You are not logged in. Please login or register.
SmoothVideo Project → Posts by dlr5668
I recently updated to 22H2 build and my mpv RIFE started lagging with >50% GPU usage. Solved that with these tweaks
vo = gpu
gpu-context = d3d11
d3d11-exclusive-fs = yes
hwdec = auto-copy
fullscreen exclusive should die. 99% DX11 apps can work in flip model where borderless is not worse than exclusive
weiw26 wrote:Thanks a lot! This is the perfect solution now. I'm using a 2080TI, and I can get full 1080p interpolated with RIFE 4.6+FSRCNNX+Krig running flawlessly. I cannot believe it! GPU memory usage is ~6GB and utilization ~80%. No need to get 4090 for now thank god!
VeniVediVeci wrote:Now I am using a project named mpv_lazy in github(https://github.com/hooke007/MPV_lazy, in Chinese), and it enables RFIE v4.6 using tensorrt. My video card is 3070 laptop and it works totally fine in 1080p real-time doubling FPS, the performance of which is much better than the SVP4. Hope the svp4 can support the trt RFIE as soon as possble.
Can confirm that this project gives me frame doubling at around ~15% GPU use with MPV. I however haven't found how to go beyond doubling in that build, since everything's in chinese.
This is extremely promising though!
Edit : I'm not entirely sure how relevant that is but I've tried speeding up the footage and I can do up to 4.5* footage speed before stutters happen (compute use goes to 65%).
mpv-lazy\portable_config\vs\rife_cuda.vpy
Find FPS_num = 2 line and increase. Thats assuming you run cuda model (800 mb archive)
Works a bit faster that current SVP for me as well
>Performance: Up to 150 fps at 4K resolution*
А разве старый похожую скорость не давал ? Никто не мерял задержку на предыдущих сериях видеокарт ?
No way it can be 50% faster. lets wait for Chainik install script since SVP lost compatibility with cuda rife
Chainik wrote:> Although, anywhere above 48 fps my GPU usage sky rockets from 46% to 90%
because x2.5 interpolation (i.e. 24->60 fps) need generating TWO times more intermediate frames than x2 (24->48)
> I added another model from github and when selected I get a flickering sensation on any video
then add another model
and post a bug report on the Github regarding the flickering one
The flickering is due to the incompatibility between current rife_vs.dll and model 4.6. I can reproduce this flickering problem on Windows 11. Since most other people are ok, maybe the problem is with Win 11 but not with Win 10.
Shouldnt be a problem. I run win 11 as well
Yep rife works properly with mpv only
Chainik wrote:I mean, switch the video profile to movie x2 and start transcoding. Will it only give 72 fps?
Yes that's the case:
Something wrong here. 4090 should be twice as fast (at least). Note that a have few vms in background + game
Hello
In SVP the option "scene change threshold" set to 15% has resolved the slowdowns in fast scenes, I think it sould be set to this value by default.
I am happy with AI model 4.6, there is a noticable improvement since model 4.0. There are still some artifacts (like on vertical bars or subtitles) maybe it could be improved with futur models.
Thanks for your work!
Subs shouldnt affect interpolation. Just get release without hardcoded ones
I found out that svp index changes constantly between 0.98 to 1.01 when i use rife. Is that normal ?
do you see dropped frames ? yes -> bad, no -> good
5) Попробуй Nvenc H265 max bitrate
4) Там есть Loseless флаг. Это точно, я даже пару раз использовал
3) 3070ти умеет 3х (та же нагрузка как 2.5) с небольшим даунскейлом вроде 900р. 2.5x кстати в 2 раза сложнее чем 2х
2) https://www.svp-team.com/forum/viewtopi … 1&p=13 мы вот тут обсуждали сборку. Если задать число пискселей видеокарта обрабатывает в секунду, то оно само даунскейлит
maxIpps = 65000000 # Max interpolated pixels per second the GPU is capable of
3070ти умеет 65кк, 1080ти около 40кк
Оно в итого неплохо работает
1) 
dlr5668 wrote:aloola wrote:you can also resize the video to 1600x900 then x3 fps. Work for me, I have a 3070ti.
also, I wish "ignore HFR source" profile has an option to turn off the "alter video frame size".
I think https://0bin.net/paste/NKODt3ka#aCSb+rh … Uu7S3e2JiD can be ported to svp as well. Ideally it should track dropped frames and reduce resolution based on these. Current iteration just uses hardcode value
maxIpps = 65000000 # Max interpolated pixels per second the GPU is capable of
I watch 720p and below (old content) x3 and everything else x2 (1080p no changes and 4k downscale to about 1440p)
I'd like to try this out. Where would I put this exactly? I already tried in mpv/scripts however, I get an error regarding the coding in mpv's console. "Unexpected symbol near "#". So I removed any "#" but then I get hit with another error: '=' expected near 'vapoursynth'. Any help would be much appreciated.
I am not sure if it can be used directly. https://www.svp-team.com/forum/viewtopi … 051#p81051 better try portable mpv. Can be unpacked anywhere
Chainik wrote:none is better or smoother
difference is more threads could rise GPU utilization in case it's too low while output fps is not enoughWith 2 threads it uses around 50-55 % of my gpu. With 4 it goes up to 60. Which one should i use ?
Thats interesting. 2 threads @ 3070ti maxes it out. I guess you need more for 3090/4090. Watch for power usage instead of load
Insindro wrote:Chainik wrote:> Although, anywhere above 48 fps my GPU usage sky rockets from 46% to 90%
because x2.5 interpolation (i.e. 24->60 fps) need generating TWO times more intermediate frames than x2 (24->48)
> I added another model from github and when selected I get a flickering sensation on any video
then add another model
Ah yes I understand now. I was just basing off from what you said on the rife AI wiki: "1080p@60-72 atleast rtx 3080". I guess a better video card is needed to achieve those frames without dealing with the drop in frames. Also the Github I was testing with was the one you provided on the wiki, all of them flicker through svp, which is odd. I wonder if I'm the only one. Nonetheless, this is better than anything I've used previously.
you can also resize the video to 1600x900 then x3 fps. Work for me, I have a 3070ti.
also, I wish "ignore HFR source" profile has an option to turn off the "alter video frame size".
I think https://0bin.net/paste/NKODt3ka#aCSb+rh … Uu7S3e2JiD can be ported to svp as well. Ideally it should track dropped frames and reduce resolution based on these. Current iteration just uses hardcode value
maxIpps = 65000000 # Max interpolated pixels per second the GPU is capable of
I watch 720p and below (old content) x3 and everything else x2 (1080p no changes and 4k downscale to about 1440p)
updated portable rife mpv for 4.4/4.6
Download https://www.playbook.com/s/vadash/ssp9p … ben53VKR1k
VT https://www.virustotal.com/gui/file-ana … M5ODg3Mg==
Original script by Elexor. It can do some cool stuff like auto scale resolution based on GPU performance. This way you can use x3 interpolation even without RTX 3080 (minimum for 1080p)
> and leave 4.4
why 4.4? just because the numbers are higher than 4.0?
ok, 4.0 was removed from the list, but 4.2, 4.3 and 4.4 are still there, w/o any changelog or something like that
is 4.4 better than 4.2?
From flowframes discord
quality 2D:
2.3 > 4.6 ≥ 4.5 > 4.4 > 4.0 > 4.3 > 2.4 > 4.1 > 3.1 > 3.8 > 3.9
quality 3D:
2.3 > 4.6 > 4.5 > 4.3 > 4.4 > 4.0 > 2.4 > 4.1 > 3.1 > 3.9 > 3.8
simple pattern resilience 2D:
4.0 ≥ 3.9 ≥ 3.1 ≥ 3.8 > 4.6 > 4.5 ≥ 4.3 > 4.1 > 2.3 > 2.4 > 4.4
complex pattern resilience 2D:
4.0 > 4.6 ≥ 4.5 ≥ 2.4 > 4.3 > 4.1 > 4.4 > 3.9 > 3.1 > 2.3 > 3.8
pattern resilience 3D:
2.3 > 3.1 > 2.4 > 4.6 > 4.5 > 3.8 > 4.3 > 4.1 > 4.4 > 3.9 > 4.0
speed:
4.0-4.4 > 4.6/4.5 > 3.9 > 3.8 > 3.1 > 2.3/2.4
note:
quality = sharpness and accuracy
2.3 CUDA > 2.3 NCNN | complex pattern resilience 2D: 4.0 > 2.3 CUDA > rest > 2.3 NCNN > 3.x
pattern resilience 2D/3D: 2.3 TTA >> everything
TTA reduces sharpness, ensemble does not
4.x is preffered over 2.x/3.x in UHD mode (default ≥1600p)> Do you plan to update SVP with the newer models?
kind of... not exactly like this
there will be only ONE model bundled (*), plus you'll be able to add whatever model you want manually, by copying model's folder(*) if you _know_ which one from 4.0 to 4.4 is the best in terms of overall quality - this is the right moment to tell
Why should we stick with one ? Remove old ones and leave 4.4 and 4.6 (17,21)
If someone wants to try new models
0 update SVP
1 Download dll and source https://github.com/styler00dollar/Vapou … /r9_mod_v3
2 Unpack models C:\Program Files (x86)\SVP 4\rife\models
Model 4.6 is 7% slower than default svp model
del
This is great news!
Until now, I thought there would be some problems based on statements from n00mkrad, who uses RIFE in his Flowframes:
n00mkrad wrote:HDR is currently not supported as the neural networks only work with 8bpp content.
I can't say if there will ever be HDR support.
I tried this one https://4kmedia.org/lg-new-york-hdr-uhd-4k-demo/
18 fps or 1/3 realtime for 4K HDR x2 fps 3070ti
and it lost HDR data
Chainik wrote:"Avisynth Filter" is a confusing naming cause there're _two_ different filters there, one for Avisynth and one for Vapoursynth
both supporting 10-bit and hdrThanks for your reply. I'm guessing you mean:
avisynth_filter: https://github.com/CrendKing/avisynth_filterSo if I understand correctly whichever combination:
AviSynthPlus-RIFE +avisynth_filter
or
VapourSynth-RIFE-ncnn-Vulkan +avisynth_filtershould be sufficient in SVP for real-time interpolation of 4K 10-bit HDR files using RIFE?
Only for 4090 4k@24fps x2
ну дай т.н. б-г
и всего-то от 1000 евроа я вообще думал что dlss это суперсэмплинг, ну и fps растет за счет того что рендер идет в маленьком разрешении
при чем тут optical flow? или тут теперь дополнительно к суперсэмплингу еще и fps прямо увеличивают интерполяцией?
ага dlss3 состоит из 2 частей - увеличение разрешения и вставка доп кадров как SVP
Новый улучшенный (и более быстрый) Optical flow
I fixed that problem. Apparently Windows has it's own version of Python hidden in "WindowsApps". I got rid of it, and now that error is gone, but it's still not working.
Now when I try to play a video it plays for a half a second and then freezes indefinitely. There aren't any errors in the logs when this happens.
EDIT: So, in Plex it freezes as I said, but in MPC-HC, it just doesn't work. The performance monitor shows the SVP Index at a constant 0.5 (normal framerate, since I have RIFE set to x2). Setting the profile back to Automatic works as expected, it's just RIFE that isn't working (both the CUDA version and the normal version of RIFE in MPC-HC, although in Plex, the normal version works, and CUDA version freezes playback).
EDIT 2: I just found the note at the bottom of the wiki page saying it doesn't work for real-time transcoding, which is what I've been testing it with. That's unfortunate. Any plans to make that work? Also maybe getting it to use the tensor cores?
pytorch version is obsolete. RTX 3xxx works better with VS
Just use x3 and it will work. For example I use x6 for 144hz display (its actually 165 but x6 multiplier is superior)
Bonus points for forcing exact rate with CRU
SmoothVideo Project → Posts by dlr5668
Powered by PunBB, supported by Informer Technologies, Inc.