Built-in hardware unlock for CMP 50HX and CMP 90HX on Linux — one command, one reboot, both cards run at full speed

More Pearl hashrate on Turing and Blackwell cards

RTX 40-series: +1.3% hashrate at unchanged power draw. Measured against 1.5.11 in a single interleaved series on one card: 139.92 -> 141.78 TH/s, 91 accepted shares, zero rejects RTX 30-series: -1.6% power draw at unchanged hashrate (122.09 -> 120.08 W, sign 6 of 6 pairs, confirmed by two independent meters). Both families gain about +1.65% efficiency Shares no longer vanish on a flaky route to the pool. A submitted share is now 26% smaller on the wire, disconnects roughly halve, and the share of submits that actually receive an answer rose from ~40% to ~70% on an affected route. Nothing about the mined work changed - only how much of it survives a bad network TLS now works everywhere. On HiveOS it was previously impossible: the launcher stripped ssl:// from the pool address, so the miner opened a plain connection to a TLS port and reconnected forever. The address is now passed through intact, the miner detects TLS by itself when no scheme is given, and --tls (or FORGE_TLS=1) is accepted for those who prefer to be explicit Disconnects now tell you why — pool closed the connection, read error, or the pool stopped answering submitted shares. A link that was working reconnects in 1 second instead of 5, and the difficulty on screen is the one the pool actually assigned Two identical installations no longer duplicate each other's work, and a pool address with a scheme no longer breaks the non-Pearl coins

Shares could be lost without any trace. If the route to your pool died in one direction — the pool's jobs still arrive, but your submits never reach it — the miner did not notice. Submits piled up unsent in the system buffer while the screen showed full hashrate, 100% accept and zero rejects. The pool, meanwhile, credited only a fraction of the work. On an affected rig this cost up to 90% of the shares until the connection eventually reset itself, minutes later Two independent recoveries now. A stalled connection is dropped within ~25 seconds instead of grinding on for up to 15 minutes, and a new watchdog reconnects when a submitted share goes unanswered for 45 seconds (normal reply latency is well under a second). Both print a clear warning — this failure used to be completely invisible Nothing else changed: kernels, hashrate and power draw are identical

Pearl activated a hardfork at block 98900. From that height the network derives share noise differently, and any miner still running the old derivation produces invalid shares — full hashrate on screen, nothing accepted by the pool. This release implements the new derivation. If you mine Pearl, update now; older versions will not earn on this chain Nothing else changed. Kernels, hashrate and power draw are identical - the fix touches only how the share seeds are derived. Confirmed live on Ada and Blackwell cards with zero rejected shares

CMP 50HX - around +19.5% hashrate at the same pinned clock, verified on an eight-card rig against a live pool. Power draw rises by roughly the same amount, so this is a mining-throughput win rather than an efficiency one. CMP boards only - every other Turing card keeps the kernel it had CMP 170HX (and A100) - around +17% hashrate and about +15% efficiency against 1.5.8, measured release-against-release on the same card at the same pinned clock. Part of that is a kernel improvement and part is a packaging fix: 1.5.8 shipped without the block-path kernels for this architecture, so those cards were quietly running the slower path CMP 90HX - micro-optimisations for stability on its mining path, and a full eight-card validation run on this build Fixed: on 1.5.5-1.5.8 a Pascal card (P102-100, P104-100) or a Tesla V100 connected, looked healthy and mined nothing - 0.00 H/s on that card, with no error shown. On a mixed rig the other cards kept working, which is why it went unnoticed for four releases Fixed two cases where the miner could run at full hashrate while submitting nothing valid - it now refuses to start and says what to do, instead of failing silently

Fixed the remaining cause of HiveOS dashboard stats freezing after many hours on multi-GPU rigs. Mining and share submission were never affected — only the dashboard readout. If the stats file cannot be written, the miner now says so once instead of failing silently.

Fixed a file-descriptor leak that could freeze HiveOS dashboard stats on rigs with many GPUs. Mining and share submission were never affected — only the dashboard readout. Fixed the clock-offset indicator overflowing its column in the dashboard.

The Turing search kernel has been rewritten: ~30% more hashrate and ~20% better efficiency on the 20 series — 2060, 2070, 2080, 2080 Super, 2080 Ti and their Super variants. The exact gain varies from card to card. The 20 series now fully supports the Pearl rank-penalty softfork. Verified on live pools: 100% share acceptance, zero rejects. RTX 30, 40 and 50 are untouched — their kernels are bit-for-bit identical to 1.5.4. Drop-in upgrade: replace the binary, keep your configuration, wallets and pools.

Pearl activates a softfork at block 96251: the bound a share must meet is scaled by 128/rank. Releases before 1.5.4 keep mining at the old rank and earn half as much for the same hashrate — an eighth for anyone who was at rank 1024. Nothing in the console shows it; only the payouts differ. 1.5.4 mines at the new rank on Pascal (P104/P106), RTX 30, RTX 40, RTX 50 and A100, and collects the full reward. Nothing to configure. Turing, Volta and Hopper stay on the previous rank: they keep working after the fork, at a lower reward. Raw hashrate is 1-2% lower by design, since the new rank costs a little more work per share. Income still doubles. Verified on live pools across the supported cards.

Pearl draws less power at the same hashrate. Measured on the same cards at the same clocks against v1.5.2: about 4% better efficiency on RTX 40, about 2.4% on RTX 30. On a rig running into a power limit, that headroom comes back as clocks. Cards under a power limit now get a profile chosen for that case. The miner reads the limit from the driver and decides per card, so a mixed rig no longer has to compromise. FORGE_EFF=1 or FORGE_EFF=0 forces the choice. Fix: 8 GB cards on Windows were losing about 4% hashrate. Any card with less than 9.5 GB was assumed not to fit and put into a reduced mode. The miner now measures what actually fits on the card instead of guessing from the label. Fix: a custom profile could leave a rig at full displayed hashrate while submitting nothing. Profiles the proof format cannot carry are now refused at start with a clear message instead of failing silently. Fix: a launcher that set the pool through the environment could hang on connect. The standard pool dialect was chosen only when the pool arrived on the command line, so the same pool written the other way left the miner connected and waiting for work with nothing on screen. The standard dialect is now the default whenever a pool is configured at all.

The Windows download is sorted by coin. One folder per coin, ready-made .bat files inside. Every pool region is now covered — Global, EU, RU and US — each with a plain and an encrypted (-SSL) variant. Lighter binary. Cleaned out leftovers that no longer served a purpose. Better documentation. Each download now carries a proper guide for its platform — Windows, Linux and HiveOS — in English and Russian. --help lists every coin and option the miner actually supports. Fix: the .bat files now launch the miner by full path, so they also work on machines where Windows is configured not to run programs from the current folder.

Pearl (PRL) is 5-6% faster on RTX 30 and RTX 40. Measured live on a rig of 8x RTX 3060 Ti and a rig of 4x RTX 4070 Ti. The periodic dips are gone. The hashrate graph is flat now: the spread between readings dropped about 3x on RTX 30 and about 5x on RTX 40. Mixed rigs. Each card now runs its own best settings instead of the whole rig sharing one compromise, so a single older card no longer holds the rest back. Fix: on 8 GB cards, turning the hashrate optimization off could abort the miner at startup.

New coin: Xelis (XEL) — mine XEL at just 1% dev fee, the lowest on XelisHash right now. Qubitcoin (QTC) — noticeably faster on RTX 30/40. Wider hardware: now runs on GTX 10-series (Pascal) and Tesla V100 (Volta) across every coin.

BTX speed: +16% on RTX 30-series, +20% on RTX 40-series, +11% on RTX 50-series. Validated on live pools with zero rejected shares. BTX on older cards: added support for Pascal (P106-100, P104-100) and Turing, including CMP 50HX. Pearl on RTX 50-series: +5.6% — 219.66 TH/s at 609.69 GH/W on an RTX 5080. Smaller package: the download is about 3x smaller — roughly 10 MB instead of 30 MB. Faster to pull onto every rig.

BTX: about +12% hashrate on RTX 30-series (3060 Ti / 3070 / 3070 Ti / 3080 / 3080 Ti and similar). BTX: RTX 50-series is faster too (about +5%). Zero rejects, identical shares — pure throughput gain.

This release improves Pearl (pearlhash) share reliability — a rare edge case in the share proof is fixed, so shares stay consistently accepted. It also removes the experimental efficient mode.

New optional --efficient mode for Pearl. It draws noticeably less power for nearly the same hashrate - on RTX 40 (Ada) cards, about 10-13% lower power for about 3% lower hashrate. The result is better energy efficiency (more hashrate per watt). For miners optimizing electricity cost or running power-limited rigs. The default mode is unchanged and still targets maximum hashrate. Optimized for RTX 40 (Ada). On other GPUs the flag safely falls back to the normal maximum-hashrate mode - no error, no change in behavior.