Hi, I want to ask about splitting sections, what do those two arguments mean? Size and padding, and what is the recommended settings for a 16gb Tesla V100 GPU when proccessing a 3GB 56min animation? Thank you.
Hi nekodaze. The 'Section Size' and 'Section Padding' under the Split Frames section (It is beside but you got the point) is a roundabout way to render frames while using reduced memory size.
It is not recommended and I would not recommend it AT ALL. It will create artifacts in fast moving objects, smokes, and similar action scenes. It will create boxes artifacts on your entire video, like for my case, when I try to render 1920x1080 video (23 hours render), I use 450 Section Size and 150 Section Padding under the split frames section, it creates 8 equal sections on my video when it is done rendered
(only seen in high intensity scenes as mentioned before).
If you still don't get it due to my 1st class English, imagine 8 TVs with frames and all, put together to make 1 big screen. You can still see the border even though it is see-thru (like table glass color).
About the 16gb Tesla, I am currently on GTX 1070 and the maximum size it can handle is 500 Pixel.
I think the VRAM can correlate to how big the frames can be rendered in linear fashion. (Without the split frames section being selected).
I have tested a GTX 1050 with 2gb of VRAM (120-140px), 1050 ti 4gb (240-260px), and GTX 1070 8gb (480-500px).
All are tested personally by me and not taken from other sources.
If we calculate the frames that your gpu is able to make, it will be 2x bigger than mine, about 960-1000px.
I know this sounds such a turn off but it is in Alpha state, and as you can see from the error codes, you can tell it is still not yet optimized properly and needed some more time to mature.
All you can do is wait it out to be better, it is just ridiculous to think you have to spend on a 24gb gpu to render a 1080p native file without the artifacts.
And as an added bonus, I do not see any difference in rendering speed in different cpus' strength for my case. The GTX 1050 2gb uses an i5-3470, 1050 ti uses an i7-3770, and the gtx 1070 uses a Ryzen 2600. When I use the same datum settings, only 120px on all setups to match the limiting factor of 1050 2gb, all render at almost the same speed (same size,frame rate, and video duration of 30 seconds), all take about 1 hour or so to finish). I have yet to test on i3 or lower tier processor since I do not have any.
I hope this helps answer your question. I am not an expert or anything, just telling from my own collected personal data, settings and experience.
Thank you very much Noraiman, I thought it is splitting video by duration instead of picture, I never thought that a few 1080p frames would take over 16GB VRAM. Looks DAIN has a long way to go.
I've tried many different size values and I think 900px may be the limit for 16gb, very close to ur calculation. If it goes higher rendering process will be unstable, some scenes may take more memory.....Or maybe just my video bit rate is too high, each frame after extraction is about 3MB, I haven't try other video so not sure if this size is normal.
I temporarily gived up proccessing this video, for it will take about 860h to rendering 🌚
About different CPUs, I noticed CPU usage is low after frames extraction, so yes cpu is probably have no thing to do with rendering speed.