zerosoul9901 wrote:
dlr5668 wrote:
Chainik wrote:

> while v2 models significantly reduces the playback seek time, improve performance

dunno, I can't see any improvements
not "significantly", not even slightly

----
unpack "rife_v2" into SVP 4\rife\models (i.e. there must be SVP 4\rife\models\rife_v2\rife_v4.6.onnx)
replace SVP 4\rife\vsmlrt.py

switch back to "v1" models - rename "rife_v2" folder to any other name, like "_rife_v2"

v2 takes 7.5/8 and crashes at seek, v1 stays at 3.5/8 as it should

Don't use dynamic shapes, it doesn't seem to work well with v2 models and requires more VRAM usage compared to v1, use static shapes instead, and use flag force_fp16 instead of fp16 during inference, reducing memory usage and engine build time during engine build, you can also add _implementation=2 flag to easily switch v2 models.

Thanks works great. Same VRAM requirements and much faster almost instant rewind

https://i.imgur.com/NM5g5Zk.png

Chainik wrote:

> while v2 models significantly reduces the playback seek time, improve performance

dunno, I can't see any improvements
not "significantly", not even slightly

----
unpack "rife_v2" into SVP 4\rife\models (i.e. there must be SVP 4\rife\models\rife_v2\rife_v4.6.onnx)
replace SVP 4\rife\vsmlrt.py

switch back to "v1" models - rename "rife_v2" folder to any other name, like "_rife_v2"

v2 takes 7.5/8 and crashes at seek, v1 stays at 3.5/8 as it should

Isnt ensemble twice as slow?

29

(3 replies, posted in Using SVP)

mark007 wrote:

Thanks a lot for that. Are those functions available within SVP (ie some BFI option) or would I need to create my own vapoursynth script?

C:\Program Files (x86)\SVP 4\script
look here

Имхо бред использовать такую старую версию. Проще держать копию mpv и смотреть аниме контент там
https://github.com/bloc97/Anime4K/blob/ … uctions.md

Xenocyde wrote:
dlr5668 wrote:
dawkinscm wrote:

I've been trying for a while now to get RIFE running smoothly with my 3080 with MPC-BE. I've given up on that so instead I tried using MPV and there was an instant improvement. It still needed a bit of work and I had to learn about MPV config files so that I could configure it to match the scaling used in my MadVR config. Now it looks great and runs very smoothly.

Nvidia should release soon https://www.youtube.com/watch?v=XA-tQpQqD7U Perfect match for Rife

Video Super Resolution support introduced today with the 531.18 drivers.

and its out. Here is mpc render for it https://github.com/emoose/VideoRenderer … ag/rtx-1.0

Not bad https://imgsli.com/MTU4Mzgx

dawkinscm wrote:

I've been trying for a while now to get RIFE running smoothly with my 3080 with MPC-BE. I've given up on that so instead I tried using MPV and there was an instant improvement. It still needed a bit of work and I had to learn about MPV config files so that I could configure it to match the scaling used in my MadVR config. Now it looks great and runs very smoothly.

Nvidia should release soon https://www.youtube.com/watch?v=XA-tQpQqD7U Perfect match for Rife

hampur6969 wrote:
Chainik wrote:

RIFE не может быть сверху от madvr
madvr выше всех
и да, с помощью madvr можно положить любую видеокарту, если задаться такой целью (т.е. выкрутить всё на максимум)
контролируйте загрузку карты...

Тут согласен. А нет ли примеров подходящих настроек madvr для лучшего сочетания с SVP?

Ставишь https://i.imgur.com/Ywh93L0.png

Zveroboy wrote:

Может кто проверить RTX3060(3070) на версии 4.6 TensorRT в SVPcode  c настройками:
- nvenc (битрейт от 10 мбит)
- фильтр tonemap

Сколько fps выдает в  разрешениях 1080p60 и 2160p60

Есть только такое, упирается в нвенс. При просмотре дает х3 норм. Сорс 4К немного зажат
Без нвенс с самым говяным пресетом дает 88 фпс енкода

https://i.imgur.com/SvMIkBs.png

earl088 wrote:

Isnt DLSS3.0 features like frame gen and nvidia reflex require HAGS to be enabled?
Or what is causing HAGS to make the real time decoding slower?

Reflex no, you can force it in most DX11/12 games for ages https://leonardo.osnova.io/ea1540f2-2fd … 84a5890f51 with GTX900 or newer

HAGS is only needed for frame generation feature of DLSS. In my experience HAGS cause problems with VR games and CUDa workload

36

(1 replies, posted in Using SVP)

Delay, not worth. Most new demanding games have DLSS2 already (gpu boost) and DLSS3 soon (Cpu boost)

37

(6 replies, posted in Using SVP)

cemaydnlar wrote:

When i activate video-sync=display-resample setting it doubles my chosen fps. Is that normal or am i doing something wrong. I'm using 120hz right now but it interpolates animes or movies to 240 fps.

I prefer to force custom res 23.976 * 6 (144). This way you can watch x2 4k, x3 2k, x6 1080p content without judders no resample needed

onurco wrote:

I have a question regarding the bottleneck when using rife tensorRT engine on higher resolutions. I have 12900K oc to 5.2ghz all cores, 6000mhz DDR5 32gb and RTX 4090. I play 1800p rife on svp with madvr on mpc-hc 4k UHD files which have bitrate around 70-100 mbps. CPU usage is around 20-30%, GPU usage around 50%, total ram usage is around 20gb at 60fps interpolation. If I try higher resolutions, then the framerate cannot keep up anymore but the CPU GPU usage are not so different than before. Can it be the RAM speed or maybe tensor cores are full on 4090 or is there some other reason? I am absolutely fine with playing videos on 1800p, I see no difference with native 4k and 1800p even on 160 inch screen with projector but still I am curious about what is causing this bottleneck.

I would check cuda (disable HAGS to see) and dram load
https://imgur.com/a/0HIN7ZQ

Zveroboy wrote:

Может кто нибудь это видео законвертить в SVP   кодировщике на версии 4.4 https://www.youtube.com/watch?v=I_163o9 … MATYVLOGSL
и выложить результат?
Такое ощущение, что мощности 1080ti  не хватает для сглаживания в сцене с небоскребами.

Это свп рвет кадр сменой сцен. Оно не совсем идеально работает

https://www.playbook.com/s/vadash/wcVXb … Ky5jaQMkLs

На обычном контенте (кино, аниме) работает лучше

onurco wrote:
Chainik wrote:

Performance boost: TensorRT only. Increase performance up to ~5% and lower memory usage in exchange for running a 5-10-minutes optimization pass in a command-line window for _every_ new video resolution. Previous runs are cached. For example, you're opening a 1080p video for the first time, you'll wait 5+ mins. Every other 1080p videos will start instantly, but when the frame size changes due to, for example, black bars cut off, you'll wait another 5+ mins.

With "perf boost" off there's only ONE optimization run for all resolutions below 1440p.


Oh I see, that's great to know that it is being cached for each resolution. I had no issues using vulkan for 1080p but at 2160p it was failing miserably, everything was slow motion. I am mostly excited to use TensorRT for 4k 2160p videos. So I won't use the performance boost mode in the beginning and wait until it is finished then. You wrote that until 1440p, it is only doing one pass without perf boost. How is the situation for 2160p videos?
Also could you send me the documentation link for TensorRT if there are other things to know? I am really excited for this update, thanks a lot for this great achievement Chainik!

Which main thread exactly dlr5668? I would like to try 1800p if I can do it real time without manually downscaling the video.

https://www.svp-team.com/forum/viewtopi … 741#p81741

Chainik wrote:

onurco

if it actually doing something (check CPU/GPU load) then you just have to wait more...

You will also want to reduce 2160p to say 1800p if you cant handle it. Check main thread for how to

42

(2 replies, posted in Using SVP)

1 and 3rd line are not errors

Blackfyre wrote:

I'm upgrading from an RTX 2070 Super to an RTX 3090 (arriving in a week). Can I use RIFE for 4K content on a 60Hz 4K TV with an RTX 3090? Looking forward to finally trying out RIFE and hopefully even more optimisations in the future too. Kudos to the devs and everyone involved, I've supported SVP since the early days and to now it is one of my most used programs <3

wort case you can apply downscaling to say 90% and it will work just fine

Zveroboy wrote:

Качество кодирование при помощи RIFE зависит от видеокарты?  Например если поменять 1080ti на 3070.
С версии 4.4 качество хорошо подтянули.

Стоит поменять потому что 3070 тянет в риалтайме 2к х2. Может на 4070 летом ? Там еще лучше 4к х2

Xenocyde wrote:

Any idea how to increase workspace memory to allow for more tactics? Or do we ignore that warning?

edit this line

clip = RIFE(clip=clip, multi=interpMulti, model=RIFEModel.v4_6, backend=Backend.TRT(fp16=True, device_id=GPU, workspace=WS_size, use_cuda_graph=True, num_streams=GPU_t))

I think 2000 is enough for 4k

grobalt wrote:

i posted some benchmark data with a sample script: https://github.com/AmusementClub/vs-mlr … nt-4630465
numbers should be valid as someone else with RTX4080 is half performance

the parameter peak_performance = True increases performance by ~60%

thanks for the parameter feedback@github smile

peak_performance is just for the bench, right ?

grobalt wrote:

I am new to SVP but due to RIFE implementation I ordered a 4090. If you guide me how to test / benchmark I will show all results

Just install everything and run encode with rife. It will show max fps

aloola wrote:
sofakng wrote:

What GPU is needed for 4k (8-bit)?

Do we need an RTX 4090 or is a “slow” 3080 Ti good enough?

my 3070ti can do x2 1440p at max.

2560*1440*24=88,473,600

3840*2160*24=199,065,600

so you might need a card that is stronger than 3070ti 2.5 times to x2 4k@24p

looks about right. I run 2432x1376 x2 with 15% to spare

sofakng wrote:

What GPU is needed for 4k (8-bit)?

Do we need an RTX 4090 or is a “slow” 3080 Ti good enough?

Evern 3090 is not enough. 4080 is probably fine with it but nobody tested it yet

50

(13 replies, posted in Using SVP)

cemaydnlar wrote:

Do we need to match the monitor hz with svp fps in order to get the best possible view experience on rife interpolation too or can we use x2 x3 x4 rife interpolation with monitor set to 144hz without any negative effects ?

If you have VRR (gsync/freesync) it shouldn't matter