AI is now capable of developing its own inference hardware

Automated excerpt

With the logits streamed back while the card runs (tools/decode_profile. py, 96 tokens), 4-bit LFM2-2. 6B: 11. 07 / 11. 02 (build B); SmolLM3-3B: 8. 92 / 8. 89 (build B); Phi-4-mini: 6. 69 / 6. 67 (build B). unit and a 120. 755 MHz clock. For LFM2 and Qwen3 the card runs one decode program compiled the logits stream back while the card is still running: the host adds 0. 17 to 0. 30 ms per token on omarchy (0. 45 to 1. 3 ms on opentpu). Decode is bound by DRAM, so a faster clock mostly helps prefill.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.