Search options (Page 5 of 16)
RickyAstle98 wrote:dawkinscm wrote:OK so after more testing my current settings are:
TensorRT: 9.2
SC: Image Detect
SCT: 25%
AdjFra: Blend
Disabled similar to when previously set to SCT: 100%
SVP motion vectors work nicely except for Hugo.
NVOP was the worse in previous testing and nothing has changed.
Idk how you test, everything works as intended, check my previous comment, everything over 15/25% is similiar to look!
I know everything over 15/25% is similar. Actually I was the first one to point out that everything over 15% is similar 
But that's what I found worked best with version of MPV that comes with SVP. I've done more testing and found that my later version behaves as before which is a little different to MPV v0.36.
With 0.37 I find that 15% is the smoothest of should mean anything 15 and above but that's not the MPV that comes with SVP so I didn't update my results.
I test with a lot of different types of movement and I've been using the same test scenes for maybe a year or more so I am very familiar with the artefacts being generated by Rife or in this case by SCT. I also watch on a very large screen so I am able to see every microstutter and even the smallest artefacts.
In the end all of this stuff is setup dependant and YMMV
OK so after more testing my current settings are:
TensorRT: 9.2
SC: Image Detect
SCT: 25%
AdjFra: Blend
Disabled similar to when previously set to SCT: 100%
SVP motion vectors work nicely except for Hugo.
NVOP was the worse in previous testing and nothing has changed.
Chainik wrote:SVP 4.6.0.272
1. added motion vectors-based scene change detection for RIFE
2. configurable "decrease to..." function (on the "Frame size" page) - selectable target resolution and downscaling filter
Thanks for the update. I decided to completely reinstall everything to remove any changes I made. The results so far are that
- SVOP and Image Comparison and disabled give very similar results.
- NVOF might be a little worse overall but still similar.
- This is the same as before except that all of them feel a little less smooth than the previous setup with SC set to 100.
- Setting SC to 100 with this new software seems to be similar to the others as well. Although when I add back the MVP interpolation code that helps. I also put back the TensorRT9.2 files.
Let's see what others say.
I think with microstuttering, some of us can see it because of a large screen size or we are sitting close to a TV or maybe some of us are more sensitive to it. But I think we all get microstuttering with Rife.
flowreen91 wrote:Xenocyde wrote:Unfortunately I can still see microstuttering
I noticed that if i exit specific background apps like TeamViewer, i suddenly get 20-30 extra fps with SVP.
I believe TeamViewer reads the memory and causes a bottleneck even when it's minimized in taskbar.
Xenocyde try to close your 200 Chrome tabs and other background apps and see if u find the specific app that causes that microstuttering. It's good to know.
Try it with the 4.17 lite version cause it's easier on the GPU:
https://github.com/AmusementClub/vs-mlr … nal-models
The v4.17 lite version is an experiment and from what I see, it's worse than even v4.17.
dawkinscm wrote:I made it explicitly clear that I used to make changes the NV3D but now everything is back on default. I always double check before I answer questions about my system. If I'm not sure then I usually say that I'm not sure.
Xenocyde wrote:OK, so Prefer maximum Power is not helping. Maybe this is not NVCP-related.
Which is exactly the point I was making.
Xenocyde wrote:dawkinscm wrote:Xenocyde wrote:I don't really think it's possible you left everything on default, but whatever, maybe I'm asking too much.
Really? What made you decide today was a good day to call me a liar?
I just want to get to the bottom of this. Was not my intent to call you a liar, I was asking for the exact options you have there because I think having Power management Mode on Normal might cause more microstutters, I'm currently testing with Prefer Maximum Power.
I made it explicitly clear that I used to make changes the NV3D but now everything is back on default. I always double check before I answer questions about my system. If I'm not sure then I usually say that I'm not sure.
Xenocyde wrote: OK, I made those two modifications with NVCleanstall, but I can still see 2-3 stutters per 40+ minute TV series episode. I don't have 3D options on default as some games require certain tweaks. Also, I don't really think it's possible you left everything on default, but whatever, maybe I'm asking too much.
I told you about those two changes which I don't think I've ever mentioned before. I also told you that they were unlikely to make any difference, and your response is:
Xenocyde wrote:I don't really think it's possible you left everything on default, but whatever, maybe I'm asking too much.
Really? What made you decide today was a good day to call me a liar?
Xenocyde wrote:dawkinscm wrote:Xenocyde wrote:So the headset would introduce more lag theoretically? Not sure why that matters. Anyway, what options do you have enabled/disabled in the NV 3D control panel?
No changes made to NV3D panel. I've made some (very) low level driver tweaks by disabling Multiplane Overlay and enabling Message Signalled Interrupts but I doubt they would make a huge difference.
Can you please show a pic with all the options in the 3D panel and maybe explain how to do the other two things you mentioned?
Hi. Please read my post above again
As for the other stuff if you don't want to Google and do them manually then there's a tool that someone in this thread mentioned a few months back called NvCleanInstall which can help. The link is here: https://www.techpowerup.com/nvcleanstall/
Xenocyde wrote:So the headset would introduce more lag theoretically? Not sure why that matters. Anyway, what options do you have enabled/disabled in the NV 3D control panel?
It's interesting because I used to make changes to NV3D panel and have exotic configurations but it looks like I've gone full circle everything because is back to default. No changes made to NV3D panel. I've made some (very) low level driver tweaks by disabling Multiplane Overlay and enabling Message Signalled Interrupts but I doubt they would make a huge difference. Yes the headset does create more load on the GPU.
The occasional micro-stutter doesn't bother me at all when compared to what it was like before Rife.
Xenocyde wrote:dawkinscm wrote:Nope I stop using v2 a while back.
Then not sure how you're getting that performance difference. Different components? OC-ed GPU? Some config tweaks? You're getting same readings in task manager too?
My CPU and RAM have minor OC tweaks but I stopped OC'in my GPU a long time ago. Yes those readings are from the Task Manager. BTW my GPU is also transmitting the SVP+MPV processed video back to my headset.
Xenocyde wrote:Are you on V2 then? I took readings from 4.15 non-V2. With 4.15 V2 or 4.17 V2 @1080p, utilization varies wildly between 40 and 60%, seems like non-V2 is more intensive but at least it stays at 60 something all the time.
Nope. I stop using v2 a while back.
Xenocyde wrote:dawkinscm wrote:The occasional micro-stutter isn't a major issue of course, but it's surprising that your GPU is hitting 67% with downscaling. For me it's around 30% with the same GPU. I use mpv player.
Haven't actually checked the 4K downscale performance since I don't have a 4K HDR TV and I only get a 4K movie when I can't find the 1080p version. 67% is for native 1080p. I don't have HAGS enabled btw.
HAGS should be off so that is OK. But my previous GPU readings was FHD3D downscaled. With 4K downscaled and playing full ATMOS (which also uses the GPU) my GPU hits 55%.
Xenocyde wrote:I have an RTX 4080 and I only play 1080p TV series or 4K downscaled to 1080p @2.5X fps with RIFE 4.15. Maximum GPU utilization I see is around 67%. Microstutters are not too often, maybe once every 15 minutes at most.
...
There was a time when microstutters did not occur at all, but who even knows what version of RIFE or what drivers I was using back then.
The occasional micro-stutter isn't a major issue of course, but it's surprising that your GPU is hitting 67% with downscaling. For me it's around 30% with the same GPU. I use mpv player.
Xenocyde wrote:I had TRT 9.1.2 and decided to give 9.2 a try because I was noticing some microstuttering at times. From what I can see, fast movement might be somewhat improved with 9.2, but microstuttering gets worse. Should I update to latest TRT version or is there really nothing I can do to get rid of microstuttering?
The latest TRT version is noted to have 4x worse performance with Rife. So unlike with other version I haven't even bothered trying.
timharvey wrote:Hi all, I use RTX 3080 and it works fine with Rife 4.6 but not with Ensemble. Can anyone with experience running with Ensemble give me advice? Or what else does the installation need to do? Thank.
Advice? Don't use it. You would struggle to get it working on a 3090ti never mind a 3080 and from personal experience of using it, I saw zero improvements or benefits.
flowreen91 wrote:oriento wrote:4.17 - 2024.05.24 | Google Drive | 百度网盘 : Add gram loss from FILM, need to fully assess the impact.
Gram loss is a type of loss function used in style transfer tasks to capture and preserve the texture and style of an image.
In the context of neural networks and machine learning, FILM stands for Frame Interpolation for Large Motion. It refers to a specific model or algorithm designed to perform frame interpolation, which is the process of generating intermediate frames between existing ones to create smooth transitions, typically in video sequences.
So i think it improves this type of scenarios where the character is hidden temporarily behind an object:
https://github.com/hzwer/Practical-RIFE/pull/23
I have scenes to test this and if v4.17 was an improvement it wasn't very noticeable. But they are still concentrating on Anime so maybe it works better for that.
Xenocyde wrote:RIFE 4.17 is out with no lite version. Testing now.
L.E.: Not sure what this new version is supposed to improve... I'm getting quite a bit more fast movement artifacts compared to 4.15 and 4.16.
The author says that he is not sure it will improve anything. I tried it first thing and found zero improvements. YMMV.
Chainik wrote:the file attached there was updated 4 times already
the last time - yesterday
I see now. That was not clear to me which is why I asked. You might want to make it clearer.
But this is great! ASM vertical artefacts are still an issue but it's about as good as I have ever got it to look but now without the stutters. Definite improvements all round.
After a little more testing with v4.15, SC=8% with nvof is generally as good as turning off SC plus it helps with certain vertical fast motion artefacts. But I'm hoping the next Rife model can improve here because nvof causes problems for another one of my other SC torture files. So back to setting SC higher than 10 or turning it off.
Is this an official SVP update? When will it become available?
You're welcome 
OK another update. My previous tests were all with blend which is how I left it from previous testing. With duplicate frames 8% works better with Spidey and as well as 6% does with dumb algo. None of them are great but I think this is an issue with the model. From previous testing a few months back, setting SC at below 6% with the dumb algo will fix this issue while causing stuttering everywhere else. It's almost as if it turns of SC for that scene.
I hope some of this helps. 8% with nvof and duplicate frames might end up being my default from now on. But then I can find at least one other scene where the best option is SC=100 and the square is solid blue so...?
Chainik wrote:%APPDATA%\SVP4\override.js
smooth.scene.limits.blocks = 40;
default value is 20
with 40 at least the Hugo intro - the fly over the city and through the train station - is better (almost no scene changes)
btw
smooth.debug.qmode = true;
will draw a color square in the frame corner, when it's red it's a scene change
APOLOGIES I MESSED THIS UP A LITTLE BEFORE. FIXED NOW!
Definite improvements at 8%
- 8% with nvof - Solid green square when Spidey descends.
Still doesn't fix the horizontal fast motion issue (see Amazing Spiderman at 1:52:48). But is otherwise the best overall and smooth with Hugo.
6% with dumb algo best for Spidey.
The others are all better than dumb algo at 6% and 8% respectively. But not as smooth as 8% with nvof or 10%+ with dumb algo.
- 6% with SVP - Flashing yellow/green square when Spidey descends.
- 6% with nvof - Solid green square when he descends.
- 8% with svp - Solid green square when he descends.
Chainik wrote:%APPDATA%\SVP4\override.js
smooth.scene.limits.blocks = 40;
default value is 20
with 40 at least the Hugo intro - the fly over the city and through the train station - is better (almost no scene changes)
btw
smooth.debug.qmode = true;
will draw a color square in the frame corner, when it's red it's a scene change
how is this easier than choosing "6%" in the gui? 
I just switched off the computer damn you Switching it back on now 
Posts found: 101 to 125 of 387