HitFilm Pro 2017 Export speed when using 32-bit color and 32xMSAA (Resolved)

edited November 2016 in Pro Support

I just purchased my copy of HFP2017 and am very happy with it. It works like a charm . That is, until you begin the mighty export process . I made a 10 second clip just for testing purposes , and the remaining export time was 1 hour. 1 hour! ! ! For a 10 second clip! ! ! HF4Express never did this, it only took a min for the same exact video. (I have a slow HDD , but not that slow! ) Anybody else experience this with the pro 2017 version? (not demo, full licence ) 


  • I must add, my computer is way more powerful than the minimum spec list. 

  • For testing , I loaded in a clip and exported it without touching it. It worked fine, but with a slight bit of editing or even tweaking the volume of the music, it goes back to 1 hour of export time . Again, if anyone else experienced this, please do comment below. 

  • edited November 2016

    You gave no information on your export setting or project setting so I just tried something on a guess.

    I exported to MP4. Specifically the Youtube 1080 preset. My export was 29.97 fps.

    I exported a simple media file about 1:01. The media file is an AVC MP4 file at 1080p29.97. I tweaked the audio volume with the NLE timeline volume bar and then exported. It exported in 1:24. 60p would export in roughly twice the 30p time.

    My machine: Window 10, 4Ghz 4770k (quad core), 16GB ram, GTX 980. Almost no GPU use registered since I used no video effects. The 980 power is not much consequence for this test.

  • edited November 2016

    It was 24fps 1920x1080 MP4-H.264 with the project settings at 32 bit color float and 32x MSAA

    All other settings were default

    @NormanPCN ;

  • Well, processing an 8-bit source file at 32-bit color depth does mean you've basically added a lot of overhead expanding the color range for effects processing (using 4 times the data of 8 bit processing) then compressing back to 8-bit output, and a 32x MSAA does mean that each frame is basically calculated 32 times. 

    So just those two settings increased the amount of data by a factor of 128.

    I'll just guess Norman was at 8 or 16 bit color and 4x MSAA. 

  • 8-bit project. 8xMSAA.

    One question to the OP is why 32-bit?

    For me, my media file is 8-bit so a 32-bit color intermediate depth does nothing for me here. I have seen 16-bit help very fine wispy mask feathers and some gradients even with 8-bit output. Also particle sim stuff with very light alpha and lots of particle layering and building up likes 16-bit. The better fine granularity lets the tons of textures blend to a better result.

    Another question to the OP is why 32x MSAA? If you are not doing 3D models then why MSAA at all.

    One clarification: 32x MSAA does not really mean a frame is calculated 32 times. It means that rasterization of vector edges to pixels are sampled "32x" to get a smoother anti-aliased edge. I am on record as to my speculation as why/how our 2D media in Hitfilm is being mucked up by the vector texture subsystem via MSAA.

  • @NormanPCN Thank you for the correction. I am guilty of oversimplifying to stress that 32x MSAA is a lot slower than 4xMsaa. :-) ESPECIALLY is your speculation that all layers are actually treated as textured polys is correct, which I think it is, given how 3D Unrolled seems to work (Calculate all layers until hitting a 2D layer, then render). A 2D layer kind of acts like a 3D poly locked at Z=0 and forcing a render of all underlying layers before moving up the stack. This would effectively make all renders "rasterization of vector[s]...to pixels." This would also explain why lights "disconnect" from underlying layers if separated by a Grade. It also seems a logical, and clever way to hack a videogame library into a compositor.

    @Danny77uk I haven't tagged you for awhile, can't ask Cedric all the techie questions? Can you confirm or correct?

  • The guys above are spot on here. There is no point upping to 32-bit color unless you have a specific need for it and have higher quality source. Even on our fastest computers I only tend to use 16-bit unless there is something I'm trying to test. Also, 32xMSAA is very slow, and depending what you're doing you may need no more than 4x or 8x - I hardly ever use more than 8x.

    A normal 1080p frame is around 6MB, 16-bit around 12MB and 32-bit 24MB... Along with 32xMSAA you're computer is doing tons of work here which it probably doesn't need to be doing and a lot of that will be shifting around and converting stuff to 32-bit for no reason.

  • edited November 2016

    @NormanPCN @JoshDaviesCEO @Triem23 ;

    Thank you all for the advice and clarification!

This discussion has been closed.