[x86-64] Dolby Vision w/FEL + Full 3D for PC (Intel, AMD & nVidia)

    Quote from hadesdevls

    Hi ashlar, CroqueMr,

    I'm glad to collaborate, with CroqueMr and with anyone else working on Dolby Vision on PC. All of the Hades code is on GitHub, so CroqueMr, use whatever helps your build, and ask me if you want to know why something is done the way it is. Hades started from your work, and the README says so.

    That said, Hades is a hobby project I started for my own setup, and much of it has nothing to do with Dolby Vision. It has privacy and safer defaults, like hiding the internal disks when booting from USB, Samba off unless you turn it on (and then with a password), no update server that could replace the build, and logs kept on the device unless you choose to upload them. There's also a Windows installer. That probably doesn't fit your build, so I'll keep working on Hades in parallel and share what works.

    On bilinear: I agree it costs fidelity, which is why Hades only steps down when it has to. It isn't a mode you choose. It starts only when the GPU is actually missing refreshes, goes back up as soon as frames fit again, is remembered per GPU and driver, and can be turned off. A GPU that keeps up gets the full picture, bit-exact, so if the N100 keeps up it never steps down there either.

    That full picture is also more faithful in RC2. It now follows the normative composer of ETSI GS CCM 001 and matches an integer reference of the spec bit for bit in every case I tested. On the profile 7 FEL stream I measured, libplacebo's reshaping, which mpv and the PC builds use (Hades too, until this release), differs from it by 4 to 15 codes out of 4095 on almost every pixel and by hundreds at high-contrast edges. If you want it, the composer and the exact Lanczos weights are libplacebo patches in the repo, and the chroma siting fix is a small change in the bridge.

    I noticed the different adjustment levels you implemented using bilinear filtering. Unfortunately, a large part of what you've done had already been explored on my side. If you look at the Fast mode in R1.0.0.Beta3, it follows a very similar approach, with comparable impacts in image fidelity, ( chrominance and high-contrast edges)

    Performance speaking, your build seems to be roughly on par with Beta 3 on the N100, where it falls back to your Level 2. I believe R1.0.0.Beta3 might preserve slightly better image fidelity, but that still needs to be confirmed through proper comparisons.

    Regarding the project itself, my goal with the Dolby Vision build in my repository isn't to integrate every possible feature, particularly things like security enhancements that fall outside the scope of the DV engine.

    That said, I'd really encourage you to contribute directly to the main repository when working on the DV engine itself. PRs would allow everyone to benefit from your improvements and help us avoid duplicating efforts, you can push PRs and I will perform all non-regression tests. Of course, you're absolutely free to maintain your own fork for additional features, such as your security enhancements.

    I think this would be a much more productive approach for everyone, and I'd be happy to see your contributions integrated into the main project.


    I’m currently finalizing the next release, it doesn’t improve performance for the native DV circuit, it features a complete rewrite of the other circuits (DV Enhanced etc...) and incorporates some of your elements, such as Nvidia support. If all goes well, it will be pushed tomorrow.

    Quote from Creator44

    CroqueMr , did my log about the DV freeze and black screen help find the issue?

    Unfortunately not so much. We can see the issue, but we can't identified the root cause. I ran an intensive AMD stress test on my 7430U and fixed a few minor issues, which might help with yours. In the meantime, I’ve adjusted the logs to capture more detailed information so we can pinpoint the root cause.

    Edited 2 times, last by CroqueMr: Merged a post created by CroqueMr into this post. (October 16, 2026 at 9:11 PM).

    I installed R1.0.0-Beta3 on my backup Libreelec box (GK3V Pro + LG C9). Still a bit flaky for everyday use but its getting there!

    For some reason, the "All Layers Test" was showing in the wrong aspect ratio. Looked like cinemascope. Also, as has been previously mentioned, the on-screen menu and Player Process/Player Debug displays were cut off.

    I did install over the top of standard Libreelec v13, not sure if that was a good idea?

    Quote from CroqueMr

    Unfortunately not so much. We can see the issue, but we can't identified the root cause. I ran an intensive AMD stress test on my 7430U and fixed a few minor issues, which might help with yours. In the meantime, I’ve adjusted the logs to capture more detailed information so we can pinpoint the root cause.

    Do you want a test clip?

    hadesdevls a hardware report for RC2 on a discrete AMD card, since "other AMD" is still untested in your list.

    ASUS NUC 12 with a Radeon RX 7600 (Navi 33), HDMI straight to a Panasonic TV (TV-led DV), clips played from local storage:

    • P7 FEL, two test clips (CM v2.9 and CM v4.0): native DV, Dolby Vision logo on the TV, the FEL text is visible
    • P7 MEL: detected as MEL, native DV
    • P8.1 and P5 (59.94): native DV
    • 240 of 240 frames per 10 s at 4K 23.976, 600 of 600 at 59.94, no step-down, about 5 ms GPU time per frame
    • Start and stop each in one mode change
    • Connector during DV: max bpc 8, full-range RGB, no HDR metadata, DVBRIDGE_REFUSED 0 (DVBRIDGE_LINK_FORMAT also reads 0)

    Two things I noticed:

    • With a settings window over the video, every frame is composed with the GUI and output drops to about 16 fps (43-47 ms GPU time per frame). Closing it brings back 240 of 240.
    • The log repeats "DVBridge libplacebo: Trying to call pl_opengl_unwrap on a pseudo-texture", with no visible effect.

    Happy to send the log.

    Quote from Creator44

    Do you want a test clip?

    Not necessary it will reflect the log :) I will ask you to send me the new log when the next release will be pushed to capture more informations. Even if your issue is solved with the next patches.

    Quote from Glennza

    I installed R1.0.0-Beta3 on my backup Libreelec box (GK3V Pro + LG C9). Still a bit flaky for everyday use but its getting there!

    For some reason, the "All Layers Test" was showing in the wrong aspect ratio. Looked like cinemascope. Also, as has been previously mentioned, the on-screen menu and Player Process/Player Debug displays were cut off.

    I did install over the top of standard Libreelec v13, not sure if that was a good idea?

    Maybe it came from a small issue that I already patched in my internal release. It would be great if you could try again on the new release I will push today and send me the Kodi logs file (kodi.log)

    Edited once, last by CroqueMr: Merged a post created by CroqueMr into this post. (October 17, 2026 at 9:51 AM).

    Quick update regarding beta4. I’ve encounter a strange issue: the "fast" mode (designed to help the N100 handle certain render modes) reduce GPU workload but also reduce the headroom on AMD CPUs and increase the drops if your iGPU can not handle the workload. I want to fully understand why this is happening before releasing the build, so publication might be pushed back by a day or so.

    Good news: I rebuilt a new pipeline from scratch called "Dolby Vision Master Reference". It makes zero compromises on image quality with a focus on being bit-perfect. The N100 handles this circuit flawlessly, even on heavy FEL remuxes or 60 FPS Dolby Vision ISO :)

    Thanks for testing it on the N100, and I'm glad the NVIDIA work helps your next release. Good luck with it.

    If you still have the log from that N100 run, I'd like to see it. Knowing why it stepped down to level 2 there would help me tune it.

    On fidelity, I agree with you that it needs proper comparisons. The way I measure is written up in the descriptions of the libplacebo patches in the repo (9930 for the composer, 9940 for the step-down levels): frames of a profile 7 FEL stream drawn on a software renderer, the composer compared with two separate transcriptions of the spec, and each step-down level with the full-quality picture. Happy to compare on the same frames whenever it's useful.

    And when I have something in the DV engine that could help your build, I'll send it as a PR or point you to it. Hades will keep going in parallel for the things outside the DV engine, and anything in it is yours to use.

    I'm away from home until Tuesday, so I'll follow up then.

    Quote from cinemaONE

    hadesdevls a hardware report for RC2 on a discrete AMD card, since "other AMD" is still untested in your list.

    ASUS NUC 12 with a Radeon RX 7600 (Navi 33), HDMI straight to a Panasonic TV (TV-led DV), clips played from local storage:

    • P7 FEL, two test clips (CM v2.9 and CM v4.0): native DV, Dolby Vision logo on the TV, the FEL text is visible
    • P7 MEL: detected as MEL, native DV
    • P8.1 and P5 (59.94): native DV
    • 240 of 240 frames per 10 s at 4K 23.976, 600 of 600 at 59.94, no step-down, about 5 ms GPU time per frame
    • Start and stop each in one mode change
    • Connector during DV: max bpc 8, full-range RGB, no HDR metadata, DVBRIDGE_REFUSED 0 (DVBRIDGE_LINK_FORMAT also reads 0)

    Two things I noticed:

    • With a settings window over the video, every frame is composed with the GUI and output drops to about 16 fps (43-47 ms GPU time per frame). Closing it brings back 240 of 240.
    • The log repeats "DVBridge libplacebo: Trying to call pl_opengl_unwrap on a pseudo-texture", with no visible effect.

    Happy to send the log.

    Thanks, this is a great report, and the first one on a discrete AMD card.

    The drop with a settings window open is the important one. Frames with the GUI go through the composited path, and 43 to 47 ms on an RX 7600 is far more than I'd expect from it. The pl_opengl_unwrap message has no visible effect, as you say, but it shouldn't be there either.

    Yes, please send the log. On the LIBREELEC partition, rename DIAG-OFF.TXT to DIAG-ON.TXT, play a Dolby Vision file with the settings window open for half a minute, shut down from Kodi, then zip the hades-diag folder and attach it to an issue on GitHub.

    I'm away from home until Tuesday. I'll look into both and try to fix them, and there will be a new update between Tuesday and Wednesday.

    Hi! I’m new here and came across this project while looking for a way to play Dolby Vision Profile 7 FEL on a PC-based setup.

    I haven’t bought any hardware yet and would appreciate a recommendation before I do.

    my setup / use case:

    • TV: LG OLED77CX9LA
    • sources: UHD Blu-ray remuxes in MKV, Profile 7 FEL (single-track, BL+EL+RPU); MP4 dual track possible too
    • goal: DV Standard (TV-led) output with full FEL reconstruction, no frame drops

    what I’ve gathered so far:

    • Intel 8th gen and newer are listed as supported. But 8th/9th gen iGPUs (UHD 620/630) only output HDMI 1.4 natively and depend on an LSPCon chip for HDMI 2.0. Does DV signaling work reliably through LSPCon at all?
    • N100/N150 are listed with occasional frame drops on some FEL files.
    • the reference hardware seems to be an i7-11390H / Iris Xe.

    my questions:

    1. is an 11th gen Iris Xe (e.g. i5-1135G7) a safe minimum for FEL, or is 12th/13th gen noticeably better?
    2. has anyone successfully used an 8th/9th gen machine with LSPCon for DV output, or should I avoid those entirely?
    3. are there specific mini PC models that users have already confirmed working (ideally with an LG OLED)?
    4. is dual-channel RAM required to get stable FEL playback on Iris Xe?

    Thanks in advance for any advice!

    As soon as Beta 4 is released (maybe today) i will test it on my N100, will post my findings here.

    Hello

    1- You CPU is safe for the FEL don't worry, since beta2 release a N100 can read without any problem and any drop a Remux with FEL and even the new pipeline "dolby vision master reference" that make no compromise on the picture quality and will be release today in Beta4, the only pipeline that a N100 do not handle is the DV adaptive HDR mapping at 50fps.

    2- You need a native HDMI output, no DV support is possible with LSPCON or other HDMI converters. Sometimes the converter is integrated directly onto your motherboard, leaving you no way to get a true native HDMI output. This is quite common on 8th and 9th gen motherboards, so you have to be careful, unfortunately, this information is always tricky to find

    3- Many miniPC works perfectly, take at least a N100 to be confortable.

    4- no dual-channel is not required, my test benchs are a N100 with 8gb of ram in single channel and a AMD 7430U with dual channel. The 11390H was my first test bench and has been replaced.

    Releases · CroqueMr/libreelec-x86-DV
    LibreELEC x86_64 with Dolby Vision support for Intel and AMD iGPUs - CroqueMr/libreelec-x86-DV
    github.com

    R1.0.0 Beta4 - Major pipeline optimizations, Master Reference and NVIDIA support

    Rendering modes

    • Dolby Vision Standard: accepts near-imperceptible image differences to deliver substantially higher rendering performance.
    • Dolby Vision Master Reference: new TV-Led pipeline aiming for bit-perfect playback, preserving the original pixels and metadata. The N100 handles all tested sources in this mode, including 4K FEL remuxes and 4K60 content.
    • Dolby Vision Adaptive HDR Mapping: replaces Natural and Signature with a single mode. Expands highlights and selected midtones using the dolby vision metadata and configured display capabilities, with shadow protection and temporal stabilization.
    • Adaptive processing also supports HDR10 input through AI-generated Dolby Vision metadata.

    Performance

    • AMD Performance, performance improvement on AMD CPUs thanks to dedicated rendering pipelines optimization.
    • Major performance improvements to HDR10 to Dolby Vision AI, reducing analysis overhead, redundant processing and preparation delays.
    • Substantially optimized Dolby Vision to HDR10 Basic and Expert, including FEL reconstruction and direct presentation.
    • Optimized Master Reference by removing redundant reconstruction, intermediate surfaces and copies without changing the reference calculations.
    • Improved Adaptive processing through reduced memory traffic, streamlined shaders and earlier FEL preparation.
    • Added automatic fast rendering, using a 2.5 ms presentation-margin threshold. Player information reports its active state.
    • Master Reference remains non-Fast. Automatic performance recovery can switch to Standard when necessary, with a notification.

    Image processing

    • Corrected Expert output-volume limiting, including BT.2020 peak-brightness overshoot.
    • Improved Expert handling of missing or inconsistent L1 statistics and compatible L8 trims.
    • Corrected the processing domain of CM4 trim commands in HDR10 Expert.
    • Improved Adaptive pixel/metadata consistency and addressed reported darkening and brightness instability.

    Hardware and 3D

    • Added an NVIDIA backend for Turing and newer, including NVDEC integration, HDMI Dolby Vision signaling and shared CB1 processing.
    • Improved Dolby Vision capability detection on supported Intel HDMI paths.
    • Corrected Blu-ray 3D stereo-output selection when native HDMI 3D modes are unavailable.
    • Added progressive 1080p half-side-by-side HDMI modes at 24, 50 and 60 Hz.
    • Retained separate native 3D playback, outside Dolby Vision processing.

    Playback and interface

    • Fixed an L10 metadata-parser freeze affecting some Dolby Vision files.
    • Improved startup, seeking, BL/EL pairing and recovery after interrupted preparation.
    • Fixed rendering-mode transitions, stale textures and unfinished-frame cancellation.
    • Corrected L5 handling with menus and overlays.
    • Improved native Kodi format reporting for skin integration.
    • Simplified Player information: actual TV-Led/Fast state, relevant display-profile fields and supported render-engine usage.
    • Quick rendering selection remains available through the O panel cycle.

    Known limitations

    • Heavy conversions and visible menus can still exceed N100 rendering capacity, particularly at 60 FPS.
    • Adaptive rendering remains subject to visual refinement.
    • NVIDIA hardware playback and AMD HDMI 3D still require further physical validation.

    Credits

    • hadesdevls: NVIDIA support, adapted from LibreELEC Hades.
    • dangerouslaser: L5 active-area fix for menus and overlays.
    • CinemaOne and popcornmix: upstream Blu-ray 3D work used for the AMD integration and stereo-output fixes.


    Next focuses

    • Move to a stable release : small clean in the repo, bug fixing and rendering pipelines fine tunning following feedbacks.