Finally I got it done! and it’s absolutely COOOOL!
I compiled a version for intel mac as well. And I adapted the hook to work in flame 2023.2.x . If I could get hands on a linux only for compiling the transcoding engine, I could get it to work with all flame boxes
2 Likes
What can I do to help?
2 Likes
The h.265 Importer is now available for Apple Silicon/Flame2025.x and Apple INTEL Mac’s/Flame 2023.x on my patreon
2 Likes
There was zero interest.
If there is any interest, I would consider rewrite the h265_decoder in c++ and compile for rocky linux with proper nvidia hardware acceleration; I have now a rocky9.5 box available with 2x rtx a5000 to test it. The apple silicon version works pretty cool since march this year and even the apple intel version for older flames(2023) on mac rocks. I think I try it anyways.. I am just curious about how the hevc/H.265 import will work on a linux box and how it compare to apples hardware engines
2 Likes
Today, I used nvidia’s actual videoenc/dec sdk to write a h265_decoder (the c++ cli converting tool). Now please give me any idea what kinda footage you would like to get into the timeline from h265/hevc footage. Prores 10bit 422 works perfectly fine on apple silicon. But on Linux Flame, there are different options. Let me know what kinda material you would like to get finally into your reel.
2 Likes
Why not use h265_decoder to directly add h.265/HEVC footage to the timeline? If intermediate encoding is still needed, I don’t think it’s the best solution.We hope Flame can directly decode H.265/HEVC footage like H.264.
Thats exactly the point. Flame does’nt support native hevc / h.265 footage to import right from any camera since ever. And yes, you can of course transcode footage with ffmpeg or other tools. But my intend is to get this video clips directly into the reels. Flame native hevc / h.265 support is a dream that did not come up since a decade or even longer. But now I can import that stuff super fast directly while the demanding client sit here and watch. NATIVE? NOPE! Workaround nearly as native? YES
3 Likes
nvidia sdk use of course the hardware on the rtx cards, and I now have a chance to play around with a potent rocky linux box. Why that’s not native implemented? I have no clue. Do I need that feature? of course. If I can import 10bit 422 hevc native, I would not think twice. But as long that is not the case… my lil plugin shine pretty bright
1 Like
Excellent! Continue optimizing, don’t stop. As long as it can perfectly decode H.265/HEVC, there is currently no need for encoding function.
The licensing fee for H.265/HEVC is not low, and I think this is one of the main reasons why Flame does not natively support H265/HEVC.
1 Like
A I stated, it’s not that easy to let nvidia cards do the same as the apple hardware does. Its kinda extremely tricky. Nvidia only supports pre-defined streams to hardware encode/decode. If that’s out of range (6k, 8k, 12bit 422 or any combination, it falls back to software encode. Yes we have 40 cores available in cpu. but thats not the point. nvidia is very picky in what kind hevc footage comes in. So I see myself working 1 more week to get it round and about.
4 Likes
To compaire that:
apple silicon hardware eat anything for breakfast and put a proper stream out that rocks in milli-seconds. Nvidia is a Diva and DONT LIKE UHD yet. real 4k works already well, 6k and 8k and 12k … just nope! forget it! Does’nt work at all. Fallback to cpu on this cases. And H.265 on cpu might be ok on fast multicore cpu’s, its SLOW like a Grandma’s Coffee Machine. I only have my old 2x RTX A5000 for now. They are NOT even mentioned in the actual Nvidia Video Codec SDK at all.
1 Like
Very insightful, and a concrete example of how Apple Silicon is the preferred environment for QuickTime.
The almost instantaneous transcoding of a ProRes QuickTime to a web compatible stream is nothing short of astonishing.
Steve Jobs used to evangelize about how the Compression / Decompression business was significantly more important than most things, hence h.263 in video conferencing, development efforts into h.264 and everything subsequent to that.
He was right.
Obviously, being right does not excuse monstrous behavior and monopolistic tendencies.
3 Likes
I got the C++ rewritten h265_decoder finally to work today on rocky linux (2xRTX a5000) with top speed & same high quality output as the apple silicon version and its now not picky anymore about what kind of hevc / h265 mov material from different camera brands and models inclusive most modern iPhones comes along. Wrapped in the flame python hook, it works now input wise with the actual nvidia codec sdk with a fallback for exotic files to use x265 sdk and cuda ffmpeg libraries with nvidia enc/dec the same way as in apple silicon.
Mission accomplished? Not yet. I compiled and tested the cli command “h265_decoder” and on almost rocky flames it should work out of the box. If it does’nt, I also made a installer as rpm ‘h265_decoder-1.0.12-1.el9.x86_64.rpm’ with all needed libraries. Now I need someone to test the cli tool as it is, and the complete hevc h265 import plugin package for flame.
2 Likes
my gpu is too old i expect. ping @mybikeislost to see if he can squeeze in a test - i think he has RTX5000 or RTXA5000 and rocky 9.x
1 Like
Hi Phil, I implemented a fall-back cascade : 1st nvidia codec sdk for proper speed and utilising the hardware inside the nvidia cards> if that not works > ffmpeg libraries with nvidia cuda support for rendering with available cuda > and if that also not goes trough > it uses the cpu’s on the available machine to transcode that hevc h.265 material. As you mentioned, Jobs did a great job on all that streaming/recording tasks with Quicktime we ever imagine on apple hardware. Nvidia on the other hand is pretty much super silent and even to get the proper codec sdk from nvidia was a bit of a challenge for me as a european living in china since 2 decades. Well, I made it work and I am happy doing that. Not, that I personally need that on a linux flame. But I learn so much about the different quirks and tweaks… I can use that for my general AI implementations. Have a nice weekend and hopefully talk soon. There is more coming up in that AI world
2 Likes