1.9.1 UPD 11月3日 Windows, Linux

下载 ForgeMiner

Windows Linux

ForgeMiner 版本历史

共 10 个版本，最新的在最前。说明原样引自开发者发布的版本说明（英文）。

ForgeMiner 1.9.1 当前版本

· 发布页面

ForgeMiner v1.9.1

Fixes for what 1.9.0 users reported, Pearl faster on the RTX 50 series, NVIDIA voltage on every RTX 20–50 card, --mclk with any clock the card lists, a fan curve per card, and a long round of fixes to pools, shutdown and the overclock flags.

Fixes for 1.9.0 reports
  • Pearl on the RTX 30 / 40 series with a driver older than 580: the default kernel made the miner restart again and again. It now loads on those drivers, and falls back to --kernel 2 on its own where it cannot.
  • Pearl on the CMP 30HX mined at 0 H/s in 1.9.0; it now has a kernel of its own.
  • KawPow on the RX 470–590: from epoch 613 on, the DAG build could hang the card. Fixed.
  • --fan-curve stayed on its first point when the curve was written with spaces (40:30, 60:60). Fixed.
Faster
  • Pearl on the RTX 50 series: up to +2% hashrate on the same overclock.
  • Iron Fish on the RX 470–590: about +8%. Quantus on AMD: +0.3…1% on every series.
Overclock, voltage, fans
  • NVIDIA core voltage on Linux: --vddc and --vddoff now work on every RTX 20 / 30 / 40 / 50 card. A card a request does not fit is skipped and the other cards are still set; the reason is shown in the dashboard.
  • --mclk on the GTX 10 / 16, RTX 20 and CMP 30–50HX: any memory clock the card lists, not only 810. A clock the card does not list is refused with the list. If the driver accepts a memory clock but does not hold it, the miner says so instead of reporting success.
  • A fan curve per card: --fan-curve 45:30,70:90/40:40,70:80/*, one curve per card in --gpu order, * leaves the card to the driver.
  • --cclk above a card's highest clock now says what the driver will actually run. --oc-reset works per card on mixed rigs and with --gpu all.
  • AMD: fans go back to their own speed on exit.
Pools and stability
  • Two failed connections move to the next pool in the list, and the first pool is taken back when it returns.
  • Every way out — Ctrl+C, miner stop, an error — hands the cards back and ends within seconds.
  • Quantus: a card that hits a CUDA error restarts by itself while the other cards keep mining.
  • NOID dev fee: 2% (was 2.5%).
  • A pool that hangs up on every full-size Pearl share gets one clear warning to mine on another pool.
  • Windows rigs without an AMD card no longer load the AMD OpenCL runtime.
  • Linux: if the AMD cards are hidden from the user running the miner, it says to run as root instead of failing on the NVIDIA path.
  • HiveOS: a --gpu in a trailing comment no longer applies, and backup pools with a scheme in the URL work.
  • The RTX 50 series, and NOID on the RTX 30 / 40 series, need NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.9.1/ForgeMiner-1.9.1.tar.gz.
Checksums (sha256)
53c515b4160b5f51945573ed15ebb69c49de8e5798c24fb5dfdc3adb1739d867  ForgeMiner-1.9.1-windows.zip
10bb590103515f8dfbea0d0bb8b15f1b6ccf9a48219fee39f1d0bf8304396fd4  ForgeMiner-1.9.1-linux.tar.gz
e92d64938cf9b5d011f6b8b9c81a9ef3e82017ac22ef51b2a5fd42cf4f866188  ForgeMiner-1.9.1.tar.gz

ForgeMiner 1.9.0

· 发布页面

ForgeMiner v1.9.0

Pearl on the RTX 30 / 40 series now mines with about 7% less board power out of the box, for a fraction of a percent of hashrate. Iron Fish is a new coin on NVIDIA and AMD. And a long list of fixes and additions below: the GTX 16 series, memory at 810 MHz, Radeon RX 400 / 500, AMD speedups, voltage, BTX.

Pearl on the RTX 30 / 40 series: efficient by default
  • Every RTX 30 and 40 card gets a new default kernel: about 7% less board power at the same clocks, for a hashrate loss of under half a percent (RTX 3060, 3070 Ti and 4070 Ti measured on the same overclock). Nothing to configure.
  • Want the last few tenths of a percent anyway? --kernel 2 brings back the previous kernel, the highest hashrate; --kernel 1 is the default. On the RTX 20 and 50 series the flag changes nothing.
  • RTX 50 series: a little less power at the same hashrate. RTX 20 series unchanged.
New coin: Iron Fish (FishHash)
  • --algorithm iron on NVIDIA (GTX 10 and newer) and AMD (RX 400 and newer); needs a card with 6 GB or more. The 4.6 GB dataset is built on the card in seconds. Dev fee 0.70%, the same as KawPow.
  • Measured against the popular miners on the same overclock: within a few percent on the RTX 30 / 40 and RX 5000 / 6000 / 7000 series; the RX 9000 series is still being tuned.
New cards: AMD Radeon RX 400 / 500 series (Polaris)
  • Pearl, Quantus, NOID, QubitCoin, Xelis, Cryptix, KawPow (Ravencoin, Quai, Neurai) and Iron Fish (8 GB cards) now run on the RX 470 / 480 / 570 / 580 / 590 through the stock OpenCL driver, on Linux / HiveOS and Windows. Conflux needs more than 8 GB and is not available on these cards.
Fixes and improvements
  • GTX 16 series and CMP 30HX, Pearl about 6x faster on the GTX 1650 / 1660 family (TU116 / TU117): these cards have no tensor cores, and 1.9.0 mines them with the kernel written for the GTX 10 series instead.
  • GTX 16 series, Quantus no longer loses half of its hashrate at a low memory clock (the 810 MHz memory setting): the search table is chosen by what fits the card's L2 cache, so the card runs at full speed with the memory at 810 MHz as well as at stock.
  • AMD faster on the same overclock: Pearl on the RX 9060 XT about +18%. NOID on the RX 9000 series about +3%. Quantus on the RX 5000 / 6000 / 7000 / 9000 series +1…1.5%.
  • --mclk 810 on the GTX 10 / RTX 20 series: the memory clock now goes back on its own after a crash, at the next start of ForgeMiner (with or without the flag), and with --oc-reset.
  • NVIDIA core voltage on Linux (explicit flags, off by default): --vddc on the RTX 20 / 30 / 50 series, --vddoff on the RTX 40 series, --voltage-info --gpu to read what a card offers, --voltage-recover to put a card back after an interrupted run. Verified on drivers 580, 610 and 615. The clocks can come from HiveOS or from ForgeMiner's own --cclk / --coff / --mclk on the same card, whichever you use. Lowering the core voltage is the next step in efficiency on any coin: the same hashrate at fewer watts.
  • BTX refuses cleanly on a machine without NVIDIA cards instead of stopping with a driver error.
  • ForgeMiner's own --cclk / --coff / --mclk / --plimit / --fan under CUDA_VISIBLE_DEVICES went to the card with the same NVML number, not to the card being mined; the card is now matched by its PCI bus id. Runs with every card visible were never affected.
  • Windows: cards whose driver reports no power to NVML (nvidia-smi prints [N/A], some RTX 4060 boards) showed 0 W; the watts now come from the same per-rail sensor GPU-Z reads.
  • Dashboard: shares that were in flight when a route died with unacknowledged data (route dead) now count as lost, not delivered, and the Delivered line says how many. AMD clocks are shown when the card reports no level; NVIDIA offsets are shown on mixed rigs.
  • Help: NOID runs from the GTX 10 / RX 5000 series, the BTX example uses btx-pool.com, --fan-curve is NVIDIA-only.
  • The RTX 50 series, and NOID on the RTX 30 / 40 series, need NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.9.0/ForgeMiner-1.9.0.tar.gz.
Checksums (sha256)
fee310c016aa2cbf0d4973e914ea8f7bb35996e07c92c9b5128c4c130a39ee87  ForgeMiner-1.9.0-windows.zip
26fefe3584328a376485978aef1f288bef9617820157fb1416f58ef639f5926c  ForgeMiner-1.9.0-linux.tar.gz
f919504b57661d5c43bb5e26ffbdbd64245705c7efff55089e2352c1633bb257  ForgeMiner-1.9.0.tar.gz

ForgeMiner 1.8.9

· 发布页面

ForgeMiner v1.8.9

AMD Radeon support for every coin, NVIDIA and AMD cards together in one miner, and faster Quantus and NOID on NVIDIA.

AMD Radeon
  • Every coin on the RX 5000, 6000, 7000 and 9000 series (RDNA 1–4), on Linux / HiveOS and Windows: Pearl, Quantus, NOID, QubitCoin, Xelis, Cryptix, Conflux and KawPow (Ravencoin, Quai, Neurai). The command lines are the same as on NVIDIA; ForgeMiner picks the right code for each card by itself. The dashboard and the API show power, temperature, fan and clocks of the AMD cards on both systems.
  • Against the popular AMD miners, on the same overclock: NOID +16…+26%, QubitCoin +5…+7%, Pearl from +0.6% (RX 7000) to +44% (RX 5000), Quantus up to +5%; KawPow and Conflux on par.
  • Mixed rigs. NVIDIA and AMD cards in one rig mine in one ForgeMiner by default, no extra flag. The cards are numbered NVIDIA first (as before), then AMD in PCI bus order; --gpu and the overclock flags follow this order. FORGE_AMD=0 mines on the NVIDIA cards only, FORGE_AMD=1 on the AMD cards only. A rig with cards of one brand is numbered as before.
  • Overclock flags for AMD: the same --cclk, --coff, --mclk, --moff, --plimit and --fan as on NVIDIA, plus the core voltage: --vddoff (offset; RX 6000 / 7000 / 9000) and --vddc (absolute; RX 5000 / 6000). On Linux / HiveOS and on Windows; the previous settings are put back when ForgeMiner exits. On Windows --plimit is a percentage of the card's power limit, not watts.
  • Good to know: Conflux needs a card with more than 8 GB, the dataset no longer fits into 8 GB (the RX 5700 XT refuses it, as with other miners). Pearl on the RX 9060 XT runs only at the stock core voltage; a negative voltage offset makes it fail. The RX 9000 series needs a recent driver (on HiveOS, a current image). On plain Linux the AMD OpenCL runtime (ROCm) has to be installed; HiveOS has it. BTX stays NVIDIA-only.
NVIDIA
  • Quantus is faster on every series, against 1.8.8 on the same overclock: GTX 10 +3.2%, RTX 20 +3.7%, RTX 30 +2.2%, RTX 40 +2.2%, RTX 50 +2.0%.
  • NOID is faster on the RTX 30 / 40 / 50 series: RTX 50 about +11%, RTX 40 about +3% at about 5% less power, RTX 30 about +2.5% at about 5% less power. Shares are also checked on the processor many times faster, so a weak rig CPU is less loaded.
  • Pearl: on the RTX 40 and RTX 50 series the same hashrate at about 1–1.5% less power; RTX 20 and RTX 30 unchanged.
  • Memory clock 810 MHz on the GTX 10 and RTX 20 series (Linux / HiveOS and Windows). NVIDIA's management library cannot lock the memory clock before the RTX 30 series; on these cards --mclk 810 now holds the memory at 810 MHz, and the previous setting is put back when ForgeMiner exits or is stopped (after a crash, at the next start with --mclk 810). On the RTX 20 series, Quantus and NOID keep their hashrate at about 9% and 13% less board power; on the GTX 10 series about 7% less power. It is off unless you ask for it: add --mclk 810 (per card, in --gpu order). On Windows run the miner as administrator. Pearl on the RTX 20 series needs its memory clock, do not use 810 there.
Also new
  • --proxy — every pool connection through a proxy (requested in #5). socks5://[user:pass@]host:port is SOCKS5 with an optional login; the pool's host name is resolved by the proxy, so a blocked local DNS does not matter. http://[user:pass@]host:port is an HTTP proxy with CONNECT. Every connection goes through it: your pools, the failover and backup links and the dev fee. TLS pools work as usual. Also as the environment variable FORGE_PROXY. A proxy adds its own latency to every share.
  • Fix — the help on Windows can be scrolled again. _help.bat and forge --help in cmd printed into the dashboard's screen, which has no scrollback, and the mouse wheel then typed garbage into the window. The help and --version now print before the dashboard takes the console; forge help and forge /? work too.
  • The RTX 50 series, and NOID on the RTX 30 / 40 series, need NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.9/ForgeMiner-1.8.9.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes.
Checksums (sha256)
7d87d84961d0fa88b22307da8e6bb26b606ed679f0799d3ca3df3865c102a468  ForgeMiner-1.8.9-windows.zip
37333017437ce3c25d90cb5728d0df240a85b4d3611fe03f4a10470032859055  ForgeMiner-1.8.9-linux.tar.gz
e4e5fe1c0e3d54c1eb78a23db47b496cba96092789837e0831ae9a4978cb73cd  ForgeMiner-1.8.9.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

ForgeMiner 1.8.8

· 发布页面

ForgeMiner v1.8.8

Pearl no longer runs out of system memory on HiveOS rigs, Quantus mines on BaikalMine, and Quantus has fewer stale shares on pools that change jobs often.

What's new
  • Fix — Pearl on Linux / HiveOS ran out of system memory. 1.8.7 took about 0.5 GB of system RAM per card for Pearl, so rigs with 4 GB could stop right after start ("Killed", exit code 137). Pearl now needs about 140 MB per card, less than 1.8.6 (about 330 MB). Windows was not affected. If you set FORGE_PEARL_LEGACY=1 as a workaround, remove it.
  • Fix — Quantus on BaikalMine (qtc.baikalmine.com:2560): it mined nothing in 1.8.6 and 1.8.7 ("Invalid Quantus difficulty"), it works now. The Windows package has a new BaikalMine-Global.bat for Quantus.
  • Quantus — the pool connection stays up between shares. A pool that closes a quiet connection (BaikalMine does after 90 s) no longer drops the miner.
  • Quantus — fewer stale shares on pools that change jobs every few seconds (BaikalMine): the card moves to a new job about twice as fast. On pools with rarer jobs (Kryptex) nothing changes.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.8/ForgeMiner-1.8.8.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes.
  • Everything else is unchanged from 1.8.7.
Checksums (sha256)
5f5f2d2039d02adfe237a47691f4204095adf8b61c23e1f5416a326a6d02a6d0  ForgeMiner-1.8.8-windows.zip
4e86be6074558610b46892614667e843d83efd684e4d6324c67b7a066b458081  ForgeMiner-1.8.8-linux.tar.gz
c2b6bcfe08be104bda256828f76a342ad4418cb2b09da37f8dbe989cfab60964  ForgeMiner-1.8.8.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

ForgeMiner 1.8.7

· 发布页面

ForgeMiner v1.8.7

Pearl and Quantus are faster on the RTX 30/40/50 and no longer care about the memory clock on Pearl, NOID is faster on every card and now runs on the GTX 10-series, RTX 20-series and Tesla V100, and a dead pool connection is noticed in seconds instead of minutes.

What's new
  • Pearl — new kernels on the RTX 30/40/50 and the CMP 70HX/90HX. At the usual memory clock, against 1.8.6: about +2% on the RTX 40-series at −12% board power, +1.3% on the RTX 30-series at −17%, +3% on the RTX 50-series at −7%, +18% on the CMP 90HX, +1.3% on the CMP 70HX at −12%. The bigger news is the memory clock: Pearl no longer needs it. Set the memory to 810 MHz and the hashrate stays the same as at 5001 while the power drops further — on a 4070 Ti that is about +55% hashrate and +27% hashrate per watt against 1.8.6 at the same 810 MHz, and about +20% per watt against 1.8.6 at 5001. FORGE_PEARL_LEGACY=1 brings the previous kernels back.
  • Quantus — new kernels on the RTX 30/40: about +2.3% on the RTX 40-series and +1.4% on the RTX 30-series; the RTX 50-series gets +0.9%. Quantus now also runs on the Tesla V100.
  • NOID — new kernels on the RTX 30/40/50: about +14% on the RTX 40-series, +11% on the RTX 30-series and CMP 70HX, +9% on the RTX 50-series. NOID now runs on the GTX 10-series, RTX 20-series, CMP 40HX/50HX and Tesla V100 too.
  • Tesla V100 mines every coin; Conflux runs on the GTX 1080 Ti and the V100.
  • Fix — a dead pool connection is noticed in seconds. Every coin's pool socket now has the same TCP timeouts the Pearl path had: a route that stops answering is dropped and redialled in about 25 s instead of after seven minutes of mining into nothing (seen on btx-pool.com).
  • Fix — Conflux and BTX keep mining when the first card has no kernel for the coin (a GTX 10 in slot 0 used to stop the whole miner).
  • Fix — --cmp-install in a flight sheet shared with rigs without CMP cards: those rigs print one line and mine as usual instead of exiting and restarting.
  • Fix — Pearl worker name: without --worker, a flag value is no longer taken as the worker; ADDRESS.NAME works as on Quantus.
  • Clearer messages: --cclk/--mclk on cards that cannot lock clocks (GTX 10; memory on the RTX 20) now say which cards can and what to use instead; Xelis and NOID on Hopper and QubitCoin on the A100 say "no kernel for this card" instead of asking for a driver update.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.7/ForgeMiner-1.8.7.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes.
  • Everything else is unchanged from 1.8.6.
Checksums (sha256)
0e7708f27d0436b2f9f769ffdc34a3dbba626d41bce935d967a5edfdbb32306b  ForgeMiner-1.8.7-windows.zip
e9707644f2bd92bfac6db615a3463526ff47a6c8e9be567e85dc7cc44d8c42c2  ForgeMiner-1.8.7-linux.tar.gz
664c5f993690e1ceca68f025e8dbd278fcacbc67d329628b95e94256c84b8f51  ForgeMiner-1.8.7.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

ForgeMiner 1.8.6

· 发布页面

ForgeMiner v1.8.6

Quantus is faster on every generation and now runs on the GTX 10-series, Pearl is faster on Pascal and the CMP cards, and the overclock is put back when forge exits.

What's new
  • Quantus — new kernels on every generation. The GTX 10-series (Pascal) mines Quantus now; 1.8.5 left these cards idle. Against 1.8.5 on the same cards: about +11% on the RTX 20-series (2080 Ti, 2060 and CMP 50HX alike), +7% on the RTX 30-series, +8% on the RTX 40-series and +14% on the RTX 50-series, with hashrate per watt up by about the same.
  • Pearl on the GTX 10-series (Pascal) — about +55% hashrate and +34% hashrate per watt on a GTX 1080 Ti.
  • Pearl on the CMP 50HX and 40HX — both now mine the same tuned kernel as the RTX 20-series: about +3% on the 50HX and +2% on the 40HX at the same power.
  • Fix — RTX 30-series: Pearl no longer stops at start with CUDA error 700. Reported on Windows (3060 Ti, 3070 Laptop) and also seen on Linux. The FORGE_FRAG=0 workaround is no longer needed.
  • Fix — Quai no longer loses shares.
  • The overclock is put back when forge exits. forge restores what it set itself through --cclk, --coff, --mclk, --moff, --plimit and --fan; a setting someone changed in the meantime (a HiveOS profile, Afterburner) is left as it is.
  • New flags: --oc-keep keeps the overclock after forge exits (the 1.8.5 behaviour); --oc-reset puts the cards back to stock (default power limit, clocks unlocked, offsets 0) and exits.
  • Clearer messages: what went wrong and what to do — the pool's own reason for a refused login or share, why a pool connection ended, hints for CUDA errors and for an overclock setting the driver refused, a GPU number that does not exist. Long lines wrap in the log pane, and a message repeated many times is shown once with a counter.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.6/ForgeMiner-1.8.6.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes.
  • Everything else is unchanged from 1.8.5.
Checksums (sha256)
e7731ac9c6d7a3c79016a9b5a08c5cfb2b3df5f1b5709eee8cd720d1960b723d  ForgeMiner-1.8.6-windows.zip
d261e209331a168451e86776b57320a4a3e92857ca2970fdfa44428eb60585cc  ForgeMiner-1.8.6-linux.tar.gz
4fd2736fa52f0660698312f7970b9d3724acde91bdd180c780a8b5bf55d37f7b  ForgeMiner-1.8.6.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

ForgeMiner 1.8.5

· 发布页面

ForgeMiner v1.8.5

A large Pearl gain on the RTX 20-series, 8 GB cards get their profile back, faster Quantus, and a found share no longer stalls the card.

What's new
  • Pearl on the RTX 20-series (Turing) — a large gain. Until now only the RTX 2060 and the two CMP boards mined the tuned kernel; every other RTX 20-series card, Titan RTX and Quadro RTX fell back to a generic one that had never been measured against anything. They all mine the tuned kernel now — about +43% hashrate on an RTX 2080 Ti, and about +30% more hashrate per watt.
  • Fix — 8 GB cards get their deep profile back. 1.8.3 and 1.8.4 counted the memory headroom twice and dropped 8 GB cards (RTX 3060 Ti, 3070, 3070 Laptop, 2060 SUPER and others) to the basic profile: about 3% less hashrate, and on one reported rig a card that then produced only rejected shares. Both are fixed.
  • Pearl on the whole RTX 30-series. The faster path is now the family default instead of a single model — about +4% hashrate and +3% hashrate per watt.
  • Windows: memory use is fitted to the card's real free memory. On Windows the driver reports a per-process budget that does not include what other programs hold, so a card shared with a game, a desktop or another miner could be overcommitted — and Windows answers that by paging, which costs a large part of the hashrate rather than failing. Forge now fits against what is really free and keeps room for the desktop.
  • A found share no longer stalls the card. Every accepted share used to pause mining while its proof was rebuilt — about 100 ms on an RTX 3060 Ti, on every share. It is about 10 ms now, so fewer shares are lost on busy pools.
  • Quantus (--algorithm quantus) is faster again — about +4% on the RTX 30-series, +2% on the RTX 40-series and +2.5% on the RTX 20-series, with hashrate per watt up by the same amount — the cards draw no more power.
  • Overclock and fan flags now work on every coin. Quantus ignored --cclk, --coff, --mclk, --moff and --plimit completely, and --fan, --fan-curve and --oc-delay worked only on Pearl. Also fixed: on coins other than Pearl, a * or an empty slot in a per-card list shifted the remaining values onto the next card.
  • Pool difficulty is taken from the job's target when the pool does not state it — affects solo and some pool setups that leave the field out.
  • Smaller: the full per-card table and ASCII-only text when there is no console (running through PSEXEC or into a pipe); the debug output no longer prints internal names.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.5/ForgeMiner-1.8.5.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes.
  • Everything else is unchanged from 1.8.4.
Checksums (sha256)
00932919ca03878a66dc433298ee715c8f473dd08be114104095cf9d75aca20c  ForgeMiner-1.8.5-windows.zip
94a442738029c7edd757add0876f29814aeedf40a463af57b0a5c7ef45fc0f6d  ForgeMiner-1.8.5-linux.tar.gz
9508eabc3d12bca32a506885a5a81d64d67e5c29bfb3f0995512a0e36d9f0fe5  ForgeMiner-1.8.5.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

ForgeMiner 1.8.4

· 发布页面

ForgeMiner v1.8.4

Share difficulty, best share and blocks found on every coin, Quantus off the CPU, and more accepted shares.

What's new
  • New — every share's difficulty, the session's best share and blocks found, on every coin. Each accepted share shows its own difficulty, the dashboard keeps the best share of the session, and a share that is also a block is flagged and counted. Best share and blocks sit in the totals row; accepted / stale / rejected is back to three numbers.
  • Quantus (--algorithm quantus) no longer keeps a CPU core busy for every GPU — CPU load drops from about one core per card to almost zero, which matters on rigs with small CPUs.
  • Quantus hashrate is slightly higher on almost every GPU.
  • More accepted shares: a share found just after the pool switched jobs is now still sent while the pool accepts it, instead of being thrown away. Biggest gain on Xelis, where jobs change every few seconds; smaller on Quantus, Cryptix, KawPow (RVN / QUAI / XNA) and qhash.
  • HiveOS custom miner — upgrade with the versioned URL: https://github.com/0xHashRaptor/ForgeMiner/releases/download/v1.8.4/ForgeMiner-1.8.4.tar.gz. HiveOS reinstalls a custom miner only when its installation URL changes, so rigs on the old .../releases/latest/download/ForgeMiner.tar.gz link stay on the version they first installed.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • Everything else is unchanged from 1.8.3.
Checksums (sha256)
c61474c1edad5052b717f5a46a1900ea4056871b042f9da97351647ac356de1b  ForgeMiner-1.8.4-windows.zip
491296d32a76eae56497830e885c1c671d20f9b43a44d14c3f1c3776eeeca192  ForgeMiner-1.8.4-linux.tar.gz
a878789682981753727623c2af61af99329bf816c94fd7ce562bc838dae74f7f  ForgeMiner-1.8.4.tar.gz

ForgeMiner 1.8.3

· 发布页面

ForgeMiner v1.8.3

Faster Quantus on every GPU, plus a startup fix for 6 GB cards.

What's new
  • Fix — 6 GB cards (e.g. RTX 2060) on Windows: in 1.8.2 Pearl could start and then show 0 H/s on a card with little free memory. Fixed — these cards mine normally again.
  • Quantus (--algorithm quantus) is faster and more efficient on every supported GPU — up to about +3% more hashrate, and more hashrate-per-watt too, depending on the card (RTX 20/30/40/50-series). Same dev fee, same pools.
  • New — skip a card in the overclock flags with *: e.g. --cclk 1500,1500,*,1500 sets the 1st, 2nd and 4th cards and leaves the 3rd untouched. Works for --cclk, --coff, --mclk, --moff, --plimit and --fan.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • Everything else is unchanged from 1.8.2.
Checksums (sha256)
0ae45c24e7a1832a48a5eba1ad3bc3cf67d6a291c5ee8234ab3f08a8098ed090  ForgeMiner-1.8.3-windows.zip
3849b75e5e0eab89aeb04895449c377cbbf43ba9baff2372f4bdbe6aa0e8008b  ForgeMiner-1.8.3-linux.tar.gz
73d0f3fc637f67c49f27662364c810f39c2c67ee7dc7bf95897a044b15298b60  ForgeMiner-1.8.3.tar.gz

ForgeMiner 1.8.2

· 发布页面

ForgeMiner v1.8.2

Pearl runs cooler and more efficient on every GPU, and faster on the RTX 50-series.

What's new
  • Pearl (--algorithm pearlhash) draws less power on every supported GPU — the same hashrate at lower wattage, so higher efficiency (hashrate per watt). Up to about 8% more efficient depending on the card. Same dev fee, same pools.
  • RTX 50-series (Blackwell): faster Pearl — about +2%.
  • RTX 20-series (Turing): higher Pearl hashrate — about +2%.
  • The RTX 50-series (Blackwell) requires NVIDIA driver 580 or newer (unchanged).
  • Everything else is unchanged from 1.8.1.
Checksums (sha256)
9a8d6371ab68f0751048e7ce5d46086e29c803472621bac52ceb6c6d53a88e27  ForgeMiner-1.8.2-windows.zip
2c8921708eeedad79f95e7ebe2e6f142c9ab28a6195bb8a884401d0b1779b299  ForgeMiner-1.8.2-linux.tar.gz
ee7936854a4e83af567691edaea4e20b423711b599d1db3a555a532e698096a8  ForgeMiner-1.8.2.tar.gz

Download the assets below · docker pull hashraptor/forge · forgeminer.org · Telegram · Discord

下载 ForgeMiner

1.9.1
Windows Linux