Workloads
Core edge workloads
Used for the home page, answers and edge rankings.
| Workload | Published rows | Platforms | Unit |
|---|---|---|---|
| object detection | 1627 | 81 | FPS |
| instance segmentation | 181 | 20 | FPS |
| pose estimation | 192 | 23 | FPS |
| image classification | 1771 | 76 | FPS |
| semantic segmentation | 264 | 35 | FPS |
| speech-to-text | 121 | 62 | FPS |
| LLM generation | 2993 | 106 | tok/s |
More edge workloads
Published edge results in further workloads. Each has its own like-for-like tables.
| Workload | Published rows | Platforms | Unit |
|---|---|---|---|
| LLM prefill | 41 | 22 | tok/s |
| vision-language models (VLM) | 78 | 22 | tok/s |
| image/video generation | 70 | 26 | FPS |
| language models (NLP) | 141 | 33 | FPS |
| embeddings (CLIP etc.) | 15 | 5 | FPS |
| audio classification | 11 | 6 | FPS |
| speech enhancement | 4 | 1 | FPS |
| text-to-speech | 5 | 5 | RTF (lower is better) |
| depth estimation | 73 | 12 | FPS |
| super-resolution | 100 | 8 | FPS |
| face analysis | 94 | 11 | FPS |
| hand tracking | 22 | 1 | FPS |
| text recognition (OCR) | 27 | 11 | FPS |
| oriented object detection | 31 | 13 | FPS |
| keypoint / feature detection | 16 | 6 | FPS |
| 3D point clouds | 12 | 1 | FPS |
| object tracking | 6 | 1 | FPS |
| re-identification | 6 | 3 | FPS |
| medical segmentation | 6 | 4 | FPS |
| recommendation | 14 | 7 | FPS |
| video action recognition | 7 | 1 | FPS |
| graph neural network | 7 | 5 | FPS |
Tiny (MLPerf Tiny, microcontrollers)
Microcontroller-class results from MLPerf Tiny, with measured energy per inference. Ranked only against each other.
| Workload | Published rows | Platforms | Unit |
|---|---|---|---|
| tiny image classification | 39 | 19 | inferences/s |
| visual wake words | 35 | 17 | inferences/s |
| anomaly detection | 34 | 16 | inferences/s |
| keyword spotting | 33 | 17 | inferences/s |
Datacenter (not ranked against edge)
MLPerf Inference datacenter results. Kept out of edge rankings.
| Workload | Published rows | Platforms | Unit |
|---|---|---|---|
| LLM serving (MLPerf datacenter) | 605 | 118 | tok/s |
How to read these numbers
Every number is labelled. Confidence: A MLPerf closed, audited · B documented third-party method · C community / review · D vendor-published, not independently verified · E estimate. Power basis: measured with the result · spec vendor spec/envelope · power n/a not published.
Machine-readable: /find.json (see llms.txt); full tables are in “Show data”. Confidence grades: A audited MLPerf closed · B documented third-party · C community · D vendor-published · E estimate. Unknown values are shown as unknown, never estimated. Generated 2026-10-16 from public sources.