[Video idea for Jeff] Comparing old/new "AI-accelerator" M.2 cards for Edge AI: object detection, video encodning, LLMs, and AI image generation?

[Hedda]

@geerlingguy Would probably make an interesting video comparing old and new M.2 cards for Edge AI; object detection, video transcoding, LLMs, and AI image generation?

I would very much love to see a comparision video for some different LLMs and image generation models on a standard Raspberry Pi 5.

Like example:

• M5Stack LLM-8850 based on based Axera AX8850 SoC

vs.

• Radxa AICore AX-M1" M.2 2280 AI Acceleration Module based Axera AX8850 SoC

vs.

• Waveshare Hailo-8 M.2 AI Accelerator (Hailo-8) (and/or the Hailo-8 M.2 AI Accelerator reference hardware)

vs.

• Geniatech AIM M2 AI Accelerator Module M.2 Form Factor based on Kinara Ara-2

vs.

• MemryX MX3 AI accelerator M.2 Module (based on 4 MemryX MX3 chips without onboard RAM)

vs.

• Axelera AI M.2 AI Inference Acceleration card (a.k.a. Metis M.2 Card)

vs.

• Raspberry Pi AI Kit (Hailo-8L) (and/or the Hailo-8L M.2 AI Accelerator reference hardware)

vs.

• Google Coral TPU M.2 Accelerator A+E Key

Note! It might need to also compare heat generated and power since some of them will have an unfair advantage with active cooling.

PS: Not sure if want to include the Raspberry Pi AI HAT+ (Hailo-8) as it is not an M.2 card, but it does directly compete with the above.