Check out post # 292 -293 from user ‘E-Support’ in this thread - you may need to update FFMPEG.
So I did. Same result…
Do you have more than 1 GPU in your system including integrated graphics on your motherboard? I made a discovery this weekend after looking through some of my render logs and it seems that as I had been reinstalling Topaz a few times trying to address my own issues with slow renders coupled with barely any utilization and low VRAM usage as shown by Task Manager. While looking at my performance graphs, I happened to notice that my onboard AMD GPU had a very small amount of load on it (like 5-7%), meanwhile my 5090 had around 23-33% load with only 7.5 GB used. The logs revealed that somehow, the system has been defaulting to my onboard AMD GPU instead of the dedicated NVIDIA GPU. This was odd considering that when I checked the presets in settings, they did show NVIDIA; however, when I looked at the ‘Codec settings’ in the main menu, sure enough, it was defaulting to AMD. I don’t know if this is a bug in the system or if it really cannot properly detect the faster, better card, but I thought it worth mentioning. I honestly haven’t had a chance to test my slowness theory as I’ve had a Proteus render running for literally days (it’s like a 3.5 hour video) because I’m at the point where it makes more sense to just let the AMD GPU finish its work rather than restart it with the NVIDIA one set.
If this the problem, you can deactivate onboard gpu into Windows device manager. I was not sure running into trouble by doing this, but for sure no there exists also CPUs without integrated gpu, so last week I did it, deactivated my AMD onboard graphic adapter driver without problems.
EDIT: you can also deactivate embedded GPU in the Bios, then OS can’t see it, its gone, this is a better solution
Tell me under what conditions will all 24 GB of RTX 3090 video memory be used without using the computer’s RAM? (when using Starlight) I process video in low resolution and the maximum that I have video is 16 GB. It’s a bit of a shame that 8 GB are idle without use.
It doesn’t matter what minimum memory volume I set 95 90 80 70 50 it doesn’t affect anything
1 Like
I have simillar effect with my 16GB card, most of time between 12.5Gb-12.7GB gets used and additional 0.1-02GB shared. This makes no sense VRAM is not maxed out while a little bit shared is used. Next I try CLI, there you can set in numbers ammount of card ram and you can go over physical VRAM
I’m trying to read these but don’t have much free time with my job. Maybe these can help the performance issues with Starlight Mini. I think I read there’s a 96GB version too.
RTX 4090 48GB
https://www.reddit.com/r/LocalLLaMA/comments/1iy7e4x/rtx_4090_48gb/
4090 48GB after extensive use?
https://www.reddit.com/r/LocalLLaMA/comments/1k1e4y4/4090_48gb_after_extensive_use/
RTX 4090 48GB vs 6000 ADA 48gb?
https://www.reddit.com/r/LocalLLaMA/comments/1k598gi/rtx_4090_48gb_vs_6000_ada_48gb/
that does work well!
I have a 9950X3D, 64G Ram and 5090 LC OC (Astral) and Cant even get passed model loading 
I did disable memory fallback for the executable and python processes, still nothing.
Trying to load source 1080p to upscale 2x to 2160p using StarLight Mini.
How did you manage to overcome the issues yourself?
Hi, sorry for any inconvenience. The long time loading issue will be fixed in the next patch version. And we optimized the performance for high-end GPUs like 5090. Hopefully, everyone will see a faster version with 100% vram setup on 5090 machines.
7 Likes
Thanks Jason, any ETA for that version to drop?
I will have a meeting with the engineering team later today. I believe that update can be very soon. I can not promise but maybe this week. If it’s not, pls don’t blame to me 
3 Likes
haha, thats okay 
improvements are always welcome.
That’s a good point.
4 Likes
2H 1080P upscale to 3840 with Starlight Mini - ETA 44d - thats accurate? Interestingly only uses about 200W off 5090 LC OC.
Download the release essentials ffmpeg binaries:
Unzip the files into a temporary directory.
Open a cmd window, and type
where ffmpeg.exe
This will list the directories that contain ffmpeg. For each directory listed EXCEPT the topaz directory, copy the ffmpeg files from the temporary directory to the directory listed by “where ffmpeg.exe”
Starlight Mini gives very good results, but it still needs some improvements. For example, there’s no pause button, which is absolutely necessary because the process takes a long time and we can’t pause it. This is a major issue at the moment. Also, if it worked in Render, we could preview how it will look beforehand. In the past, there used to be a new version released every week, but it’s been a long time now—there’s neither a beta version nor a final release. I wonder when the new version will come out?
3 Likes
Please don’t leave us 4090 users behind 
7 Likes
I don’t have onboard video…just one video card, so I don’t think this affects me.
Maybe it’s the video file that’s faulty (or just somehow incompatible)?
I have the exact same behaviour, but only with one specific video file (of the ones I tested up to now).
The others work - and I now have done some short, if you can call 5-6 hours short, but also three long renderings over 2-4+ days (concert movies with up to somewhat over 2 hours and that at 0.3 fps) which all succeeded.
