Changelog
redPeaRL 0.7.0-beta.4 (17 Oct 2026) - usage stats fixed; first public-beta build
First build of the public beta. Download from GitHub Releases; the zip's SHA-256 and the exe's are listed there and in BUILD.txt. Internal builds made the same day while this was being chased were never distributed.
Fixed
- Usage stats stopped after a few reports. The report sender hung after each exchange when it tore down its connection objects on Windows; one hung report silenced stats for the rest of the session while mining carried on, in every session since beta.1. The connection machinery (event loop, name resolver, TLS context) is now created once and kept for the life of the process, only the per-report stream comes and goes, each report runs in its own thread under a 90-second watchdog, the response is read under a 15-second deadline, and the log says when a report is refused, stuck or left a clean-up pending. Verified on the home card: reports every five minutes for 81 minutes with a flat thread count.
Documentation
- POOL-EXAMPLES.md: Kryptex custom share difficulty through the password line (
d=N), with the pool's formula.
Unchanged
- Everything else is 0.7.0-beta.3. Not code-signed yet; one GPU per process; Windows only; the self-test needs the GPU to itself.
redPeaRL 0.7.0-beta.3 (17 Oct 2026) - Rabid Pool preset and pool examples
Same build as 0.7.0-beta.2 plus a pool preset and a setup file.
Added
- Rabid Pool preset:
pool = "rabidmining"connects to the PPLNS port; SOLO on that pool ispool = "stratum+tcp://prl.rabidmining.com:9934". Checked on 17 October from our own card: login, compressed proofs and shares accepted. - POOL-EXAMPLES.md in the package: a ready-to-paste
redpearl.tomlblock for each supported pool (Kryptex, HeroMiners, LuckyPool, PearlFortune, Rabid Pool PPLNS and SOLO) and the pattern for any other stratum pool.
Unchanged
- Everything else is 0.7.0-beta.2. Not code-signed yet; one GPU per process; Windows only; the self-test needs the GPU to itself.
redPeaRL 0.7.0-beta.2 (17 Oct 2026) - start-up fix and smaller proofs
Replaces 0.7.0-beta.1, which never reached a tester. Same package layout; SHA-256 of every file in BUILD.txt.
Fixed
- The beta.1 exe did not mine. A start-up lookup ran with an empty coin symbol and, because the product build blanks the names of every coin it does not mine, the empty symbol matched those blank rows and replaced Pearl. Found on the first run of a packaged build with
redpearl.toml; the lookup now runs only when a symbol was typed.
Changed
- Proofs are a third smaller on the wire. The gzip encoder gained dynamic-Huffman blocks; a proof for the default shape compresses about 98x instead of 64x, and the round-trip test against Python's gzip now runs in CI.
- A small efficiency gain on power-limited cards, from an internal change to how the kernel is fed; measured on our own card, not on anyone else's yet.
Known issues and limits
- Unchanged from 0.7.0-beta.1: not code-signed yet; one GPU per process; Windows only; the self-test needs the GPU to itself.
redPeaRL 0.7.0-beta.1 (16 Oct 2026) - first closed-beta build
The first build meant for people other than us. Windows, AMD Radeon RX 9070 XT and RX 9060 XT (RDNA4). Linux build in development. Package: redpearl-miner.exe + kernels\ + redpearl.toml + Start-redpearl-miner.bat + HELP.md + EULA.md + THIRD_PARTY_NOTICES.md + LICENSE + BUILD.txt (commit, build time, SHA-256 of every file).
SHA-256 of the package files: see BUILD.txt inside the zip; the exe's hash is repeated on the download page.
What changed since 0.6.0
Getting started
- First start without a wallet writes a commented
redpearl.tomlnext to the exe and stops with a plain message; the launcher no longer needs editing (wallet, worker, pool and all switches live in the config file;RP_*environment variables and command-line flags override it). HELP.md: quick start, what a healthy start looks like, every config key, pools, the fee, GPU settings, benchmark, troubleshooting, logs, how to report.- Licence acknowledgement:
accept_eula = trueinredpearl.tomlafter readingEULA.md; the miner does not mine without it. EULA 1.0 (beta) ships in the package.
Pools
- Presets for Kryptex (full support: TLS, every region, worker names, compressed proofs), HeroMiners, LuckyPool and PearlFortune with the login form each pool wants;
pool = "luckypool:eu1"style region pick orregion = "auto"(fastest from here).--pool-testtimes every region of every preset. Plain-language help after three failed connections in a row (which step failed, what to try). - Share lines carry the submit-to-reply time; one
job:line per new job at verbose.
Developer fee (1 %)
- Mines on Kryptex as worker
fee-<your pool>; if Kryptex cannot be reached for three attempts it mines on your pool, and tries Kryptex again after an hour. Your mining never waits for the fee connection (verified: the fee fell back and cycled while the user side kept every share). - Pools closing an idle fee login is routine now: reconnect in about a second, no red error.
Logging and status
- Log levels
quiet | normal | verbose | debug; start-up at normal is 8 lines; the 30-second status line shows TH/s, 15-minute average, accepted/rejected, GH/J, watts, GHz, mV, edge/hotspot/memory temperatures and fan. Hashrate is pool-equivalent (1 TMAC/s = 1 pool TH/s) and measured from an exact work clock; power is averaged over the 30 s. - The miner writes its own log:
miner.lognext to the exe, a ring of 5 x 10 MB files, timestamps with time zone, colours stripped;[log]keys for path, size and count. - Sensors come from AMD's ADLX runtime (read-only); the miner never writes to the card.
GPU and engine
- No GPU tuning in the product at all: power limit, clocks, voltage, memory timing and fan belong to Adrenalin. The ADLX/ADL/RSMI writers,
tune.txtand every--gpu-*knob are gone. Reason: five "undervolt applied but not effective" incidents on the home card, two with no tuner write; the applied tune lives in Adrenalin's own stack. - Default batch 4 (was 16): about +0.5 % on a power-limited card, about -0.2 % uncapped; batch 4-32 via
[advanced] batch. - Wide shape 4096 x 65536 x 8192 (0.6.0) kept; the self-test (CPU/proof oracle) runs before the first share on every start; low-memory fallback halves the batch when the buffers do not fit.
- Controlled shutdown: Ctrl+C, console close or logoff finish the batch in flight, give queued shares up to 2 s, flush the log, exit 0; a second Ctrl+C forces exit.
Anonymous usage stats (on by default, opt-out)
- Every 5 minutes while mining, one small record: random install id, miner version, OS, GPU model / architecture / VRAM, your settings (pool, batch, shape, log level, fee pool, EULA state), hashrate, shares, GH/J, power, clock, voltage, temperatures, fan, uptime, self-test result, error counts, last error, pool reconnects; one crash record on a fatal error. Never: wallet, worker name, password, computer name, files; wallet / worker / password are masked out of any text; no IP stored.
stats = falseturns it off; the start-up line says which it is. Full list in HELP.md. No work inside the mining loop; the sender runs on its own thread with a 10 s budget.
Product build hygiene
- Developer modes, per-batch diagnostics, runtime-compiler loaders, upstream names and other coin families are compiled out of the product; a strings audit (
tools/audit-product-strings.py) fails the CI product build on any of them. Symbols stripped, no build paths.
Verified on this build (home RX 9070 XT, 16 Oct 2026)
- First run: template written, exit 2 with the message; config present but no wallet: message, exit 2; licence not accepted: message, exit 2;
--pool-test kryptex: 7 regions ranked. - Mining: self-test PASS, shares accepted, 0 rejected.
- Fee routing: fallback after 3 empty attempts in 9 s, return attempt at the retry window, cycling; user side 12 shares / 0 errors in 8 min.
- Stats: first record 71 s after start, HTTP 204, visible in our collector; mining unaffected.
- Product strings audit: 0 hits.
Known issues and limits
- Not code-signed yet; signing is planned for the release. The SHA-256 of every file is in BUILD.txt.
- One GPU per process. Several cards: one process each with
device = Nand separate worker names. - Windows only in this beta; the Linux build (including a container image) is in development.
--benchmarkand the self-test need the GPU to themselves; another miner or heavy program on the same card trips the 5-second self-test watchdog.- The Linux/Salad image is a separate build and not part of this package.
Reporting
Email support@redpearlminer.com with your card, your Adrenalin power/voltage settings, what you expected, and miner.log (or its last 200 lines). Thank you for testing.