Edge AI chips compared on speed and power draw
Curated sources
Sources are organized around the requested evidence needs.
Edge AI inference speed, power, and energy benchmarks
Curated sources for comparing inference speed with power or energy. **BrainChip report:** includes latency and energy per inference for Akida, MCU-class devices, NVIDIA Jetson Nano, and Google Coral; downloadable raw results and a complete reproduction package are not confirmed, and some results were unverified.[[cite:1]][[cite:2]][[cite:3]][[cite:4]] **MLPerf Inference Edge:** pairs benchmark performance with whole-system average AC power for submitted edge systems; downloadable result files and reproduction materials are not confirmed here.[[cite:5]][[cite:6]][[cite:7]] **Early Experience in Benchmarking Edge AI Processors:** compares Edge TPU, NVIDIA Xavier, NovuMind NovuTensor, and GTX 1080 Ti using latency and images-per-second-per-watt; methodology and benchmark tables are published, but downloadable raw data, code, and reusable artifacts are not confirmed.[[cite:8]][[cite:9]][[cite:10]] **Edge Devices Inference Performance Comparison:** reports inference time and frames per second for Jetson Nano, Intel Neural Stick, and Google Coral devices, with a downloadable CSV and substantially described methodology, but no paired measured power or energy results.[[cite:11]][[cite:12]][[cite:13]][[cite:14]] **TI TDA4 application note:** an official technical source on performance and power benchmarking for the TDA4 platform; it is useful for methodology, but the supplied evidence does not confirm a downloadable cross-processor dataset.[[cite:15]]
https://brainchip.com/wp-content/uploads/2023/01/BrainChip_Benchmarking-Edge-AI-Inference-1.pdf
Créez votre compte pour conserver cette réponse et la reprendre plus tard.
Examinons les alternatives :
- Modifier la requête.
- Démarrer une nouvelle conversation.
- Supprimer des sources (si elles ont été ajoutées manuellement).