GPU Memory is full on a GTX 1070Ti - Buy new Card or....??

Hi!

I am working on some videos with a High Dynamic Range, and unfortunately it's going to take Noise Reduction on every clip to fix the noise.

When I render, I always get this message, whether it's a 30 second clip or 10 minute long final video.

Your GPU Memory Is Full
Try reducing the timeline resolution or the number of correctors.

Here are basic computer stats:

Intel i7-7820X
GTX-1070Ti
32GB RAM
Operating System is on an M.2 SSD Drive
Scratch files are on an M.2 SSD Drive


I know Noise reduction is one of those things that takes a lot of processing, but I'm getting some artifacts in my final product... I'd call it 'stutter' or 'jitter', usually during a transition from one clip to another.

Any ideas (besides doing better camera work)

- what is your timeline resolution?
- what is your deliverable?
- can you reduce the timeline?
- if not, can you move the NR nodes to first and use node cacheing?
or move the NR to last node and cache the node before?
- are you useing any OFX?
- last eval if you really need to use that much NR, i've done entire feature films without touching NR.... it's destructive, it's just a case of eval'n what's the least worst, artifacts from NR, or noise, usualy i'd prefer to add film grain to the shots around a noisy shot and not touch NR

[Peter Chamberlain]

I'd try doing a pass with no NR, render all individual clips and then bring them back in to do a separate NR pass.

I'd try doing a pass with no NR, render all individual clips and then bring them back in to do a separate NR pass.

That is a good idea.

I admit I am new to Resolve, and while this is a bit daunting-sounding to me, I can probably figure it out.

The biggest point of confusion with all this is: All the different render options, formats, codecs, etc. etc. If I do an initial pass with no NR, what settings should I use?

And thanks for replying; it's great to have a Blackmagic employee reply

My timeline/project info is roughly:


- 4K UHD (3840x2160)
- 23.976 fps
- Deliverable will be 2 videos, one @ 1080p and the other at 4K, both will be uploaded to YouTube & Vimeo
- The timeline length is approximately 10:21 (minutes:seconds)

Depending on your source (and available disk space) you should export to DNxHR 422 or 444.
An alternative would be Cineform.

- what is your timeline resolution?
- what is your deliverable?
- can you reduce the timeline?
- if not, can you move the NR nodes to first and use node cacheing?
or move the NR to last node and cache the node before?
- are you useing any OFX?
- last eval if you really need to use that much NR, i've done entire feature films without touching NR.... it's destructive, it's just a case of eval'n what's the least worst, artifacts from NR, or noise, usualy i'd prefer to add film grain to the shots around a noisy shot and not touch NR

Hey there; thanks for replying! I really appreciate it

To answer your LAST question FIRST: Yeah, I know NR isn't the ideal thing to do. And I'm a one-man operation, working in the marketing department of my company. And we haven't found any professional production companies who want to do our work.

That might sound weird, but the video projects I started working on recently have a long time frame; meaning I might be at a location for a day, then go back a month later for 2 days, then..... It's common that it would take 6 months just to get all the raw footage. (This is in the construction/industrial industry, so we are at the mercy of the construction manager and customer's time frame, and the actual project we are filming can easily take 6 months to complete since the filming has to be spread out over such a long time period.)

Anyways, for THIS project, most of it was shot in what I would consider an HDR situation. Much of it was at night time, so pitch black, but with a LOT of very bright lights all over. Or in a poorly lit room (that I could NOT bring any lighting into.) Maybe one day we'll get a RED camera (Or a BM Ursa! hehe) or some other camera with more stops to handle the dynamic range. For now we are using (2) Panasonic GH-5s.

I bet you are all wondering why I wouldn't add lighting to the room(s), etc., but the nature of the location was a live site, and I was 'lucky' to get in 2 cameras, a few mics, tripod, monopod, and a small LED light. They wouldn't let me bring in bigger lights like my Aputure 120ds, etc. But I still don't think that would have been enough, at least not with me being the ONLY person to operate the camera, sound, setup, etc. etc. One room was medium sized, and another was, well not even a room. More like a HUGE warehouse type place. Next time, I'll frame my shots better. But the fact is, all the footage has more noise than I'd like

When I was on location, I did my best to capture the darks, but there were ALWAYS bright lights, everywhere. I captured in Panasonic's V-Log, and went with the lowest ISO I could, but I think that, watching the footage in Resolve, I'm seeing a lot of noise, and my guess is that it's from an ISO that was 'too high' on the camera. (Yet, when I went lower on the ISO in test shots, I wasn't getting everything I wanted, so I had to turn up the ISO a bit.)

I hear what you're saying about adding some grain to it all, but this is more along the lines of a corporate video, and I have to be careful about getting 'creative' with adding film grain.


My timeline/project info is roughly:

- 4K UHD (3840x2160)
- 23.976 fps
- Deliverable will be 2 videos, one @ 1080p and the other at 4K, both will be uploaded to YouTube & Vimeo
-The timeline length is approximately 10:21 (minutes:seconds)
- I don't think I can reduce the length of the timeline. I have to show the entire process, and since this real-world project took ~3 months, it's hard to squeeze it down (at least for me, the world's most novice producer!)
- I put a NR node on every clip, because there is a significant amount of noise on every clip (due to the camera's ISO setting when I recorded). I'm new to Resolve, so I'm sure there is a more efficient way of adding noise reduction at the end, or like was suggested, render everything WITHOUT NR, then add NR and render again.
- Node caching: Yes I was experimenting with this a bit, but I kept running into issues with filling up my drive! I have a 512GB SSD dedicated just for the cache, but it was filling up and leaving me with 20MB free space on the drive! Yeah, I wish this wasn't a 10 minute video, but it includes an interview with some VPs, and my boss wants the entire interview. (I already cut out about 5 minutes of the interview, but I don't feel I can cut out anymore without missing key points.)

Sorry for the long reply!

Thanks : )

11min of UHD @ DNxHR_444 should fit, with room to spare
maybe try change cache to that? or even DNxHR_HQ if you are tight

i rarely to never suggest to producers/directors what coulda / shoulda been done on set, i know everyone is trying to make the best trade-off's, i trust that anyone who come into my suite really cares about the quality

try Paul Dore's plug-in that anylises noise and only reduce what is needed, usualy it's shadows in the blue channel, so try a keyer to limit the NR, and therefore the artifacts, and keep the bad stuff away from the good stuff that did not need NR in the first place?

not sure if there's any advantage to running out a submaster to DNxHR compared to cacheing the last node before NR in DNxHR, one disavantage is temporal NR uses handles to calculate what is nose and what is detail in the content, you lose that advantage when running out a flatened master

if you do run out a master i'd try running out with the Pp XML option and 5fr handles, then temporal NR has access to the media it needs to do the best job it can

in theory an output @ DNxHR444 and a cache @ DNxHR444 should be the same, but the cache version can be quickly changed, the master not so easly