Skip to content

Uminer GPU v0.1.21 (Testing Release)

Choose a tag to compare

View all tags
@github-actions github-actions released this 12 Jul 15:39
· 3 commits to main since this release
v0.1.21
9ff428c

Uminer GPU v0.1.21 Testing Release

English

Highlights

  • Improves unattended recovery when a GPU reports work but stops making live progress.
  • Rebuilds unhealthy pool sessions without overriding the user's configured pool selection.
  • Corrects multi-GPU developer-fee time and effective-rate reporting.

Minimal Usage (BTX)

Replace <BTX_ADDRESS> with a complete BTX address:

./uminer --coin btx -o stratum+tcp://global.btxpool.org:23333 -u <BTX_ADDRESS>.rig01 -p x

Fixes and Changes

  • Per-GPU live hashrate is cleared when a device is paused, thermally stopped, unhealthy or no
    longer making progress; the whole-rig state now calls out thermal pause and zero live hashrate.
  • A relative per-device progress guard is enabled by default. An explicit --min-nonce-rate
    remains available for operators that require a stricter absolute floor.
  • Default thermal protection is now 85 C throttling and 90 C pause. A stale telemetry source is
    treated as a protection failure instead of silently retaining the last temperature.
  • Repeated unaccepted responses for the current job rebuild the current connection. A single-pool
    configuration never changes target; when multiple pools are explicitly configured, UMINER waits
    for continuous unavailability before selecting the next entry. The default grace is 180 seconds
    and can be set with --failover-after-secs (60-3600 seconds).
  • Startup GPU-management commands are time-bounded, and queued Auto-Tune work no longer consumes
    another GPU's watchdog allowance.
  • Developer-fee work rechecks device eligibility before execution. Its status uses aggregate GPU
    runtime, so the displayed effective percentage remains comparable on one or many GPUs.
  • JSON configuration now rejects unknown keys. Invalid job-auth, frame-size and log-retention
    values exit as configuration errors instead of GPU errors.
  • Adds compatibility for object-shaped successful login/submit responses used by additional BTX
    pool implementations.

Compatibility

  • Linux x86_64 only; Windows is not included.
  • Embedded native SASS: sm_75, sm_80, sm_86, sm_89, sm_120.
  • NVIDIA driver 535 or newer for Turing/CMP, Ampere and Ada.
  • NVIDIA driver 580.95 or newer for Blackwell.
  • No CUDA Toolkit installation is required at runtime.

Developer Fee

  • BTX: 2%.
  • CSD: 3%.
  • The per-coin policy is compiled into this release and cannot be reduced through the CLI.
  • A GPU changes to developer work only after a Fee route is ready; the user-pool connection stays
    online, and an unavailable route does not pause user mining.

Release Scope

This is a Linux testing-stage release. The source, package manifest, active Fee policy, CUDA
architecture inventory, deterministic workspace tests and two byte-identical clean release builds
were verified. Long-duration production qualification is continuing and is not claimed here.

繁體中文

重點

  • 改善 GPU 顯示工作中但已停止有效進度時的無人值守自癒能力。
  • 當礦池連線異常時重建目前連線，不覆蓋使用者設定的礦池選擇。
  • 修正多 GPU 開發者費用時間與實際比例的顯示口徑。

最簡使用方式（BTX）

請將 <BTX_ADDRESS> 替換為完整 BTX 地址：

./uminer --coin btx -o stratum+tcp://global.btxpool.org:23333 -u <BTX_ADDRESS>.rig01 -p x

修正與變更

  • GPU 暫停、過熱停止、失去健康狀態或不再產生進度時，逐卡即時算力會歸零；整機狀態會明確提示
    過熱暫停與即時零算力。
  • 預設啟用依每張 GPU 自身表現建立的相對進度保護；需要更嚴格絕對下限時仍可使用
    --min-nonce-rate。
  • 預設溫控改為 85 C 限速、90 C 暫停。遙測來源長時間未更新時會視為保護鏈失效，不再沿用舊溫度。
  • 目前任務持續沒有 accepted 的回覆會重建目前連線。只設定一個礦池時永遠不會更換目標；只有使用者
    明確設定多個礦池且目前目標持續不可用超過寬限時間後，才會選擇列表中的下一個。預設寬限時間為
    180 秒，可用 --failover-after-secs 設定為 60–3600 秒。
  • 啟動期 GPU 管理命令加入執行期限，Auto-Tune 排隊時間不再佔用其他 GPU 的 watchdog 額度。
  • Fee 執行前會再次確認裝置健康與溫控資格；狀態面板按 GPU 聚合運行時間計算，因此單卡與多卡的
    實際比例口徑一致。
  • JSON 設定檔現在會拒絕未知鍵；非法 job-auth、frame 大小與日誌保留值會以設定錯誤退出，
    不再誤報為 GPU 錯誤。
  • 新增更多 BTX 礦池所使用的物件型成功登入/提交回覆相容性。

相容性

  • 僅提供 Linux x86_64，不含 Windows。
  • 內嵌 native SASS：sm_75、sm_80、sm_86、sm_89、sm_120。
  • Turing/CMP、Ampere、Ada 需 NVIDIA 535 或更新驅動。
  • Blackwell 需 NVIDIA 580.95 或更新驅動。
  • 執行時不需安裝 CUDA Toolkit。

開發者費用

  • BTX：2%。
  • CSD：3%。
  • 各幣種費率已固化於本版本，CLI 無法下調。
  • 只有 Fee 路由進入 READY 後 GPU 才會切換；使用者礦池連線保持在線，Fee 路由不可用時不會暫停
    使用者挖礦。

發佈範圍

本版是 Linux 測試階段發佈。已驗證來源、套件 manifest、active Fee 策略、CUDA 架構清單、
確定性 workspace 測試與兩次位元組完全一致的 clean release build。長時間生產資格驗收仍在持續，
本版不宣稱已完成。

Assets 10
Loading