo_model_zoo v5.4.0, DFC v5.4.0 | |
| 30 | Intel Core i5-1335U | ≥4.21Chip TDP · Accelerator only · floor | 231 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 12-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 31 | Intel Core i7-1185GRE | ≥7.97Chip TDP · Accelerator only · floor | 223 FPS | 28.00 WAccelerator only, host not includedChip TDP · max of range 12-28 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 32 | Terasic DE10-Agilex Development Board (Agilex 7, FPGA AI Suite) | 220 FPS | — not published | resnet-50-tfnot stated in table (set by .arch file) · batch 1 | D 2026-08-10 (published) | Altera FPGA AI Suite Handbook (performance tables)FPGA AI Suite 2026.1.1 (doc 863373) | |
| 33 | AMD Zynq UltraScale+ ZCU102 (ZU9EG) | 214 FPS | — not published | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| 34 | Intel Core i5-1235U | ≥3.64Chip TDP · Accelerator only · floor | 200 FPS | 55.00 WAccelerator only, host not includedChip TDP · max of range 15-55 W | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 35 | Intel Celeron 6305E | 13.26Chip TDP · Accelerator only | 199 FPS | 15.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 36 | AMD Zynq UltraScale+ ZCU104 (ZU7EV) | 183 FPS | — not published | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo | |
| 37 | Hailo-8L (M.2/PCIe, x86 host) | 138 FPS | — not published | resnet_v1_50INT8 (Hailo quantized) · batch 1 | D 2026-10-15 (page last edited) | Hailo Model Zoo public models (Hailo-8L)hailo_model_zoo v2.20.0, DFC v2.20.0 | |
| 38 | Intel Atom x7425E | 10.48Chip TDP · Accelerator only | 126 FPS | 12.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 39 | Rockchip RK3588 (NPU, single core of 3) | 113 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 40 | Intel N100 | 18.39Chip TDP · Accelerator only | 110 FPS | 6.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 41 | AMD Kria KV260 Vision AI Starter Kit (K26 SOM) | 6.77Module TDP · Module | 102 FPS | 15.00 WModuleModule TDP | tf_resnetv1_50_imagenet_224_224_6.97GINT8 · batch unknown | D 2023-01-13 (published) | Vitis AI 3.0 Model Zoo performance tableVitis AI v3.0 model zoo |
| 42 | Rockchip RK3576 (NPU, single core of 2) | 96.00 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 43 | Rockchip RV1126B (NPU) | 78.20 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 44 | Altera Agilex 5 FPGA E-Series 065B Modular Development Kit (FPGA AI Suite, OFS PCIe) | 64.00 FPS | — not published | resnet-50-tfnot stated in table (set by .arch file) · batch 1 | D 2026-08-10 (published) | Altera FPGA AI Suite Handbook (performance tables)FPGA AI Suite 2026.1.1 (doc 863373) | |
| 45 | Intel Atom X6425E | 4.91Chip TDP · Accelerator only | 58.96 FPS | 12.00 WAccelerator only, host not includedChip TDP | resnet-50INT8 · batch unknown | D 2026-10-01 (published) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| 46 | Rockchip RK3562 (NPU) | 55.00 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 47 | Rockchip RK3566/RK3568 (NPU) | 38.60 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 48 | Rockchip RK1808 (NPU) | 37.10 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 49 | Rockchip RV1126 (NPU) | 36.20 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) | |
| 50 | Rockchip RV1109 (NPU) | 24.40 FPS | — not published | resnet50-v2-7INT8 · batch 1 | C 2026-10-13 (page last edited) | Rockchip rknn_model_zoo README benchmark tablerknn_model_zoo v2.3.2 (README @ 837db8e) |
Excluded from ranking (15)
Partial or suspected partial models, fps/W outliers and rows HWBench Data flags (rank_note) are kept out of the rankings. Rows with a published power figure that Data does not mark for the default fps/W ranking still appear under By speed. Every row stays in /find.json.
| Platform | Model | Result | Power | Left out of | Why | Source |
|---|---|---|---|---|---|---|
| Hailo-8 AI processor (chip, HNC18BI11BH) | ResNet-50 | 1371 FPS | 3.60 WAccelerator only, host not includedVendor-measured (typical/avg) | Efficiency ranking | fps/W outlier: fps/W 380.8 is 65.7x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured-by-vendor, accelerator-only) | Vendor page · Hailo-8 Datasheet Revision 1.9.2 (April 2026)Hailo-8 Datasheet Revision 1.9.2 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| Hailo-8 M.2 module | ResNet-50 | 1371 FPS | 4.00 WModule Measured | Efficiency ranking | fps/W outlier: fps/W 342.8 is 59.2x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured, module) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p03-ina228), 2026-05-05 |
| Hailo-8 M.2 Key M ET module (HM218B1C2HAE / HM218B1C2FAE) | ResNet-50 | 915 FPS | 3.30 WModuleVendor-measured (typical/avg) | Efficiency ranking | fps/W outlier: fps/W 277.3 is 47.9x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured-by-vendor, module) | Vendor page · Datasheet (mirror)Hailo-8 M.2 Key M ET Module Data Sheet Rev. 2.2; Table 3 - Power Consumption for typical configurations (typical at 25 C) |
| Axelera Metis M.2 card | ResNet-50 v1.5 | 2050 FPS | 7.60 WModule Measured | Efficiency ranking | fps/W outlier: fps/W 269.7 is 46.6x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured, module) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p04-axelera-metis), 2026-05-07 |
| Hailo-8L AI processor (chip, HNC1LBI11BH) | ResNet-50 | 500 FPS | 1.90 WAccelerator only, host not includedVendor-measured (typical/avg) | Efficiency ranking | fps/W outlier: fps/W 263.2 is 45.4x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured-by-vendor, accelerator-only) | Vendor page · Hailo-8L Datasheet Revision 1.9.3 (April 2026)Hailo-8L Datasheet Revision 1.9.3 (April 2026); Table 12. Power consumption (average power, 25 C, VDD 0.8 V, normal power mode) |
| Hailo-8L M.2 Key B+M ET module | ResNet-50 | 433 FPS | 1.90 WModuleVendor-measured (typical/avg) | Efficiency ranking | fps/W outlier: fps/W 227.9 is 39.3x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured-by-vendor, module) | Vendor page · Datasheet (mirror)Hailo-8L M.2 Key B+M ET Module Data Sheet Rev. 4.0; Table 3 - Power consumption for typical configurations (typical at 25 C) |
| DEEPX DX-M1 M.2 LPDDR5x2 module | ResNet-50 | 1009 FPS | 4.70 WModule Measured | Efficiency ranking | fps/W outlier: fps/W 214.7 is 37.1x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured, module) | Mario Bergeron, Edge AI Power BenchmarkingMario Bergeron, Edge AI Power Benchmarking (edge-ai-power-p05-deepx-m1), 2026-06-10 |
| Axelera Metis Embedded 113m M.2 card | ResNet-50 v1.5 | 1849 FPS | 9.00 WModuleModule TDP · max of range 3.5-9 W | Efficiency ranking | fps/W outlier: fps/W 205.4 is 35.5x the median 5.8 of 109 powered rows for model_key "resnet-50" (module-TDP-spec, module) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| NVIDIA Jetson AGX Orin (MaxQ) | MLPerf ResNet | 3526 FPS | 23.57 WWhole system Measured | Efficiency ranking | fps/W outlier: fps/W 149.6 is 25.8x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured, whole-system) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| Axelera Metis Edge 130p PCIe card | ResNet-50 v1.5 | 1825 FPS | 15.00 WModuleBoard TDP · max of range 8-15 W | Efficiency ranking | fps/W outlier: fps/W 121.7 is 21.0x the median 5.8 of 109 powered rows for model_key "resnet-50" (board-TDP-spec, module) | Axelera Voyager SDK model zoo (docs/reference/models/model-zoo.md)Voyager SDK 1.9.0 (model-zoo.md @ f184b09) |
| MemryX MX3 M.2 module | ResNet-50 | 1148 FPS | 10.00 WModuleModule TDP | Efficiency ranking | fps/W outlier: fps/W 114.8 is 19.8x the median 5.8 of 109 powered rows for model_key "resnet-50" (module-TDP-spec, module) | MemryX Model Explorer (SDK 2.2)MemryX SDK 2.2 model explorer |
| NVIDIA Jetson Orin NX 16GB (MaxQ) | MLPerf ResNet | 1682 FPS | 14.95 WWhole system Measured | Efficiency ranking | fps/W outlier: fps/W 112.5 is 19.4x the median 5.8 of 109 powered rows for model_key "resnet-50" (measured, whole-system) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| NVIDIA Jetson Orin NX 16GB | MLPerf ResNet | 2641 FPS | 40.00 WModuleModule TDP · max of range 10-40 W | Efficiency ranking | fps/W outlier: fps/W 66.0 is 11.4x the median 5.8 of 109 powered rows for model_key "resnet-50" (module-TDP-spec, module) | NVIDIA Jetson Benchmarks (MLPerf v3.1 results table)MLPerf Inference v3.1 (JetPack 5.1.1) |
| Intel Core Ultra 9 288V | resnet-50 | 1981 FPS | 37.00 WAccelerator only, host not includedChip TDP · max of range 17-37 W | Efficiency ranking | fps/W outlier: fps/W 53.5 is 9.2x the median 5.8 of 109 powered rows for model_key "resnet-50" (chip-TDP-spec, accelerator-only) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |
| Intel N100 | resnet-50 | 110 FPS | 6.00 WAccelerator only, host not includedChip TDP | Efficiency ranking | fps/W outlier: fps/W 18.4 is 3.2x the median 5.8 of 109 powered rows for model_key "resnet-50" (chip-TDP-spec, accelerator-only) | OpenVINO official performance benchmarks (graph-data-ov.json)OpenVINO 2026.4.1 |