Repository navigation
FreeForGem 1.7.7 — release notes
English
Files: freeforgem-1.7.7_hiveos.tar.gz (HiveOS Installation URL
https://github.com/pepsykolya/FreeForGem/releases/download/v1.7.7/freeforgem-1.7.7_hiveos.tar.gz),
freeforgem-1.7.7-linux-x86_64.tar.gz, freeforgem-1.7.7-windows-x64.zip. Check the SHA-256 (MANUAL.md, section 9).
| File | SHA-256 |
|---|---|
freeforgem-1.7.7_hiveos.tar.gz |
b61356d568428290c0060e6c8a169a820f970a9c9a5d2ae41da6b322cad6ebcd |
freeforgem-1.7.7-linux-x86_64.tar.gz |
fd0697eec486add58fc3f9b7d2cefd2be71ef70f103692cf0d414ce89c701c31 |
freeforgem-1.7.7-windows-x64.zip |
0ea283ec8c9b5406ecfed86637e6f60d0cb0170c6fac041ff93d82b769193f68 |
Small efficiency gains and two new helpers.
- NOID: about +0.6 % MH/W (RTX 30 and RTX 40,
--noid-mode eff): a new eff kernel (t512vs) for all supported architectures, same hashrate
within 0.15 %, about 0.7 % less power. The saved auto-uv profiles of NOID are re-checked once (the kernel name is part of the profile);
NOID_KERNEL=t512vabrings the old kernel back. - Pearl, RTX 30: about +0.3 % hashrate at the same power. The RTX 30 kernel works on a larger jackpot tile (16 x 16 instead of 8 x 16). With
--target-mode jackpot(the pool sends a threshold made for the smaller tile) the miner runs the previous 8 x 16 kernel, also when the mode
changes while mining. Ada, Turing and RTX 50 are unchanged. The profiles of auto-uv for RTX 30 are re-checked once. - Pearl, RTX 30: a hint in the log to lock the core at 1530 MHz (
--lock-cclock 1530): about 3.6 % better TH/W for about 2 % less hashrate on an
RTX 3070. The miner only prints the hint; it sets nothing. A new lock gets its own auto-uv search. - New:
--cmp-driver(HiveOS, off by default): installs the unlocked NVIDIA driver build for CMP cards. In the flight sheet add the word
cmp-unlock-driverto Extra config arguments. It replaces the kernel driver of the whole rig and reboots it: read MANUAL.md, section 20, first. - HiveOS: when a newer release is out, the dashboard shows
need update!!next to the version (1.7.7 0.652 TH/W, need update!!). - CMP 50HX: the auto-uv limit is raised from +225 to +330 (
--uv-maxstill overrides it; RTX 20 / GTX 16 stay at +225). A saved result that stopped at the old limit (no failing step, not the voltage floor) is searched further up automatically from that offset at the next start; Pearl, NOID and CMFD. - Memory clock lock check: the miner now warns (yellow, once per card) when a memory clock lock does not hold under load. Example:
--lock-mclock 405on an RTX 3070 is accepted by the driver, but the card computes at 810 MHz; the miner used to say only "locked to 405". - auto-uv: a repeated search now starts one coarse step below the last result of the card instead of from the bottom; the "voltage floor" verdict is given only at a comparable card temperature (a card that is still warming up is no longer reported as having hit the floor); on a rig with little free RAM, a check that could not start is planned again once the miner has settled (the check plan keeps 128 MB spare instead of 256).
- Windows: the console window no longer freezes the miner (QuickEdit is switched off, so a click in the window does not stop it), the output goes through a bounded queue with its own writer thread, and the statistics line is no longer painted red after one rejected share (only real events are coloured).
- Pearl on Ada (RTX 40) uses the same kernel as 1.7.6 and was not re-tested on hardware in this release.
Русский
Файлы: freeforgem-1.7.7_hiveos.tar.gz (HiveOS Installation URL
https://github.com/pepsykolya/FreeForGem/releases/download/v1.7.7/freeforgem-1.7.7_hiveos.tar.gz),
freeforgem-1.7.7-linux-x86_64.tar.gz, freeforgem-1.7.7-windows-x64.zip. Проверьте SHA-256 (MANUAL.md, раздел 9).
| Файл | SHA-256 |
|---|---|
freeforgem-1.7.7_hiveos.tar.gz |
b61356d568428290c0060e6c8a169a820f970a9c9a5d2ae41da6b322cad6ebcd |
freeforgem-1.7.7-linux-x86_64.tar.gz |
fd0697eec486add58fc3f9b7d2cefd2be71ef70f103692cf0d414ce89c701c31 |
freeforgem-1.7.7-windows-x64.zip |
0ea283ec8c9b5406ecfed86637e6f60d0cb0170c6fac041ff93d82b769193f68 |
Небольшие выигрыши в эффективности и два новых помощника.
- NOID: около +0,6 % МН/Вт (RTX 30 и RTX 40,
--noid-mode eff): новое eff-ядро (t512vs) для всех поддерживаемых архитектур, хэшрейт тот же
(в пределах 0,15 %), мощность примерно на 0,7 % ниже. Сохранённые профили auto-uv для NOID перепроверятся один раз (имя ядра входит в профиль);
NOID_KERNEL=t512vaвозвращает старое ядро. - Pearl, RTX 30: около +0,3 % хэшрейта при той же мощности. Ядро RTX 30 работает на большем jackpot-тайле (16 x 16 вместо 8 x 16). При
--target-mode jackpot(пул присылает порог под меньший тайл) майнер запускает прежнее ядро 8 x 16, в том числе когда режим меняется на лету.
Ada, Turing и RTX 50 без изменений. Профили auto-uv для RTX 30 перепроверятся один раз. - Pearl, RTX 30: подсказка в логе про лок ядра 1530 МГц (
--lock-cclock 1530): около 3,6 % лучше TH/Вт при примерно 2 % меньшем хэшрейте на
RTX 3070. Майнер только пишет подсказку и ничего не ставит. Для нового лока auto-uv ищет свой профиль. - Новое:
--cmp-driver(HiveOS, по умолчанию выключено): ставит разблокированную сборку драйвера NVIDIA для карт CMP. В полётном листе добавьте
словоcmp-unlock-driverв Extra config arguments. Заменяется драйвер ядра на всём риге и рига перезагружается: сначала прочитайте MANUAL.md, раздел 20. - HiveOS: когда вышла новая версия, на панели рядом с версией пишется
need update!!(1.7.7 0.652 TH/W, need update!!). - CMP 50HX: предел auto-uv поднят с +225 до +330 (
--uv-maxпо-прежнему главнее; RTX 20 / GTX 16 остаются на +225). Сохранённый результат, упёршийся в старый предел (сбойного шага не было, это не voltage floor), при следующем старте автоматически дотюнивается вверх от этого offset; Pearl, NOID и CMFD. - Проверка лока памяти: майнер теперь предупреждает (жёлтым, один раз на карту), если лок частоты памяти не держится под нагрузкой. Пример:
--lock-mclock 405на RTX 3070 драйвер принимает, но карта считает на 810 МГц; раньше майнер писал только «locked to 405». - auto-uv: повторный подбор теперь стартует на один крупный шаг ниже прошлого результата карты, а не с нижней границы; вердикт «voltage floor» выносится только при сопоставимой температуре карты (прогревающаяся карта больше не считается упёршейся в пол); на риге с малым запасом RAM проверка, которая не смогла стартовать, планируется заново после прогрева майнера (план проверок оставляет запас 128 МБ вместо 256).
- Windows: окно консоли больше не замораживает майнер (QuickEdit выключен, клик в окне его не останавливает), вывод идёт через ограниченную очередь с отдельным потоком записи, строка статистики не краснеет после одного отклонённого шара (красятся только события).
- Pearl на Ada (RTX 40) работает тем же ядром, что в 1.7.6, и в этом релизе на железе повторно не проверялся.