Skip to main content
Participant
August 7, 2026
Question

AI Denoise never completes on Intel Arc 140T (Core Ultra) — GPU at 100% compute, progress stuck — worked in ~1–2 min on earlier LrC versions — LrC 15.4.1 + Intel 32.0.101.8860

  • August 7, 2026
  • 4 replies
  • 41 views

AI Denoise consistently fails to complete on my machine. This is 100% reproducible and has persisted across multiple Lightroom Classic and Intel driver versions, including the current pairing of both.

System:

  • Lightroom Classic 15.4.1 [202606201310-b9f148a4], Camera Raw 18.4
  • Windows 11 Home 26200
  • Intel Core Ultra laptop, 16 logical cores, 32 GB RAM
  • Intel Arc 140T integrated GPU (16 GB shared), driver 32.0.101.8860 WHQL (current, verified in Device Manager)
  • Files: Fujifilm RAF (X-Trans)

Symptom:

  1. Click Denoise. The background progress indicator (top-left) immediately shows roughly 10% progress, then never advances — waited well over an hour.
  2. The photo becomes locked: "AI processing is in progress for this photo. Editing is temporarily unavailable."
  3. Task Manager shows the Arc 140T near 100% compute utilisation for the entire time, so the GPU is actively executing something — it just never finishes.
  4. Exiting Lightroom during this state leaves orphaned Lightroom processes that must be killed in Task Manager.
  5. This is a regression. On earlier LrC versions (a couple of months ago, modal Denoise dialog era), Denoise on this same machine with the same type of RAF files completed in roughly 1–2 minutes when it ran. It was already unstable and frustrating to use - it would regularly freeze mid-run, requiring Lightroom to be killed in Task Manager, and typically took a few attempts per photo - but successful runs at least completed in minutes. So the Arc 140T is demonstrably capable of executing this workload in minutes. In the current version (and possibly the one immediately before it), Denoise never completes at all: the GPU computes at 100% indefinitely.

Diagnostics already done:

  • Windows System event log (filtered Critical/Error/Warning): no Display/TDR events, no Event 4101, nothing GPU-related during Denoise runs.
  • Reliability Monitor: no LiveKernelEvent 141, no hardware errors on test days.
  • So this is not a driver crash/reset loop — the GPU computes indefinitely without completing. This matches Intel community reports of Denoise on Arc taking 12–92 minutes per photo (e.g. Arc A310 thread).
  • Mains power, Windows "Best performance" power mode: no change.
  • Tried with Preferences > Performance > Use Graphics Processor set to Auto, and then also to Custom with "Use GPU for image processing" unticked, and also Off: no change - as documented elsewhere, Denoise ignores this preference and uses the GPU regardless. There is currently no supported way to run Denoise on CPU, and DisableGPUInference.txt explicitly does not cover Denoise per the ACR GPU FAQ.
  • Timed test (7 Aug 2026): Denoise started on a single RAF at ~21:05. As of 23:00+ it had still not completed — over 1 hour 50 minutes on one photo, progress indicator unmoved from its initial position throughout. So far I have not had a denoise complete, it appears to be processing, but without progress, maybe even in an infinite loop, can’t tell. 
  • GPU utilisation logged with typeperf (60-second samples, one full hour): the Arc 140T's Compute engine sat at 95–100% in every single sample, attributable to one process (Lightroom), with zero dips, pauses, or resets - the GPU is continuously executing without ever finishing. Log CSV available on request.
  • CPU comparison test (Arc disabled in Device Manager, forcing CPU denoise) to follow in a comment.

Related reports:

Requests:

  1. Please investigate what changed in the Denoise/AI inference path in recent LrC / Camera Raw releases with respect to Intel Arc. Since earlier versions completed in 1–2 minutes on this exact hardware and the current version never completes, this looks like a software regression rather than a hardware capability limit. I am paying a monthly fee, and expect this to work on what is standard and powerful hardware. There was never any indication on purchase that intel Arc is not supported. 
  2. Please coordinate with Intel on the Arc inference path — Intel has publicly stated this requires joint work between the companies.
  3. Please provide a supported option to run Denoise on CPU. On this hardware the CPU would complete in minutes; the GPU never completes, and there is no user-accessible fallback short of disabling the GPU in Device Manager.

Full LrC System Info available on request, but I think you’ll agree I’ve provided detailed info above, so please investigate as a priority.  

    4 replies

    Community Expert
    August 8, 2026

    >driver 32.0.101.8860 WHQL

    The most recent driver is 32.0.101.8864. https://www.intel.com/content/www/us/en/products/sku/241598/intel-arc-b580-graphics/downloads.html 

    You might want to try updating that first. Windows update rarely has the most recent drivers for GPUs and Lightroom is very sensitive to GPU drivers as you might know.

    DaddyB99Author
    Participant
    August 8, 2026

    Thanks - tested 32.0.101.8864: identical behaviour, 30 minutes at 100% GPU compute, no progress - progress bar stuck at about 10%, and no warnings or errors in event viewer. So the current and previous WHQL drivers both exhibit this same problem. Really hoping someone from Adobe can respond and take this up, as there will be a significant community of people on Intel Arc graphics that are impacted, unable to use the features they are paying for. 

    Community Expert
    August 9, 2026

    Hopefully somebody else with similar hardware can weigh in too to see if this is reproducible. I am responding just to get this back up to the top to see if Adobe folks pick up on it. You might want to submit a bug report instead of a question on this referring to this thread. I can’t do turn this thread into a big report unfortunately.

    DaddyB99Author
    Participant
    August 8, 2026

    Follow-up with completed test results.

    GPU endurance test (7–8 Aug): Denoise started at 21:05 BST on a single RAF. GPU utilisation logged via typeperf at 60-second intervals: the Arc 140T Compute engine ran at a continuous 95–100% (single process: Lightroom) in every sample for 2 hours 9 minutes, with the progress indicator never advancing beyond its initial position. No dips, no resets, and no Display/TDR/LiveKernel events anywhere in the Windows event logs across the entire run. The system then slept on its normal power timer; an overnight Windows maintenance restart cleared the job. The photo was not denoised. Result: Denoise on GPU did not complete in 2h09m of continuous 100% GPU compute.

    CPU comparison test (8 Aug): Disabled the Arc 140T in Device Manager (the only way to force CPU processing, since Denoise ignores the Lightroom setting: Use Graphics Processor preference). Same photo, same catalogue: Denoise completed successfully in ~5 minutes at 50–80% CPU utilisation. This is not a viable workaround, however - I can’t disable my GPU every time I want to execute any Lightroom AI function. 

    Summary - same machine, same type of file:

    • Earlier LrC (approx. 2 months ago), GPU: 1–2 minutes (unstable — frequent freezes requiring Task Manager kill — but fast when it ran)
    • LrC 15.4.1, GPU: does not complete (>2h09m at 100% GPU compute, zero progress)
    • LrC 15.4.1, CPU (GPU disabled at OS level): ~5 minutes, completes reliably

    So in the current release, the GPU inference path on this hardware is at minimum ~25× slower than the CPU on the same laptop, and I have not had an instance of it actually completing, so it might even be in an infinite loop for all I can tell.

    Repeating the two requests: (1) please investigate what changed in the Denoise/AI inference path for Intel Arc in recent releases, and provide a permanent solution; (2) please provide a supported option to run Denoise on CPU - disabling the entire GPU in Device Manager cannot be the intended workaround, as it also disables GPU display acceleration for the whole system.