DeepX Tapes Out Second-Gen Edge AI Chip Running 100B Models Under 5 W

1 reports archived1 independent sourcesupdated 11/10/2026, 02:33:58

Synthesis & Latest Updates

1 Sources Cross-Validated

South Korean edge AI chip startup DeepX has taped out its second-generation accelerator, the DX-M2, using Samsung's custom processor-in-memory technology to run models of up to 100 billion parameters on-device at under 5 W, delivering 10 to 30 tokens per second. The key leap over the mass-produced, vision-focused DX-M1 is a tripling of memory bandwidth, targeting robots, drones and other generative AI edge devices.

LATEST/South Korean edge AI chip startup DeepX has taped out its second-generation accelerator, the DX-M2, using Samsung's custom processor-in-memory technology to run models of up to 100 billion parameters on-device at under 5 W, delivering 10 to 30 tokens per second. The key leap over the mass-produced, vision-focused DX-M1 is a tripling of memory bandwidth, targeting robots, drones and other generative AI edge devices.

Hourly heat curve

3 fully observed hours

Now

14.8

Peak

15.103:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Hacker News AIT2Nov 10, 02:33·78 pts
DeepX Tapes Out Second-Gen Edge AI Chip Running 100B Models Under 5 W
- DX-M2 supports 20B to 100B parameter models (roughly 3B to 20B active) at 10-30 tokens/s within a 5 W budget
- Memory bandwidth tripled versus the first-gen DX-M1, the key enabler for on-device LLM inference
- Built on Samsung custom processor-in-memory technology; the vision-focused DX-M1 is already in mass production
View Related Reports & Analysis Original Source