You are not logged in. Please login or register.
SmoothVideo Project → Using SVP → New RIFE filter - 3x faster AI interpolation possible in SVP!!!
all the math (with any artifacts) are inside RIFE itself, Vapoursynth or Avisynth just runs it
Honest question , Do you mean every version use this math in the exact way? Because I test 10 movies of my collection and look better than before.
What do you think?
>rife_v4.26.7z 18.7 MB Sep 22, 2024
It's been 9 months. When 4.27?
>rife_v4.26.7z 18.7 MB Sep 22, 2024
It's been 9 months. When 4.27?
Hello
Please tell, what RTX 5xxx GPU is enough for 4K@30fps & FHD@60fps real-time RIFE frame doubling?
Is there any comparison testing of modern video cards in SVP RIFE?
I am using RIFE 4.6 model and this allows me to do 60fps 4k with madVR but I am left with a decent amount of glitches on the screen with latices or straight lines or text. These visual problems are distracting. I tried 4.25 but too many frames were being dropped. What can I try? RTX 5080. GPU threads of 2, performance boost enabled, scene change detection disabled and duplicate frames removal do not remove.
Any plans for supporting other vsmlrt backends? MIGX support for Radeon gpus has improved a lot and even TensorRT is being replaced with TRT-RTX.
> MIGX support for Radeon gpus has improved a lot
is it faster than ncnn/Vulkan ?
Hello
Please tell, what RTX 5xxx GPU is enough for 4K@30fps & FHD@60fps real-time RIFE frame doubling?
Is there any comparison testing of modern video cards in SVP RIFE?
Using RIFE 2.6 (v2) - found it on the net... https://github.com/AmusementClub/vs-mlr … e_v2_v4.7z
NO: 4K 24fps -> 60fps on my RTX 4090 over 100% GPU core load.
YES: 4K 24fps -> 60fps on my RTX 5090 for about 70% GPU core load.
YES: FHD 24fps -> 120fps on the RTX 4090 with a high GPU core load.
YES: FHD 24fps -> 120fp on the RTX 5090 at 35% for the GPU core load.
Using RIFE 2.26 (v2)
NO: 4K 24fps -> 60fps on my RTX 5090 due to over 100% GPU core load.
YES: FHD 24fps -> 120fps on my RTX 5090 at 50% GPU core load.
As for GPUs
5090 is minimum 30% faster in Tensor computations than 4090
4090 is perhaps 10-15% faster in Tensor computations than 5080
Can Smooth Motion be used with RIFE or SVP somehow? It just got officially enabled for RTX 4000 GPUs in addition to the initial RTX 5000 models.
Can Smooth Motion be used with RIFE or SVP somehow? It just got officially enabled for RTX 4000 GPUs in addition to the initial RTX 5000 models.
I have smooth motion on globally but I can't get my media player or anything to recognize it. However this reminded me to turn on smooth motion option in mpv and the even when I'm using remote desktop as well as the TV, the micro-stutters are reduced or gone. I used this setting before, but I probably over-complicated it by using multiple options when I should have just left it at default. So if this remains the case then I don't need RTX smooth motion anyway 
Xenocyde wrote:Can Smooth Motion be used with RIFE or SVP somehow? It just got officially enabled for RTX 4000 GPUs in addition to the initial RTX 5000 models.
I have smooth motion on globally but I can't get my media player or anything to recognize it. However this reminded me to turn on smooth motion option in mpv and the even when I'm using remote desktop as well as the TV, the micro-stutters are reduced or gone. I used this setting before, but I probably over-complicated it by using multiple options when I should have just left it at default. So if this remains the case then I don't need RTX smooth motion anyway
Where is the smooth motion option in MPV? Does it need to be added to the config file?
dawkinscm wrote:Xenocyde wrote:Can Smooth Motion be used with RIFE or SVP somehow? It just got officially enabled for RTX 4000 GPUs in addition to the initial RTX 5000 models.
I have smooth motion on globally but I can't get my media player or anything to recognize it. However this reminded me to turn on smooth motion option in mpv and the even when I'm using remote desktop as well as the TV, the micro-stutters are reduced or gone. I used this setting before, but I probably over-complicated it by using multiple options when I should have just left it at default. So if this remains the case then I don't need RTX smooth motion anyway
Where is the smooth motion option in MPV? Does it need to be added to the config file?
Yes add the command "video-sync=display-resample" to your config.
This command by itself is not software interpolation and different command options can be used to be more like it. But when combined with SVP I had an instant reduction of micro-stutters, frame lagging and an overall improvement in smoothness, even with my GPU threads turned down to 1. YMMV.
and even TensorRT is being replaced with TRT-RTX.
it's not "replaced", it's in testing...
so far the vsmlrt scripts are buggy, but the main problem is what I can see on my laptop is TRT-RTX gets 23 fps where TRT gets 40 fps
if and when it WILL work correctly, the only reason to use it is runtime size of 150 MB instead of 1500 MB for full CUDA runtime
I replaced my graphics card from an RTX 3080 to an RTX 5080. However, it takes a long time to generate cache.
Has anyone had a similar experience or knows of a way to reduce the time it takes?
rtx3080
4.25_v2 opt level 3 > engine build time ~60 seconds
rtx5080
4.25_v2 opt level 3 > engine build time ~170 seconds
4.25_v2 opt level 1 > engine build time ~110 seconds
rtx3080
4.25_v2 opt level 3 > engine build time ~60 seconds
rtx5080
4.25_v2 opt level 3 > engine build time ~170 seconds
4.25_v2 opt level 1 > engine build time ~110 seconds
Not sure, but build time on the new card doesn't seem excessively higher.
Is anyone else experiencing multiple crashes with the last few versions of SVP? It just disappears from my Windows system tray and when I look in Task Manager, the process is no longer there.
Is anyone else experiencing multiple crashes with the last few versions of SVP? It just disappears from my Windows system tray and when I look in Task Manager, the process is no longer there.
Think I updated to latest version a few days ago IIRC. No problems here.
dawkinscm wrote:Is anyone else experiencing multiple crashes with the last few versions of SVP? It just disappears from my Windows system tray and when I look in Task Manager, the process is no longer there.
Think I updated to latest version a few days ago IIRC. No problems here.
It's been crashing the last couple of versions for me. If no one else is experiencing this then maybe it doesn't like something on my system.
Is RIFE v4.26v2 pretty much peak RIFE capability?
I know the author moved on to other things but do we have the best thing now anyway?
Perhaps we just need GPU power to catch up
Is RIFE v4.26v2 pretty much peak RIFE capability?
I know the author moved on to other things but do we have the best thing now anyway?
Perhaps we just need GPU power to catch up
4.25.
Not the best. But it's the best we have.
GPU power won't solve artifacts. Better models will.
it's really unfortunate that the author moved on to something else. It is still not perfect. I would really love a progress in ai interpolation. With rife or something else (real time)
The heavy models should have been the best, but they turned out to be worse than the "lite" models. v26 was supposed to be better than v25 but it isn't. He did try but guess he got as far as he could.
GPU power won't solve artifacts. Better models will.
Maybe the SVP team should look into implementing something like this? https://www.svp-team.com/forum/viewtopic.php?id=7466 This post is so underrated and probably slipped under the radar.
Hi. New to SVP. Looking for general opinions from others about this configuration. Open to ANY suggestions. Video quality is more important than transcode time and file size but naturally less time and reduced file size would be welcomed without compromising visual quality.
Goal is high bit rate to enhance motion via interpolated extra frames. Most of my original files & Topaz outputs are in the 60 bit rate range. I'd like to retain near that after transcoding with SVP.
RTX 5090, i9-14900k & 64GB RAM.
TensorRT
4 GPU threads
4.25 (4.25 heavy FAILS)
H.265/HEVC/NVENC
opts:
b=60000000 (I assume this is another buffer setting) Yes/No?
preset=slow
rc=cbr_hq (using cbr instead of vbr to increase overall bit rate)
level=5.1 (will use 6.1 for 8k)
tier=high (helps honor minrate but falls below although higher than without. Also prevents FAILS using all of these settings)
qmin=15
cq=15
maxrate=63000000
minrate=58000000
bufsize=60000000
color_primaries=bt2020
color_trc=smpte2084
colorspace=bt2020nc
Questions:
(1) I'm transcoding UHD 23.976 fps to 60 fps aka 59.94x. Is it best to select x2.5 instead of 60fps?
(2) Using MediaInfo & other OSD's, all of my untouched & touched videos show video properties like HDR nits, Formats, Writing library, Mastering display info's, Light levels, etc. After transcoding using SVP, none of this info is present anymore. How can I force SVP to retain it? I have seen SVP outputs by others that indeed retain them. No idea how it was done.
Thanks for any suggestions re: my setup and/or questions.
SmoothVideo Project → Using SVP → New RIFE filter - 3x faster AI interpolation possible in SVP!!!
Powered by PunBB, supported by Informer Technologies, Inc.