$250 FPGA board hits 12,000 tokens/s inference

Lamb Labs demonstrated a proof-of-concept achieving 12,000 tokens/s inference on a $250 AMD KV260 FPGA board. While the small model cannot yet answer questions coherently, it showcases the potential of ultra-fast local inference on FPGAs.

2026-08-17 ~ 2026-08-18 · 3 related posts

• 12,000 tok/s inference achieved on a $250 FPGA board — ycombinator · 2026-08-17
• $250 FPGA Board Achieves 12,000 Tokens/s Generation — generativist · 2026-08-18

1 near-duplicate retellings: ycombinator