Mujina mines on all three RDS hashboards for two hours
2026-09-28, 12:57:40Z to 14:57:41Z. Mujina, the open-source (GPL-3.0-or-later) Bitcoin mining software, held all three hashboards of an RDS unit, 300 BZM2 ASICs, mining to a public pool for a full two-hour hold, then stopped them safely. This matches the length of the stock software's own reference run on this unit.
At the start of the hold the chain tip was block 968,999
(000000000000000000000a20227e54154bb16bc17858da76ca8857eb6fee12e5), so the hold
began mining toward block 969,000. At its end the tip was block 969,011
(0000000000000000000050ab3c26e1b6e127364362095f09dbbb1dbe57aa52c8). The hold
therefore covered blocks 969,000 to 969,011, twelve new blocks, and Mujina
moved to fresh work at each one. Two independent block explorers, mempool.space
and blockstream.info, agree on both tips' heights and hashes.
What ran
- Hardware: one RDS unit, three hashboards of 100 BZM2 ASICs each on their own 9-bit, 5 Mbaud chains, sharing one supply rail.
- Start-up: the unit's stock software powered and calibrated the boards. Mujina then took all three chains over without dropping power, and began dispatching pool work to each in turn.
- Pool: OCEAN, Stratum V1, share difficulty 262,144, under a worker named
mujina.
What it measured
Over the 7,200 s hold:
| shares accepted / rejected | 546 / 1 |
| the one rejection | a share for a block that had just been found (stale-prevblk) |
| trips, arms, scrams | none |
| the stop | orderly: dispatch paused, load off, every board dark afterwards |
Every criterion written for this case before the run was met: case MJ-HASH-07,
attempt 4, run 20260928T124110Z.
Figures published by explicit decision (Ronald Bynoe, 2026-09-28): the share counts, the window and the block range. No hash rate, efficiency, power or temperature is published before the first multi-day run of at least 48 hours; those are in the project's case records, for the eventual retrospective.
What it took
Two earlier attempts were stopped by the test harness, not by Mujina. The harness's own long-running shell stalled on a blocked read in one attempt and exited unexpectedly in the other. Both times the safeguards brought the machine safely dark. The harness was then rebuilt so that a read cannot block a phase, the power-supply listener lasts as long as the window, a stop can always be run from outside, and the harness records how its shell ends. This run is the first two-hour hold on that harness.