Performance Data for Consumer Edge Use Cases
Loss Prevention Retail Benchmark Results
Find the latest performance data by choice of Intel® processors for the vision-enabled workloads that drive loss prevention.
Detection and Classification
Pipeline: Non scans, Hidden items, Items in basket, Multi product scans, Product switching, Sweethearting.
Mode: Power save
| Platform Name | SKU | Target XPU1 | Model | Video Type/Resolution | Precision | Cumulative Throughput2 | Use Case Density3 | Stream Density4 | Latency (ms)5 | Batch Size | Average Package Power (Watts) |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Raptor Lake Refresh(Gen 14) | i5-14500 | GPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
133.99 | 3 shopping lanes | 18 | NA | 8 | 52.19 |
| Raptor Lake Refresh(Gen 14) | i5-14500 | CPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
187.98 | 6 shopping lanes | 36 | NA | 1 | 59.55 |
| Meteor Lake (Series 1) | Ultra 5 125HL | GPU:NPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
279.34 | 7 shopping lanes | 42 | NA | 8 | 26.5 |
| Meteor Lake (Series 1) | Ultra 5 125HL | GPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
194.11 | 5 shopping lanes | 30 | NA | 8 | 23.91 |
| Meteor Lake (Series 1) | Ultra 5 125HL | CPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
158.75 | 5 shopping lanes | 30 | NA | 1 | 42.82 |
| Arrow Lake (Series 2) | Ultra 5 235H | GPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
554.92 | 6 shopping lanes | 36 | 345.603 | 1 | 36.7843 |
| Arrow Lake (Series 2) | Ultra 5 235H | GPU:NPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
664.7 | 8 shopping lanes | 48 | 374.225 | gpu:1,npu:1 | 43.1124 |
| Panther Lake (Series 3) | Ultra X7 358H | GPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
914.26 | 10 shopping lanes | 60 | 313.295 | 1 | 32.8929 |
| Panther Lake (Series 3) | Ultra X7 358H | GPU:NPU | detection - yolov11n classification - efficientnet-b0 face-detection-retail-0004 face-reidentification-retail-0095 |
RTSP Camera (AVC HD @15FPS) |
INT8 INT8 FP16 FP16 |
927.8 | 12 shopping lanes | 72 | 420.951 | gpu:1,npu:1 | 40.3939 |
You can run the performance benchmarks using existing models or your own models.
Hardware and software configuration (measured Dec 19, 2025):
Hardware configuration for 1-node, Intel Corporation Raptor Lake Client Platform, 1x Intel(R) Core(TM) i5-14500, 14 cores, 65W TDP, HT ?, Turbo On, Total Memory 64GB (2x32GB DDR5 4800MT/s [4800MT/s]), BIOS Intel(R) Core(TM) i5-14500, microcode 0x3a, 1x , 1x Ethernet Connection, 1x 114.6G SanDisk 3.2Gen1, 1x 931.5G Samsung SSD 970 EVO Plus 1TB, Ubuntu 24.04 LTS, 6.14.0-35-generic. Test by Intel as of Thu Nov 13 02:31:08 PM IST 2025.
Hardware configuration for 1-node, Intel Corporation Meteor Lake Client Platform, 1x Intel(R) Core(TM) Ultra 5 125HL, 14 cores, 45W TDP, HT On, Turbo On, Total Memory 64GB (2x32GB DDR5 5600MT/s [5600MT/s]), BIOS Intel(R) Core(TM) Ultra 5 125HL, microcode 0x24, 2x Unknown NIC, 1x AX88179 Gigabit Ethernet, 1x 232.9G WD_BLACK SN770 250GB, Ubuntu 24.04.3 LTS, 6.14.0-27-generic
Hardware and software configuration (measured Jun 26, 2026):
Hardware configuration for 1-node, Intel Corporation Arrow Lake Client Platform, 1x Intel(R) Core(TM) Ultra 5 235H, 14 cores, 200W TDP, HT Off, Turbo On, Total Memory 64GB (2x32GB DDR5 6400MT/s [6400MT/s]), BIOS Intel(R) Core(TM) Ultra 5 235H, microcode 0x118, 2x Unknown NIC, 1x 114.6G SanDisk 3.2Gen1, 1x 476.9G INTEL SSDPEKNU512GZ, Ubuntu 24.04.3 LTS, 6.14.0-27-generic.
Hardware configuration for 1-node, Intel Corporation Panther Lake Client Platform, 1x Intel(R) Core(TM) Ultra X7 358H, 16 cores, 200W TDP, HT Off, Turbo On, Total Memory 64GB (2x32GB DDR5 6400MT/s [6400MT/s]), BIOS Intel(R) Core(TM) Ultra X7 358H, microcode 0x11b, 1x Ethernet Connection, 1x AX88179 Gigabit Ethernet, 1x 465.8G Sabrent SB-RKT4P-500, Ubuntu 24.04.4 LTS, 6.18-intel.
- Wall power refers to platform power consumption.
- Accuracy (if listed) was validated with the specified dataset.
- efficientnet-b0 uses the pre-quantized model from the DLStreamer Pipeline Zoo, distributed as FP16-INT8.
1 Target XPU. The compute engine or engines the pipeline was assigned to. GPU:NPU means the pipeline was split across both, with different stages running on different engines.
2 Cumulative Throughput. Total frames per second across all streams at the reported density.
3 Use Case Density. The number of complete Loss Prevention deployments the platform sustains while meeting the target frame rate. Each deployment covers one shopping lane and runs six camera streams.
4 Stream Density. Total camera streams, being Use Case Density multiplied by six cameras per deployment.
5 Latency (ms). The median time one frame takes to pass through the analytics pipeline, from the video source, through decode and inference, to the result at the end of the pipeline. Measured with the Intel DL Streamer latency tracer. Figures are measured at the use-case density shown on the same row, so a higher figure on a row with more shopping lanes comes from the extra load and not from a slower platform. NA means latency was not captured for that run.