I deeply respect the work of djnice which has really opened new possibilities for the S928X, but it’s software rendering with limitations instead of hardware decoding. Even he admits that the FEL is 10bit instead of 12bit because of hardware limitations and it can only work for 24p videos.* The challenge is that this requires a lot of custom code that I’m not sure CE would want to support for a long time, especially since this is not proper 12 bit FEL. One change in Kodi and everything breaks. This is a lot for the CE team to handle over the long term. So now we have more forks for people to choose from and even a fork of a fork.
Honestly, if anything gets pulled in from djnice’s recent work, I would hope it would be the software 3D decoding, which seem to be only 3 relatively self-contained commits, to overcome the limitations of the Amlogic MVC codec.
“The S5 display path takes at most 10 bits, so the FEL output is 10 bits plus dither, not a true 12-bit signal.”
Yes, I read his explanation before he posted the build so I had tempered expectations but DMDReview’s results shows that I’m probably not going to be able to see the difference with my own eyes.
Like bluesea mentioned, there is only one 4K 60Hz movie (Gemini Man?) and it’s MEL.
Completely understand CE team not wanting to take on a lot of code for one specific platform - I wasn’t really expecting that.
Yes, the DMDReview result are quite impressive. I think this is most important because it creates a path to implement FEL on Dolby Vision hardware without native FEL support. I’m particually excited about the idea of this coming to RockChip RK3588S2-DV devices. Of course, that would be for LE, but it would be amazing to have such a powerful CPU with FEL. Exciting times!
What’s the use of 12-bit output if most TVs (especially OLEDs) struggle to even display 8-bit signals without increased color-banding/posterization? Dithering is almost mandatory either inside the display or source device. Dithering from 12 to 10 bit isn’t going to make any noticeable difference and may actually improve the quality on displays with poor internal dithering.
I really didn’t want to sound as if I was disparaging your work. From a real world perspective it is probably perfect. I was more pointing out the technical limitations. I’m sorry if I caused you grief.
Yes, this question is very interesting. Djnice and I spent an entire day checking and comparing the reality of these 12‑bit streams. I started pondering this question about three weeks ago, trying to figure out whether there are truly 12‑bit streams in DV FEL. I had to develop a utility together with ChatGPT (by the way, Claude refused to help me, saying it was illegal); this utility provides a real, lossless capture of video from the player’s HDMI output. And now we’ve been able to compare three implementation options. hw amlogic, hw mediatek, and sw-hw are some kind of hybrid; I can say that you won’t see any difference between them with your eyes. But if you’re looking for a reference, then Oppo can produce a very accurate result for the original image in grayscale. s922x produces a larger variation in the output; this is probably due to how the merging is done, which algorithm is applied, and how rounding is carried out, since the transformation process is multiple within DV. yuv-lms-rgb-itp. A rounding error occurs at each step. And here, the s928x indeed shows more accurate results due to dithering than the hw amlogic. Moreover, technically, it may even look better on real images on uniform surfaces than the Oppo, because it generates noise even where it’s not encoded, so it will look even better than real 12‑bit processing. Yes, these are not the original source bits, but we will never see this difference. Plus, we need to take into account that the Oppo has many other problems with DV, so there’s no need to rush to buy it. It is important that s928x ensures accurate brightness output, i.e., the restoration is complete.
Where is the source from the open source project?
It’s a own coded proprietary dovi.ko, even encrypted so nobody can analyse it.
OSMC is since a long time changed to C(losed)SMC…
And the worst part is that they use the free Kodi forum to advertise their product, which is overpriced, by the way, for something that doesn not even provide real Dolby Vision.
It’s shameful to cheat users just to get money out of their pockets…
I haven’t compared it to Libreelec yet, but it’ll be interesting. I’ll try to find an Intel PC to test it on.
In theory, 12th-generation Intel processors support 12-bit internally, so the output image should be more accurate.
I tried the Aerial HDR screensavers again with 20261005, sadly I get the same red screen, like there’s a mode switching issue in the screensaver code somewhere. Everything else works great (AM9 Pro). Would I be right in thinking this might be a relatively easy fix now there’s been substantial work on that part of CoreELEC recently? I saw the first RC release went out at that weekend. Amazing work people
Yes, sorry, that’s what I meant. If the Aerial screensaver or Video screensaver plays an HDR video file from the Apple Aerial screensavers, playback fails with a red screen, which I think is a mode switching bug. The same happens with the same footage if I use the Video playback screensaver with the same 4K HDR footage.
If I play that same footage from the file browser the files play fine so I think it’s something specific about the screensaver code and mode switching.
