Edge AI Benchmarks
github.com/chorylee/edge-ai-benchmarks- Category
- Developer Tools
- Rank
- No. 3930Tools index
Previous survey · No. 3705 ·
- Type
- TOOL
- Use case
- Model & Agent Evaluation
- Date
About
Edge AI Benchmarks is a GitHub repository compiling vendor-reported FPS-per-dollar figures for YOLO object detection across low-cost edge AI boards such as Orange Pi 5 Max, Hailo-8 M.2 modules, and Jetson Orin Nano, priced roughly $75 to $215. The maintainer notes the FPS numbers come from vendor datasheets rather than independent lab tests and flags that TOPS figures across NPUs (sparse vs. dense INT8) are not directly comparable.
At a glance
Edge AI Benchmarks is a small reference table for object-detection hardware, not a runnable benchmarking suite. Its README and CSV place vendor-reported video-processing rates beside September 2026 price assumptions. It can help form questions about a board or accelerator, but the rows do not establish a fair performance ranking.
Good fit
Use it as a quick introduction to why a headline operations-per-second figure is a poor buying shortcut. Its strongest contribution is the caveat beside the numbers: different object-detection models and different definitions of hardware throughput make apparently comparable figures misleading.
Consider something else when
You need independent measurements, current quotations or proof that one device will run your camera workload faster. The CSV compares YOLOv5s with the larger YOLOv5m and leaves the Jetson frame-rate cell empty. A missing result cannot support a ranking.
Start here
Open benchmarks.csv beside the README. Check model, frame rate, price note and source note together before comparing rows. Then locate the named vendor’s original measurement for your model and input resolution. The CSV names its vendor sources but does not supply row-level source URLs or a complete test setup.
Cost and requirements
Reading the repository or CSV needs only a browser or spreadsheet. No board purchase or software installation is needed to inspect it. The README offers the data under CC BY 4.0 with attribution to RoboKrunch. Hardware prices in the table are dated assumptions, not current offers.
Limits to know
Vendor-reported results have not been independently reproduced by this project. Some advertised throughput counts sparse operations and some dense operations, so dividing those headline numbers does not create a common scale. The table also mixes a board with an accelerator module. Verify the full host, power and software requirements before budgeting a system.
Sources checked 2026-10-30. Source-based assessment.
Why it made the leaderboard
It gives a comparative FPS-per-dollar table across cheap edge AI boards, sourced to vendor datasheets, useful groundwork for anyone choosing hardware to run vision models at the edge.
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.