github NortheBridge/luminalshine 26.08.0-beta.7
LuminalShine 26.08.0-beta.7

pre-release2 hours ago

LuminalShine 26.08.0-beta.7

Changes since 26.08.0-beta.4 (26.08.0-beta.5 and beta.6 were staged but
never published; their contents ship here).

Fixed — stability

  • Recurring host crashes traced to a use-after-free in the pairing
    flow.
    Submitting the PIN form twice — or retrying an abandoned
    pairing — freed the pairing session record mid-operation and kept
    writing through it, corrupting the heap. The pairing session table is
    now protected by a lock and the freed record is never touched.
    (#104)
  • Clients warned "HEVC / HDR / YUV 4:4:4 is unavailable" at stream
    start
    (macOS most visibly) whenever the service had started before
    logon — i.e. after every boot. The startup encoder probe deferred
    until the desktop was ready and then never retried, so the server
    advertised H.264-only until the first stream, which also paid the
    entire probe cost mid-launch. The probe now retries until the desktop
    is ready, so full codec support is advertised before any client
    connects and first launches start much faster.
    (#117)
  • Valid admin password rejected mid-uptime. Config hot-applies
    reset the in-memory credential record's key-derivation fields.
    (#95)
  • Crash when capture started against a virtual display that never
    activated
    — now falls back cleanly instead of corrupting the heap.
    (#97)
  • Zombie driver sessions on displays Windows never surfaced are now
    destroyed instead of stacking.
    (#109)
  • Spurious "crash" dumps on service stop when Windows Explorer was
    dead: the tray thread now observes shutdown instead of pinning the
    exit path until the force-shutdown watchdog fired.
    (#114)
  • Slight lag when starting to move the cursor, and the black box
    around the text-caret (I-beam) cursor
    .
    (#98)
  • About page "Driver version: -" and empty Troubleshooting readouts
    for LuminalVGD.
    (#94)
  • Faster session start — third-party integration work no longer
    runs on the stream-announce path.
    (#99)

Fixed — GPU-reset (TDR) behavior

After a GPU hang whose Windows recovery failed, previous versions
misdiagnosed the dead display stack in an endless loop for as long as
the machine stayed up. Now:

  • Failures are classified by evidence, and the classifier itself was
    battle-tested against a live incident.
    A device-creation failure is
    cross-checked against Windows' display API; a machine-wide failure is
    declared only on the confirmed wedge signature (verified twice, with
    a settle delay), announces "Display stack down — reboot required"
    in the log, session refusals, and the Troubleshooting page — and the
    verdict now clears itself automatically if the stack recovers.
    (#105,
    #112)
  • The host no longer restarts itself into a dead display stack to
    reclaim leaked encoder resources — the restart waits for the stack to
    return (and never interrupts an active WebRTC session).
    (#113)
  • The topology lifeboat: the moment a GPU hang is detected
    (~100 ms), a physical display is activated alongside the virtual one
    so Windows always has a hardware-backed display to recover onto.
    (#110)
  • The Virtual Display Driver card and restart button actually work
    now
    , targeting LuminalVGD directly with a working two-level
    restart.
    (#106)
  • The display helper escalates instead of polling silently: a
    persistent display-stack failure produces one loud reboot-required
    message and an orphaned helper exits within 5 minutes.
    (#107)
  • Recovery hardening from the 0x10E incident: encoder stalls marked
    within ~100 ms, capture reinitializes after a reset, the service
    self-configures restart-on-failure, GPU resets charted on Session
    Details.
    (#101)

Added

  • HDR peak brightness (nits) slider in the LuminalVGD Control
    Panel: 0–1000 nits (790–1000 is the HDR zone), default 800, saves
    automatically and applies to the next streaming session. The virtual
    display's EDID advertises the chosen peak to Windows and HDR-aware
    games. Requires the bundled driver build 15.
    (#118)
  • Driver updates no longer need a reboot. The service now owns the
    virtual display device: fresh installs create it at service start,
    and upgrades rebind the new driver automatically (in place when
    possible, recreate when not) — no more WGC fallback until the next
    restart of Windows.
    (#116)
  • Crash self-reporting. Any crash writes a minidump to a known
    location and logs the exception, module, and dump path.
    (#108)
  • Driver flight recorder. The installer registers a small
    boot-rotating ETW trace of the LuminalVGD driver, and crash bundles
    automatically include it — GPU-reset incidents are now diagnosable
    from the driver's own point of view.
    (#115)
  • Redesigned Session Details view: status pills, identity panel,
    multi-range charts with JSON export, full-screen last-10-sessions
    view.
    (#100)

Changed

  • Virtual-display defaults: backend Automatic, mode Per client,
    layout Exclusive, activate On, revert-on-disconnect On, refresh
    doubling On.
    (#93)

Internal

  • Config-consistency suite gates in CI
    (#96);
    vendored web server moved to a hardened NortheBridge fork
    (#104);
    driver pin → v0.1.0-alpha.5
    (#119).

Notes

  • Bundles LuminalVGD v0.1.0-alpha.5 "Build 15" (0.1.0.15):
    proto 0.4 with host-configurable HDR peak luminance, plus build 14's
    TDR duck-out (the driver removes its virtual monitors during a
    GPU reset so Windows' recovery can never wait on them, restoring them
    with identity intact on recovery) and build 13's cold-boot activation
    and I-beam cursor fixes.
  • Requires Windows 11; 24H2+ for full HDR support.

Don't miss a new luminalshine release

NewReleases is sending notifications on new releases.