You are not logged in. Please login or register.
SmoothVideo Project → Posts by Santr
Pages 1
Should I change my AI model to 4.26 Heavy?
I'd suggest trying 4.22 Lite and 4.25 Heavy. Both are good quality-wise but while 4.25 Heavy is slightly better, it comes at a very high performance cost. 4.22 Lite has become my default.
With 4.26 models, I've experienced some pretty bad artifacts, so would not recommend. YMMV.
Anyone using 48fps or 72fps videos on a 120hz monitor? It seems to be smooth at times, and 120fps does not seem to have a big difference in comparison
If we're talking 24fps videos interpolated to 48 or 72, then yes, although mine is a 165Hz monitor. But x4 (96fps) is a noticeable step up from x3 (72fps) to my eye. More then that I can also notice but it's not as critical as getting to that x4.
One thing to note is that SVP is only half of the equation. The other half is to draw those frames in sync with your monitor, otherwise even 160fps will remain jittery.
Is there any kind of configuration that works well for cases like this one?
4.25 Heavy produced the least artifacts for me in this clip but there are still some. It's not universally best, though - most content I watch prefers 4.22 Lite.
Out of curiosity, I also tried this clip with older methods using OpticalFlow and straight CPU - and they really don't compare. Hard to imagine I used to watch stuff artifacting like that back in the days.
I unchecked everything except SVPtube
And in that window it explicitly said that you install everything you check and uninstall everything you uncheck. Once you calm down, you can just reinstall everything the same way.
Просто хочу сказать огромное спасибо за то, что допилили SVP под Линух. Давно ждал, но как-то это событие прошло меня стороной. А тут раз - и внезапно полноценный RIFE на Тензорах. Красота! 
О, спасибо, что обратили внимание на настройки. "Без прокси" ничего не изменило, а вот переключение на китайский сервер дало обновиться.
Видимо, основной сервер (CDN) у нас таки блокируют.
При попытке обновления, утилита долго висит в статусе "Подготовка к загрузке метаданных", после чего выдаёт ошибку: "Не удалось загрузить удалённое дерево: Исходники пакета не заданы для данного приложения".
Фаервол отключал, доступ в интернеты проверял. Всё должно фурычить, но нет.
Было подозрение, что это очередной подарок от чудаков из РКН (нахожусь в РФ), но добавление адреса, до которого пытается достучаться утилита обновления, в списки Zapret результата не дало.
I probably should've posted in this topic from the start, but I'm getting this error on the last step of mpv build on Ubuntu 25.04 (everything before went without a hitch):
user@comp:~/mpv-build$ sudo ./install
Install data not found. Run this command in build directory root.
Same exact issue. I've been following the guide to the letter (on Ubuntu 25.04) and getting stuck on "Install data not found. Run this command in build directory root."
v2 models should be put in rife_v2 folder
Thanks, that got it working.
Yep, 4.15l v2 handles 1440p60 no problem. Fantastic.
This is strange, my 4070 handles 4.15l v2 1440p 60x2 without issues...
I'm probably doing something wrong because I can't get v2 model (downloaded here) to run at all:
Last bits of trtexec_250207_183007.log:
[02/07/2025-18:30:10] [W] [TRT] Could not read timing cache from: C:\Users\...\AppData\Roaming\SVP4\cache\Program Files (x86)/SVP 4/rife\models\rife\rife_v4.15_lite.onnx.2560x1440_fp16_no-tf32_trt-100400_I-fp16_O-fp16_NVIDIA-GeForce-RTX-4080-SUPER_b38f6198.engine.cache. A new timing cache will be generated and written.
[02/07/2025-18:30:10] [E] Error[2]: [castLayer.cpp::nvinfer1::CastLayer::validateTypes::45] Error Code 2: Internal Error (Assertion !mOutputTypes.at(0).hasValue() || mOutputTypes.at(0).value() == params.toType failed. )
[02/07/2025-18:30:10] [E] Engine could not be created from network
[02/07/2025-18:30:10] [E] Building engine failed
[02/07/2025-18:30:10] [E] Failed to create engine from model or file.
[02/07/2025-18:30:10] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v100500] [b18] # C:/Program Files (x86)/SVP 4/rife\vsmlrt-cuda\trtexec --onnx=C:/Program Files (x86)/SVP 4/rife\models\rife\rife_v4.15_lite.onnx --timingCacheFile=C:\Users\...\AppData\Roaming\SVP4\cache\Program Files (x86)/SVP 4/rife\models\rife\rife_v4.15_lite.onnx.2560x1440_fp16_no-tf32_trt-100400_I-fp16_O-fp16_NVIDIA-GeForce-RTX-4080-SUPER_b38f6198.engine.cache --device=0 --saveEngine=C:\Users\...\AppData\Roaming\SVP4\cache\Program Files (x86)/SVP 4/rife\models\rife\rife_v4.15_lite.onnx.2560x1440_fp16_no-tf32_trt-100400_I-fp16_O-fp16_NVIDIA-GeForce-RTX-4080-SUPER_b38f6198.engine --shapes=input:1x11x1440x2560 --fp16 --tacticSources=-CUBLAS,-CUBLAS_LT,-CUDNN,+EDGE_MASK_CONVOLUTIONS,+JIT_CONVOLUTIONS --useCudaGraph --noDataTransfers --noTF32 --inputIOFormats=fp16:chw --outputIOFormats=fp16:chw --layerPrecisions=*:fp16 --layerOutputTypes=*:fp16 --precisionConstraints=obey --builderOptimizationLevel=3The 4.4v2 model best on performance, but overall, quality remains questionable! Try everything from 4.4v2 to 4.9v2 or switch to 4.15v2 lite (4.9 perf)!
Tried all of them but it seems current RIFE that comes with SVP isn't quite there to handle it. Both 4.15l and 4.16l fall just 15 fps short of hitting the 120 target.
Now I'm eagerly awaiting for the update as that looks like just what's needed to get there.
Is there a model that can do 2x with a 1440p60 video on a 4080S? I know I can downsize with SVP but I'd like to avoid doing that.
Small PSA: updating AMD drivers from 24.6.1 to 24.7.1 broke it again. Had to revert to 24.6.1 to get the performance back. Sigh, why does AMD hate RIFE so much...
It would be good to know, how to solve this in other players...
Yep. Unfortunately, only those players' developers can solve this issue, since it's tied to the player's renderer. I don't know about PotPlayer's devs but MPC-BE devs have never been receptive to fixing issues they don't encounter themselves, so we're most likely out of luck with them.
That uploaded DLL by Chainik? For me also allowed any threads (4 too) but couldn't fast forward or jump to another time. Is it working for you? same driver version too.
Yep, MPC-BE crashes every time, so I switched to MPV that comes with SVP - no problems with it whatsoever. Works better than MPC-BE, in fact: faster and stable as a rock. But I had to tinker with the config file to set it up to my liking, which is pretty much its biggest downside. The manual is pretty detailed, though: https://mpv.io/manual/stable/#options
> to see if gerappa can use 2 threads if that line was turned into a warning
still the device will be reporting queueCount = 1
https://registry.khronos.org/vulkan/spe … rties.htmlI really doubt there're any performance benefits settings "thread" count higher than device caps
but here you are...
That DLL finally allowed my 7900XTX to work 4 threads and do x4 framerate (24->96). Fantastic. That's with currently latest AMD drivers v24.6.1.
For my RTX 4080 it becomes watchable if i resize it to 540p using
resize -9600540
https://gyazo.com/52607f2009d1dae6c41a2939d7b62c04
Ah, thanks, then it's normal. I'd rather not downsize it, so I guess it's better to stick with old interpolation methods for such heavy cases - those can handle x4 for 1080p60 even at max settings, which is sufficiently smooth.
Я только начал пользоваться RIFE, не знаю его возможностей, поэтому хотел бы узнать: то, что текущая версия в SVP не справляется с удвоением кадров для видео 1080р60 на AMD 7900XTX - это нормально или что-то с моей системой не то? Версия дров 23.11.1, обработка работает в 2 потока.
Никто не в курсе, отчего madVR (205) может люто дропать кадры на связке 7800X3D + 7900XTX? Система при этом едва ли загружена: https://i.imgur.com/hxQ05pk.gif
Отключал FreeSync в дровах и мониторе. Принудительно включал верт.синхронизацию для MPC-BE. Включал Exclusive Fullscreen в madVR.
При использовании MPC VR всё норм, то есть дело явно в отрисовщике.
По типу как это уже реализовано с Nvidia Optical Flow?
Последняя просто отлично проц разгружает. Очень хотелось бы и с АМД таких же плюшек вкусить.
Since you have Nvidia, you could enable its Optical Flow option in SVP - it's a bit less taxing on the CPU.
Try a custom profile like this:
The biggest performance/quality trade here comes from "Motion vectors grid". Smaller is smoother; bigger is less taxing. Personally, I find 8 to be optimal, but YMMV.
Damn, I'm stupid. I entered the full path to filters instead of plugins and didn't even notice it in the error message.
Thanks, now it's working.
Still no go.
I also tried "-R a+rwx". Properties for the folder and files inside show Read&Write permissions for everyone, but SVP still gives the same error.
Linux noob here, running Xubuntu 18.04.1.
I'm having trouble giving SVP write permission for VLC's video_filter folder.
I run
sudo chmod 777 /usr/lib/x86_64-linux-gnu/vlc/plugins/video_filter
in the console and it appears to work fine - no errors or other messages.
But every time I try to activate "SVP in VLC (replace 'deinterlace' filter)" in SVP's menu, I get this error:
Operation failed. Check that SVP Manager has write access to '/usr/lib/x86_64-linux-gnu/vlc/plugins/video_filter/video_filter'.
What am I doing wrong?
Pages 1
SmoothVideo Project → Posts by Santr
Powered by PunBB, supported by Informer Technologies, Inc.
