FreeForGem is the multi-coin successor of FreeForGemPear (the Pearl miner this thread started with): one binary
freeforgem for three coins, own GPU kernels, a CPU check of every share before it is sent,
1% dev fee and the optional per-GPU automatic undervolt
--auto-uv. Saved
--auto-uv results of FreeForGemPear are imported on the first start. FreeForGemPear continues as the Pearl-only miner in
its own repository.
Coins| Coin | HiveOS hash algorithm | GPUs |
| Pearl (PRL) | pearlhash | NVIDIA RTX 20/30/40/50, GTX 16, CMP; AMD RX 6000+ (Windows) |
| NOID (Parano1d) | parano1d | NVIDIA RTX 30/40/50 (sm_80+), driver 580+ |
| Common Foundry (CMFD) | forgematrix_v4 | NVIDIA RTX 20 or newer, AMD Radeon (Windows); ~7 GB free GPU memory |
Linux x86_64 (Ubuntu 20.04+), HiveOS (custom miner
freeforgem) and Windows 10/11 x64.
What's new in 1.7.6 (2026-11-01) · video tutorial:
English ·
Pyccкий- One shared auto-undervolt climb for Pearl, NOID and CMFD and one shared profile store (state files keep the 1.7.5 format, so profiles carry over). The search climbs in 30 MHz steps and refines with 15 MHz steps only inside the first failed interval: the same edge in about half the steps (--uv-coarse 0 = the old single step size). A driver error on a coarse step is not refined again (no second crash).
- Better start points: RTX 40 starts at +200 and RTX 50 at +300 for NOID and CMFD too; after a crash the search stays 30 MHz below the crashed offset in all three algorithms.
- NOID and CMFD get what Pearl had: periodic re-validation, a step back on a burst of rejected shares, detection of a tuned offset left on the card by an interrupted run.
- Windows idle mode (--idle) for NOID and CMFD (Pearl only before). CMFD has no light load, so its engine pauses.
- New --temp-limit C / --temp-stop C2(Pearl, NOID, NVIDIA) for laptops and mobile GPUs: the GPU works only the share of the time that keeps it below C.
- CMFD no longer applies the HiveOS profile's memory lock (for example Pearl's 810 MHz) — it made CMFD several times slower on one rig. An explicit --lock-mclock still applies.
- Measured: NOID on 8x RTX 3070 — tuning of 8 cards about 6.5 min (was about 11 min in 1.7.5), 336.7 MH/s at 707 W (0.476 MH/W), 297/297 shares accepted. Pearl on 8x RTX 3070 (HiveOS) — 646.7 TH/s at 1168 W (0.554 TH/W), 155/155 shares, 0 rejects. CMFD on RTX 3070: +8.6...+10.3 % FW/kWh after tuning.
Chat:
t.me/freeforgem_chat (Russian-speaking, English welcome) · channel:
t.me/freeforgemWhat's new in 1.7.4 (2026-10-31)
- Kryptex username as the wallet (Pearl): mine to your Kryptex account username instead of a prl1... address and Kryptex pays in BTC/USDT with auto-conversion. The miner logs in as username/WORKER (Kryptex's format).
- Windows: the start file accepts a username on a Kryptex pool — put it into WALLET of a start-kryptex-*.bat (1.7.3 and older stopped with "WALLET does not look like a PRL address"). HiveOS: the username as the wallet, template %WAL%.%WORKER_NAME% as usual.
What's new in 1.7.3 (2026-10-28)
- Pearl on low-RAM rigs: far less RAM — the miner 925 → 205 MB, an undervolt check 2.4 → 1.7 GB; proofs share the RAM (at most 4 at once, one at a time when RAM is low). A 4 GB rig with 5 GPUs no longer swaps, and the OOM killer takes a check, not the miner.
- NACRE (NACR) merge mining on HimPool and AlphaPool: wallet prl1...+nacr1p... (HimPool: also +NOCK_ADDRESS).
- AlphaPool: login fixed (1.7.2 kept reconnecting with no jobs).
- HiveOS: without --lock-cclock / --lock-mclock the OC profile's locks are set before mining; GPUs start mining one by one, 1.5 s apart.
- Pearl on HiveOS, --auto-uv, lock 1805/810: 8× RTX 3070 — 647.8 TH/s, 1179 W; 4× RTX 3070 — 326.2 TH/s, 547 W.
What's new in 1.7.2- Undervolt profiles per core clock lock (Pearl, NOID, CMFD): switching the lock reuses that lock's verified offset, a new lock is tuned once. --uv-reset forgets the current lock, new --uv-reset-all forgets all.
- Self-heal after a crash (Xid): the miner names the failing GPU, steps its offset down 30 MHz and restarts itself (at most 12 times an hour) — no tuning from scratch, no restart loop. An offset that failed a check is never applied.
- GPUs that are not tuning keep mining; progress per GPU, coloured log. Tuning on RTX 3070 starts near the edge: ~11 min instead of 20 for an 8× RTX 3070 rig.
- NOID: load average 2–3 instead of 24+ on an 8-GPU rig (HiveOS reboots at LA > 28); stale shares dropped before sending, no stale-share flood at start.
- Leftover offsets of another miner are reset to the HiveOS OC profile; HiveOS shows the rig TH/W (Pearl) or MH/W (NOID) next to the version.
- Fixes the known issue of 1.7.1-beta.
NOID on 8× RTX 3070 (HiveOS, memory 810 MHz,
--auto-uv, rig totals)
| Mode | Core lock, MHz | MH/s | W | MH/W |
| eff (recommended) | 1300 | 257.0 | 523 | 0.491 |
| eff | 1400 | 277.6 | 606 | 0.458 |
| speed (recommended) | 1500 | 336.0 | 710 | 0.473 |
| speed | 1750 | 393.0 | 1014 | 0.388 |
CMFD on the same rig:
--auto-uv found +240..+315 per card, +9..+12% efficiency, ~184 FW/s at 1720 W. Single cards: RTX 3070 ~20 FW/s, RTX 4070 Ti ~35 FW/s.
Pearl: RTX 3070 @1560 / 810 ~70.5 TH/s at ~107 W (~86 W with
--auto-uv); RTX 4070 Ti @2445 / 810 144.7 TH/s at ~169 W; RTX 2060 SUPER @1605 / 5001 49.9 TH/s; AMD RX 6800 XT 34.4 TH/s stock, ~30–32 TH/s at 160–172 W with
--auto-uv (Windows).
Why --auto-uvNear the stability limit a GPU does not crash and reports no Xid — it starts returning silently wrong results, which the pool never shows you as rejects.
--auto-uv (optional, needs root / Administrator) finds the deepest safe V/F offset of every GPU with bit-exact checks under full load, keeps one step of margin, re-checks while mining and restores your settings on exit. RTX 3070 on NOID eff @1400: 87.5 W → 72.8 W at the same hashrate (+20% MH/W).
Quick start — HiveOS (new flight sheet, Miner: Custom → Setup Miner Config)
Miner name : freeforgem
Installation URL : https://github.com/pepsykolya/FreeForGem/releases/download/v1.7.6/freeforgem-1.7.6.tar.gz
Wallet template : %WAL%.%WORKER_NAME%
Pearl : hash algorithm pearlhash pool pearl-eu2.luckypool.io:3360 extra: --auto-uv
NOID : hash algorithm parano1d pool stratum+ssl://eu2.innovlab.cc:19601 extra: --algo noid --noid-mode eff --lock-cclock 1300 --lock-mclock 810 --auto-uv
CMFD : hash algorithm forgematrix_v4 pool cmfd+tls://159.69.194.46:29445?pin=9dfb51083f287726117f689f87bc7a878792efcca58ca8b6f6e7c05ac5d152e9 extra: --auto-uv
HiveOS caches the archive by file name, so every version has a new file name — always paste the URL of the current release. CMFD needs its own OC profile without the memory lock. Coming from FreeForGemPear: create a new flight sheet.
Windows — unpack
freeforgem-1.7.6-windows-x64.zip to a folder without spaces, open the start file for your coin and pool (
start-luckypool-*.bat,
start-kryptex-*.bat,
start-noid.bat,
start-cmfd.bat), put your wallet into
WALLET, run (as Administrator for
--auto-uv). Only the GPU driver is needed.
Linuxtar xzf freeforgem-1.7.6-linux-x86_64.tar.gz && cd freeforgem-1.7.6
./freeforgem --pool pearl-eu2.luckypool.io:3360 --wallet YOUR_PRL_ADDRESS --worker rig1 --auto-uv
./freeforgem --algo noid --pool stratum+ssl://eu2.innovlab.cc:19601 --wallet o1YOUR_ADDRESS --worker rig1 --noid-mode eff --lock-cclock 1300 --lock-mclock 810 --auto-uv
NOCK merge mining on LuckyPool: wallet
prl1...+YOUR_NOCK_ADDRESS, no hashrate loss.
Video tutorial for 1.7.6: English ·
Pyccкий (also on the release page:
EN mp4 ·
RU mp4, subtitles .srt)
Dev fee — 1%: Pearl / NOID — one 60-second slot every 100 minutes on your own pool under a separate login, shown in the log; if the fee login can't be served, the slot is skipped. CMFD — 36 s per hour per GPU inside the engine.
Trust: the host program (pool client, job and proof logic) is open source; the GPU kernels ship as binary libraries. Every share is checked on the CPU before it is sent (Pearl: official zk-pow verifier). SHA-256 of every file — SHA256SUMS on the release page; download only from GitHub releases.
1d282d61b88537090be2fd4e426ac99b8119bbb08a15ea50383cc7ac0a7916f6 freeforgem-1.7.6.tar.gz (HiveOS)
f0cebb0ae3898384b3f659d5faa3ea59989547191ea3f1bc038cf726c182fb60 freeforgem-1.7.6-linux-x86_64.tar.gz
ec7919dd663f7b224832488e8fadeedc222076983a02e56ae8b8749e133104f0 freeforgem-1.7.6-windows-x64.zip
Questions, logs and hashrate reports (especially RTX 40/50 on NOID, AMD on Pearl/CMFD) — post here. Disclosure: I'm the developer of FreeForGem.
