Uminer GPU v0.1.14 (Testing Release)

Uminer GPU v0.1.14

Testing-stage release published under an explicit qualification exception. Linux only.

English

Highlights

• Replaces per-GPU pool connections with one process-level user-pool session shared by all GPUs.
• Adds bounded reconnect/failover, submit routing, panic isolation and half-failed-pool recovery.
• Adds a 120-second whole-rig/per-GPU operations panel with hashrate, power, clocks, memory,
  temperature, shares and host-resource totals.
• Adds CSD/BTX developer-fee READY gating: the user GPU is not switched until a trusted Fee route
  is authorized and has a valid job.

Developer Fee

• BTX: 2%.
• CSD: 3%.
• The rates are compiled per coin and cannot be reduced through CLI.
• The user pool remains connected during Fee work. A route failure does not pause user mining and
  missed Fee time is not recovered.

Reliability Fixes

• Verifier and GPU-worker panics can no longer leave a miner reporting ONLINE while silently
  discarding all shares.
• Three consecutive submit timeouts or ten consecutive verified-share rejections trigger a pool
  reconnect instead of indefinite zero revenue.
• Fee windows have a monotonic watchdog and always restore the device pause state.
• Unhealthy GPUs are isolated, excluded from rig-wide minimum-rate restarts and periodically
  reprobed; the process exits only when no healthy GPU remains.
• CSD collects and submits multiple hits from one CUDA batch instead of retaining only the first.
• Temporary authorization rejection uses cooldown and retry instead of permanently disabling all
  pools.

Compatibility

• Linux x86_64; glibc 2.30 or newer.
• NVIDIA driver 535+ for sm_75/80/86/89; driver 580.95+ for sm_120.
• Embedded native SASS for sm_75, sm_80, sm_86, sm_89, and sm_120.
• Plain and TLS Stratum URLs for BTX and CSD; transport follows the configured URL.
• Pearl and Windows are not included.

Qualification Disclosure

This version is being published directly for testing by explicit approval. Automated source,
workspace, architecture, active-fee and package checks passed. Production signing-control assets,
the complete multi-machine matrix, 12-hour Fee accounting and 24/72-hour soak evidence are waived
for this testing release and remain required before a production-qualified claim.

繁體中文

版本重點

• 將每卡獨立礦池連線改為全進程多 GPU 共用一條使用者連線。
• 新增有界重連/備援池、submit 路由、panic 隔離和礦池半故障自愈。
• 預設每 120 秒輸出整機/逐卡運維面板。
• Fee 必須在受信任路由已授權且取得有效 job 後才切換 GPU。

開發者費用

• BTX：2%。
• CSD：3%。
• 費率按幣種固化，CLI 不可下調。
• Fee 執行期間使用者礦池保持連線；Fee 路由失敗不暫停使用者挖礦，也不追補。

穩定性修正

• verifier/worker panic 不再造成顯示 ONLINE 但靜默丟失全部 share。
• 連續 3 次 submit 超時或 10 次已本地驗證 share 被拒絕時主動重連。
• Fee 時間窗納入單調 watchdog，任何退出路徑都會恢復卡狀態。
• unhealthy GPU 按卡隔離、不進入整機 min-rate 重啟口徑，並會週期重試。
• CSD CUDA 一個 batch 內的多個命中會全部逐個提交。
• auth 短暫拒絕改為冷卻後重試。

相容性

• Linux x86_64，glibc 2.30 或更新版本。
• sm_75/80/86/89 需 NVIDIA 535+；sm_120 需 580.95+。
• 內嵌 sm_75、sm_80、sm_86、sm_89、sm_120 native SASS。
• BTX/CSD 都可使用明文或 TLS Stratum URL，傳輸由 URL 決定。
• 不包含 Pearl 與 Windows。

驗收披露

本版根據明確特例直接作為測試版發佈。源碼、workspace、架構、active Fee 與驗包自動門禁
已通過；生產簽名控制資產、完整機型矩陣、12 小時 Fee 對帳與 24/72 小時長穩在本測試版中
經特例豁免，日後聲明生產級通過前仍必須補齊。