When float_diff is enabled, the pool can now restrict float precision to sub-1 difficulty values only, sending integer difficulty for values >= 1. This is controlled by the new float_diff_below_one setting (default: true).
This resolves a firmware compatibility issue discovered on Canaan Nano3S (and potentially other ASIC devices) where float difficulty at high magnitudes (100K+) caused CRC/COM_CRC errors on the chip data bus. Embedded miner firmware typically uses float32 internally, which only supports ~7 significant digits. A difficulty like 103297.0000 exceeds float32 precision, causing the firmware to misinterpret the value and produce hardware errors.
With float_diff_below_one enabled:
- Difficulty < 1 (e.g.,
0.0038,0.022) — sent as float with configured precision. Ideal for tiny miners like NerdMiner and ESP32 devices. - Difficulty >= 1 (e.g.,
4096,103297) — rounded to integer. Safe for all ASIC firmware including Canaan and AxeOS devices that truncate or can't process float difficulty.
To use the previous behavior (float at all magnitudes), set "float_diff_below_one": false in your vardiff config.