Skip to content

GPU Task Manager v0.4.5

Latest

Choose a tag to compare

@github-actions github-actions released this 21 Nov 21:19
· 4 commits to main since this release

Upgrade: uv tool upgrade gputaskman or pipx upgrade gputaskman. Portable builds: download the archive for your platform below and replace the old folder. Check what you have with gputaskman --version.

What's Changed

  • Preserve unknown WSL host counters instead of reporting zeros by @shanselman in #115
  • Mark failed macOS Metal client queries as unavailable by @shanselman in #113
  • Report NVML process-list failures instead of treating them as empty by @shanselman in #121
  • Invalidate WSL context ownership on immediate PID reuse by @shanselman in #118
  • Handle malformed runtime API responses by @shanselman in #120
  • Keep same-name GPU adapters in separate summary runs by @iamdatvt94 in #91
  • Serialize unobserved model RAM residency as unknown by @shanselman in #119
  • Show process memory becoming known or unknown in text diffs by @shanselman in #116
  • test: cover summary exclusion of inferred VRAM by @shanselman in #117
  • Scan all eligible processes in one-shot collection by @shanselman in #123
  • Label Intel GPU clock charts as GT by @shanselman in #124
  • Keep unknown macOS footprints unattributed by @shanselman in #126
  • Avoid private DRM attribution without client identity by @shanselman in #122
  • Test PID reuse during an in-flight macOS footprint read by @shanselman in #125
  • Avoid stale model logs as current allocations by @shanselman in #129
  • Preserve Windows fallback counter provenance by @shanselman in #130
  • Avoid certain container PID matches by count by @shanselman in #127
  • Test Linux NVML aggregation through Collector by @shanselman in #128
  • Restore container runtime API discovery without claiming ownership by @shanselman in #131
  • Hide Microsoft Basic Render Driver adapters by @shanselman in #135
  • Read Intel Xe package temperature from GPU hwmon labels by @Prada369 in #132
  • Don't show Windows counter notes for Linux and macOS compositors by @ardazeybek in #87
  • Resolve llama-server's relative model_path and --log-file against its cwd by @ardazeybek in #86
  • Cap GGUF header strings instead of reading whatever length the file claims by @ardazeybek in #88
  • Show compressed graphics peaks in summaries without adding them to VRAM by @kearns2000 in #89
  • Document TEXTUAL_THEME for choosing TUI colors by @Sergi-us in #90
  • feat: report vLLM prompt and generation token rates from /metrics by @raravind007 in #85
  • Keep runtime API probes on the original listener by @shanselman in #114
  • Keep hidden NVML process VRAM out of measured totals by @shanselman in #112
  • Show the version: --version, TUI subtitle, upgrade notes in releases by @shanselman in #136

New Contributors

Full Changelog: v0.4.4...v0.4.5