Render Server Upgrades - Supporter Early Access
Added 2020-06-19 01:46:04 +0000 UTCThis video has been too long in the works. I'm really sorry. I kept running into problem after problem and it just kept going.
But it's finally done and now that my server is back up and running and I actually have somewhere to put footage I can get back to filming. I've been going crazy waiting for all the parts to arrive so this could be done. Unfortunately, I have to go out of town for a week so the next video will be a little delayed again. But I have a mail video queued up next to be shot, a last ditch hope at getting a scanner to work, and a really boring monitor that I accidentally found the best use for. So I will have a lot of videos in production when I get back.
This video is going to have a sponsorship by Linode in it, but this was mostly done before I got the go ahead to add that. So this is going to be a supporter only upload, when I get the sponsor segment added in I'll be reuploading.
This next part is for after you've watched the video...
This video was rendered on the server! It didn't have the full noise reduction processing because it's not the final upload, but it took ~43 minutes to render. I expect that to approximately tripple with noise reduction but I'm pretty happy with that.
Comments
Ah yeah, I hadn't considered that for that part. I just thought it would be funny to try and read through it all as fast as possible. If I hadn't been heading out of town I would have tried to spend more time on getting subtitles ready to make that easier to understand. I'll go back and get that done when I'm back home at work.
Tech Tangents
2020-06-24 15:38:12 +0000 UTCFirst time I had to use the 0.5x Playback speed to follow your fast talking intro (non native English speaker here). :)
Bernd Kilga
2020-06-22 13:58:45 +0000 UTCWhen I toggle the noise reduction on and off in Resolve I can see a huge difference. It's mostly in dark areas but that ends up being a lot of things I film since there was a lot of black plastic. After doing a little more testing I was able to get the 1080ti and 1070 to perform much more similarly with noise reduction enabled. If I tune how strong I have it set so that it doesn't need to swap data out of VRAM constantly, it works much better. I haven't compared different output renders from Resolve before for perceived noise. I should do that some time before I make an update on my linux production workflow to see if I can skip it. One thing I have noticed with renders with and without NR, is that the renders themselves can be 20% smaller with it enabled: https://twitter.com/AkBKukU/status/1263827075191857153 On my Really slow 35Mbps up internet that makes a noticeable difference on 10GB+ uploads. I haven't ever really measured my light output or really have a way to. But I've got 4x105W CFL studio lights and 4x 60W equivalent LED house lights in two large soft boxes. I shoot with a CPL at 3200 ISO. More light is something I feel I could always use. I'll mention that this camera is dual native ISO as well. So it's not just "lower is better". 400 and 3200 ISO are native and have almost the same dynamic range. I would have to get down to 500 to get more dynamic range from the lower native ISO: https://cdn.fstoppers.com/styles/full/s3/media/2019/02/23/bmpcc4k-dynamic-range-to-iso-chart.jpg I don't have enough light for that with the CPL to do that. I'm shooting at 12:1 BMRAW as well so I can get some real recording times out of my SSDs. I could bump that up and get a bit more quality, but it's a huge trade off. I'm also not using a great lens, a Nikkor Zoom 18-55, with a metabones speed booster. That's why the chromatic aberration. It was a headache to put it all together but it turned out really well in the end. There were a bunch of different things I tried, so it definitely wasn't a smooth process. But I think what I have setup here is going to work really well for a long time now.
Tech Tangents
2020-06-21 14:04:50 +0000 UTCDo you think the noise reduction is really worth it? (I don’t for 3x render time) I’ve been looking hard at this video (VP9 2160p60 4:2:0 2.68 GiB) and to me all I see is typical macroblock artefacts, colour banding and some chromatic aberration. I do see a typical amount of gaussian color noise on the macroblock level but I can’t see any pixel level noise of note. Considering how VP9 (and other codecs) DCT lossy compression inherently smooths data most minor pixel level noise is removed even without any denoising. Out of interest how many lux do you typically have falling on your scenes for a given ISO? Good to see the sever has worked out and configured in just 3 weeks, when setting mine up I went through several ZFS pool configurations and several OS installs over 3 years and only this year have I been reasonably happy with it :) But now I have to upgrade my pool for new ZFS features…
WizardTim
2020-06-20 14:31:55 +0000 UTC
