[ANN] FreeForGemSaseul (ffgs) 0.5.4 — SASEUL (SL) GPU miner | NVIDIA + AMD | Lin

[pepsykolya]

FreeForGemSaseul (ffgs) — SASEUL (SL) GPU miner for NVIDIA and AMD
v0.5.4 · GitHub · Download · Manual (EN / RU / 中文) · Pool: Kavatar SL Pool

A fast CUDA / OpenCL miner for SASEUL (SL) — the resource-chain PoW of saseul.com (single SHA-256 over previous_blockhash + header + nonce).

Features

• Linux x86_64, HiveOS (custom miner package, NVIDIA and AMD rigs) and Windows x64
• NVIDIA Pascal … Blackwell (GTX 10xx – RTX 50xx), driver 525+ (on Windows only the driver is needed, no separate CUDA install)
• AMD Radeon: OpenCL, tuned for RDNA2, same binary picks the backend itself (--backend auto|cuda|amd); HiveOS detects AMD cards and runs mixed NVIDIA + AMD rigs
• Pools: Kavatar SL Pool (ours, the default), rabbitminer.cc, saseulmine.com, saseulpool.com, and solo on your own node via the bundled solo proxy; failback to the primary pool and a no-accepted-shares watchdog
• Built-in GPU clock control (NVIDIA, per GPU or as per-GPU lists for mixed rigs, x = leave alone, =0): --lock-cclock, --lock-mclock, --coffset (undervolt), --moffset, --plimit. Safe stop: the GPU work stops first, then the clocks are restored (on exit and after a crash). ffgs clock options win over the HiveOS OC profile
• --auto-uv: finds the undervolt per GPU (30/15 MHz steps, ~3.5–11.5 min, stops at the voltage floor), a profile per core clock lock, the other GPUs mine while one is tuned; --auto-lock: picks the core clock lock per GPU (best MH/W within 10% of the top GH/s)
• GPU stall / hashrate-drop watchdog and self-restart (Linux and Windows); --temp-limit for laptops; local read-only --api (JSON / Prometheus); rig MH/W in HiveOS; Windows autostart and a new-version notice
• Every share is re-checked on the CPU before it is sent; a self-test against a real mainnet block runs at every start
• Colored console: accepted shares green, errors red, hashrate white, MH/W blue, temperature by threshold; a table every 60 s and your wallet as the pool sees it every 5 min (--color auto|always|never)

Kavatar SL Pool — kavatar.kasplay.online

• Stratum sl.kasplay.online:3333, login <SL address>.<worker>, no password
• PPLNS (2 × network difficulty), pool fee 2%, automatic payouts every 6 h from 10 SL, own SASEUL node
• Site: pool and network hashrate charts, blocks; wallet lookup — workers with stale / rejected shares, address hashrate, your share of the PPLNS window and how full it is (with a forecast of when you reach your full share), coins on the address in the chain, credits and payouts; a ready HiveOS flight sheet for Import from Clipboard

Other pools

• rabbitminer.cc — nl/fi.rabbitminer.cc:1931 (TCP), pool fee 10%, payout from 100 SL
• saseulmine.com — --pool saseulmine (HTTPS), daily payouts, vardiff
• saseulpool.com — --pool saseulpool (NATS), PPLNS 90 min, pool fee 5%, payouts every 6 h from 10 SL
• Solo — your own synced SASEUL node + bundled solo/sl_solo_proxy.py (MANUAL §8)

Recommended settings (measured with ffgs 0.5.4, memory locked 810, Kavatar SL Pool)

Card	Goal	Settings	Hashrate	Power	Efficiency
RTX 3070	best MH/W	1500 + --auto-uv	5.62 GH/s	76 W	73.9 MH/W
RTX 3070	balance	1700 + --auto-uv	6.40 GH/s	106 W	60.4 MH/W
RTX 3070	max hashrate	1800 + --auto-uv	6.74 GH/s	121 W	55.7 MH/W
RTX 3080 Ti	best MH/W	1600 + --auto-uv	10.32 GH/s	180 W	57.4 MH/W
RTX 3080 Ti	balance	1700 + --auto-uv	10.98 GH/s	212 W	51.9 MH/W
RTX 3080 Ti	max hashrate	core not locked	11.54 GH/s	332 W	34.8 MH/W
any NVIDIA	let ffgs choose	--lock-mclock 810 --auto-lock	best MH/W within 10% of the top GH/s	—	3080 Ti: it chose 1700

Other cards (ffgs 0.5.2, same kernel as 0.5.4)

Card	Settings	Hashrate	Power	Efficiency
RTX 3070 Laptop	1400 / mem 810	4.60 GH/s	53 W	87 MH/W
RTX 4070 Ti	2520 / mem 810	11.98 GH/s	173 W	69 MH/W
RX 6800 XT	stock	9.86 GH/s	238 W (ASIC)	41 MH/W
RX 6800 XT	2300 MHz / 881 mV	9.69 GH/s	164 W (ASIC)	59 MH/W

GTX 10/16, RTX 20 and RTX 50 run the RTX 40 settings — not measured by us yet, reports welcome. AMD numbers are from 0.5.0 (the AMD kernel is unchanged since). rabbitminer / saseulpool use 16-hex nonces: RTX 30 about 10% lower than on Kavatar.
Every card differs — if "hardware error (bad candidate)" lines or rejects appear, lower the offset or the core clock (or let --auto-uv do it). Memory clock does not matter for SL — lock it low.
Updating from 0.5.2 or older with an undervolt: the last stop is still done by the old version; if a GPU shows [unavailable] after the update, reboot once.

Quick start — Linux

Code:

tar xzf ffgs-0.5.4-linux-x86_64.tar.gz && cd ffgs-0.5.4
./ffgs --wallet <SASEUL address>.<worker>                       # Kavatar SL Pool (default)
./ffgs --pool saseulmine --wallet <SASEUL address>
./ffgs --pool nl.rabbitminer.cc:1931,fi.rabbitminer.cc:1931 --wallet <SASEUL address>.<worker>
sudo ./ffgs --wallet <address>.<worker> --referrer <address> --lock-cclock 1500 --lock-mclock 810 --auto-uv

Quick start — Windows
Unzip ffgs-0.5.4-windows-x64.zip, open start.bat (Kavatar SL Pool) in Notepad, put your address after WALLET= (and after REFERRER= for a 1% fee), save, double-click. Other pools: start-rabbitminer.bat, start-saseulmine.bat, start-saseulpool.bat. Start at logon: install-autostart.bat (remove: remove-autostart.bat).

Quick start — HiveOS
Easiest: Flight Sheets → Import from Clipboard with the JSON from kavatar.kasplay.online (HiveOS section, copy button). By hand: miner Custom, installation URL https://github.com/pepsykolya/FreeForGemSaseul/releases/download/v0.5.4/ffgs-0.5.4.tar.gz, hash algorithm saseul, wallet template %WAL%.%WORKER_NAME%, pool URL sl.kasplay.online:3333, extra config --referrer <your address> --lock-cclock 1500 --lock-mclock 810 --auto-uv (RTX 3070; ffgs keeps these clocks even if the OC profile has other values). AMD undervolt values for the HiveOS OC profile (core, VDD, VDDCI, MVDD, SoC): HIVEOS.md.

Fee: 2% = 1% dev + 1% referral (two 60 s windows every 100 min, the first after 15 min). The referral minute goes to --referrer ADDRESS — whoever recommended ffgs to you. Put your own mining wallet there and the total fee is 1% (the dev minute is then mined on Kavatar SL Pool). Without --referrer both minutes go to the dev. Never mined on your own node.

Video tutorials (0.5.4, EN / RU / 中文): English · Russian · Chinese (中文) — what's new, setup on Windows / Linux / HiveOS, the referral fee, undervolt, auto-lock and measured hashrates.

SHA256

Code:

091da888d3db5889629853ad837495ddf28259ea10a20c608e94862d5fe75c67  ffgs-0.5.4.tar.gz
f46d3975757762d888a1f045e909839edee9d406a43b4e7e52d9add4c77130ea  ffgs-0.5.4-linux-x86_64.tar.gz
142c65ce509183dda70a432651e5b599272f428b9cd8d0bfcbef5a9bb12466a2  ffgs-0.5.4-windows-x64.zip

Questions, results on other cards (especially RDNA3/RDNA4, Linux AMD, older Pascal/Turing) — post them here, we add them to the manual.

[pepsykolya]

ffgs 0.5.0 released — Kavatar SL Pool, +9% on RTX 30, HiveOS for AMD, referral fee

Download: GitHub release v0.5.0 (HiveOS, Linux, Windows + SHA256SUMS)
Video tutorials (two hosts): English · Russian

What's new

• Kavatar SL Pool — our own pool, now the default: sl.kasplay.online:3333, PPLNS, 2% pool fee, automatic payouts every 6 h from 10 SL, own SASEUL node. Stats, charts and wallet lookup: kavatar.kasplay.online
• Faster kernel on RTX 30: +9% on 40-hex pools (Kavatar, saseulmine, solo). RTX 3070 @1400 MHz / mem 810 / --coffset 325: 5.05 GH/s at 74 W (68 MH/W), was 4.63 GH/s. rabbitminer / saseulpool unchanged
• HiveOS for AMD and mixed rigs: AMD cards detected, AMD OpenCL checked, CUDA and AMD run side by side; AMD undervolt values for the HiveOS OC profile (core, VDD, VDDCI, MVDD, SoC) in HIVEOS.md
• HiveOS flight sheet in one paste: Flight Sheets → Import from Clipboard with the JSON from the pool site (copy button)
• Fee 2% = 1% dev + 1% referral: the referral minute goes to --referrer ADDRESS (whoever recommended ffgs to you). Put your own wallet there and the fee is 1%. Without --referrer both minutes go to the dev
• Console colors: accepted shares green, errors red, hashrate white, MH/W blue, temperature by threshold; --color auto|always|never

HiveOS installation URL: https://github.com/pepsykolya/FreeForGemSaseul/releases/download/v0.5.0/ffgs-0.5.0.tar.gz
Full changelog: CHANGELOG.md · Manual (EN / RU / 中文): MANUAL.md

[pepsykolya]

ffgs 0.5.2 released — faster CUDA kernel: up to +5% on RTX 30 and +12% on RTX 40 (Kavatar SL Pool, saseulmine, solo; same clocks, 0.5.1 → 0.5.2)

GPU	Clocks	0.5.1	0.5.2
RTX 3070	1400 / 810, offset +325	5.045 GH/s, 68.4 MH/W	5.306 GH/s, 73.4 MH/W
RTX 3080 Ti	1400 / 810	8.60 GH/s, 166 W	9.05 GH/s, 162 W
RTX 3070 Laptop	1400 / 810	4.38 GH/s, 54 W	4.60 GH/s, 53 W
RTX 4070 Ti	2520 / 810	10.71 GH/s, 165 W	11.98 GH/s, 173 W

rabbitminer / saseulpool (16-hex nonces): RTX 30 +2.5%, RTX 40 unchanged. Selftest at start, every share re-checked on the CPU, 0 hardware errors in all tests. AMD unchanged. GTX 10/16, RTX 20 and RTX 50 get the RTX 40 settings — not measured by us, reports welcome.

Video (EN, 12 min): https://youtu.be/xMoA-iAuHuA — the coin, wallet, Kavatar SL Pool, setup and hashrates.

Download: v0.5.2 on GitHub · HiveOS installation URL: https://github.com/pepsykolya/FreeForGemSaseul/releases/download/v0.5.2/ffgs-0.5.2.tar.gz

[pepsykolya]

ffgs 0.5.4 released — safe stop, auto-uv about twice as fast, --auto-lock, watchdogs, local API

Update recommended for everyone who uses an undervolt (--auto-uv, --coffset, a HiveOS OC offset): 0.5.2 and older put the clocks back while the GPUs were still mining, and stopping at a high offset could cause driver errors (Xid 13 / 31 / 43), a GPU unavailable until a reboot or a hang of the rig. 0.5.4 stops the GPU work first, then restores the clocks (0 driver errors in 30+ test stops at offsets up to +495). The last stop during the update is still done by the old version: if a GPU shows [unavailable] afterwards, reboot once.

What's new (0.5.4 includes the unreleased 0.5.3)

• --auto-uv: 30 MHz steps with a 15 MHz check only around the failing step (~3.5–11.5 min per GPU), stops at the voltage floor when the watts stop falling, a profile per core clock lock (applied at once when you switch back), the other GPUs mine while one is tuned
• --auto-lock: ffgs measures several core locks per GPU, each at its own undervolt, and keeps the best MH/W within 10% of the top GH/s
• Per-GPU lists for mixed rigs (--lock-cclock 1500,1600, x = leave that GPU alone, =0 = offset exactly 0), --dry-run
• ffgs clock options win over the HiveOS OC profile (set again within 10 s, one log line); a GPU crash while another tool changes the clocks is not counted against the undervolt
• GPU stall / hashrate-drop watchdog, failing GPU named from the driver's Xid, self-restart on Linux and Windows; failback to the primary pool and a no-accepted-shares watchdog
• --temp-limit (mainly for laptops), local read-only --api (JSON / Prometheus), rig MH/W next to the version in HiveOS
• Memory clock fallback for cards without 810 (CMP 50HX → 5001); RTX 20 auto-uv fixed (bad offset range from the driver); HiveOS Extra config over several lines no longer cut after the first line
• Windows: install-autostart.bat / --install-autostart (Task Scheduler at logon) and a new-version notice in the log (nothing sent, nothing downloaded)

Measured (memory lock 810, Kavatar SL Pool)

GPU	Best MH/W	Balance	Max hashrate
RTX 3070	1500 + --auto-uv: 5.62 GH/s, 76 W, 73.9 MH/W	1700 + --auto-uv: 6.40 GH/s, 106 W	1800 + --auto-uv: 6.74 GH/s, 121 W
RTX 3080 Ti	1600 + --auto-uv: 10.32 GH/s, 180 W, 57.4 MH/W	1700 + --auto-uv: 10.98 GH/s, 212 W	not locked: 11.54 GH/s, 332 W

RTX 3070 in HiveOS Extra config arguments: --lock-cclock 1500 --lock-mclock 810 --auto-uv, or --lock-mclock 810 --auto-lock.

Video tutorials (0.5.4, two hosts): English · Russian

Download: v0.5.4 on GitHub · HiveOS installation URL: https://github.com/pepsykolya/FreeForGemSaseul/releases/download/v0.5.4/ffgs-0.5.4.tar.gz
Full changelog: CHANGELOG.md · Manual (EN / RU / 中文): MANUAL.md