26 Reply by aloola

(2,112 replies, posted in Using SVP)

oriento wrote:

When i tried rife 4.14 v2 (i just copied the onnx file)
it gives an error on screen, and on a log file
[03/02/2024-21:34:57] [W] Could not read timing cache from: C:\Users\Attila\AppData\Roaming\SVP4\cache\Program Files (x86)/SVP 4/rife\models\rife\rife_v4.14.onnx.min64x64_opt1920x1088_max2560x1440_fp16_no-tf32_trt-8502_cudnn_I-fp16_O-fp16_NVIDIA-GeForce-RTX-2080_9c726c00.engine.cache. A new timing cache will be generated and written.
[03/02/2024-21:34:57] [E] Error[4]: [network.cpp::nvinfer1::Network::validate::3100] Error Code 4: Internal Error (input: for dimension number 1 in profile 0 does not match network definition (got min=11, opt=11, max=11), expected min=opt=max=7).)
[03/02/2024-21:34:57] [E] Error[2]: [builder.cpp::nvinfer1::builder::Builder::buildSerializedNetwork::751] Error Code 2: Internal Error (Assertion engine != nullptr failed. )
[03/02/2024-21:34:57] [E] Engine could not be created from network
[03/02/2024-21:34:57] [E] Building engine failed
[03/02/2024-21:34:57] [E] Failed to create engine from model or file.
[03/02/2024-21:34:57] [E] Engine set up failed

rtx 2080 driver 554.61

build in older versions of rife (like 4.4) works

edit : thank you, i missed that i should create a folder. Sorry for that. I have searched my error on the forum and this one wasn't listed ^^'

edit2: actually my 8gb rtx 2080 is not compatible with 4.14 or 4.14 lite, it tries to allocate exactly 8gb vram, but it can't as the vram already contains a little bit of something

[03/02/2024-23:38:14] [E] Error[2]: [virtualMemoryBuffer.cpp::nvinfer1::StdVirtualMemoryBufferImpl::resizePhysical::145] Error Code 2: OutOfMemory (no further information)
[03/02/2024-23:38:14] [E] Error[1]: [virtualMemoryBuffer.cpp::nvinfer1::StdVirtualMemoryBufferImpl::resizePhysical::132] Error Code 1: Cuda Driver (invalid argument)
[03/02/2024-23:38:14] [W] [TRT] Requested amount of GPU memory (8589934592 bytes) could not be allocated. There may not be enough free memory for allocation to succeed.
[03/02/2024-23:38:14] [W] [TRT] Skipping tactic 1 due to insufficient memory on requested size of 8589934592 detected for tactic 0x0000000000000001.
Try decreasing the workspace size with IBuilderConfig::setMemoryPoolLimit().

edit3: enabling "performance boost" fixes the memory problem

because of this bug https://github.com/AmusementClub/vs-mlrt/issues/72

if you use model v2 v4.7+ you need to turn performance boost on else you will get the error.

Go to forum: Using SVP

27 Reply by aloola

(2,112 replies, posted in Using SVP)

VRR (GSync and Freesync Premium) enabled in the Display and NVCP = Crushed blacks and banding with d3d11va-copy running at 24FPS (even outside the VRR range, so VRR is not even working).

have you tried to turn on "G-sync compatible indicator" in Nvidia control panel?

because I tested with various mpv options, and g-sync always turn on in full screen for me. I also LG C2 48''

the green text on top right is a proof that G-sync is active
https://cdn.discordapp.com/attachments/ … 7ab4d&

ah, I forgot.  press the green button 7 times on your TV remote will show you VRR information. check if the fps there sync with your video fps

Go to forum: Using SVP

28 Reply by aloola

(2,112 replies, posted in Using SVP)

Blackfyre wrote:
Chainik wrote:

Blackfyre

you'd better ask about the differences between d3d11va and d3d11va-copy in relation to VRR and black levels in mpv's Github.

Thanks for the recommendation, yeah, I'll try to write something later tonight.

EDIT:

Just to narrow it down even further, I just done a lot of testing before I write and report the issue later today.

First, I decreased the code to this:

ontop
fullscreen=yes
d3d11-exclusive-fs=yes

vo=gpu-next
gpu-api=d3d11
hwdec=d3d11va-copy
video-sync=audio
gpu-context=d3d11
fbo-format=rgba16hf

target-peak=800
target-contrast=inf
target-colorspace-hint=yes

HDR Enabled:

No SVP running in the background.

VRR (GSync and Freesync Premium) enabled in the Display and NVCP = Crushed blacks and banding with d3d11va-copy running at 24FPS (even outside the VRR range, so VRR is not even working).

Changing to d3d11va solves the problem. No crushed blacks or banding.

DDU uninstall nVidia drivers. Shut down the PC. Turn off GSync and Freesync from the display. Turn the PC on and install NVidia drivers.

Run again with both d3d11va-copy and d3d11va, no problems at all with both. No crushed blacks, and no banding. Everything running smoothly.

All of this without SVP.

Conclusion:

Just enabling VRR causes d3d11va-copy to crush black levels and introduce banding, even when the video playing is outside the VRR range (my range is 40Hz to 120Hz), with the HDR video playing at 24 FPS.

EDIT 2:

What's with all the crazy details they want. I just want to report an issue that is very clearly written without the need to go into so much detail they are requesting.

I click on Report a Windows Issue and it's full of so much details they want, including a LOG. Also if this is the wrong section, the post will be deleted.

So I got to the Ask a Question section, and it says if you report an issue in the ask a question section, your post will be deleted.

Wtf lol

EDIT 3:

Ended up asking it in the Questions Section, so hopefully it wont get removed @Chainik

https://github.com/mpv-player/mpv/issues/13467

have you tried
hwdec=dxva2 and dxva2-copy ?

Go to forum: Using SVP

29 Reply by aloola

(2,112 replies, posted in Using SVP)

RickyAstle98 wrote:

Really why SVP developers cant develop RIFE-SVP swapchain? It will suit for everyone, any type of content, any resolution, almost everything! My general idea about this quiet simple, SVP algorithm almost 10 times less resource-sensitive, but intermediate frames quality rises at high frame rate video source, 2K 24>144fps ~40-60 watts with 8-12 px grid and 21 shader, whats the problem RIFE@4K 2x and then SVP with higher pixel grid? Its going to be really smooth!
With this swapchain, 24>96@4K easy! Double peformance for 4K sources...

it's simple to do but not a good idea. it would produce a lot of artefacts.

Go to forum: Using SVP

30 Reply by aloola

(2,112 replies, posted in Using SVP)

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite
Output 8000 frames in 63.61 seconds (125.76 fps) 4.13
Output 8000 frames in 50.31 seconds (159.01 fps) 4.13 lite
Output 8000 frames in 64.95 seconds (123.18 fps) 4.14
Output 8000 frames in 65.58 seconds (121.99 fps) 4.14 lite

4.14 lite is the most resource-heavy, dunno about the quality though.

Go to forum: Using SVP

31 Reply by aloola

(2,112 replies, posted in Using SVP)

RickyAstle98 wrote:

Sometimes force exact value prevent RIFE working, with new models too (mpv_cfr argument can also kick RIFE from workload) but I only use X factors because I have GSync monitor tech, I have smoothness all the time!
Force exact value = issues
Fixed frame rate = issues
Frame rate factor = no issues

are you using the newest vsmlrt.py?
then it won't work with fraction numbers.
https://github.com/AmusementClub/vs-mlr … 1847153063

Go to forum: Using SVP

32 Reply by aloola

(5 replies, posted in Using SVP)

why do you need to change it there when you can easily do it here?
https://cdn.discordapp.com/attachments/290709370600423424/1192755072694747187/image.png

Go to forum: Using SVP

33 Reply by aloola

(2,112 replies, posted in Using SVP)

TechnoStone wrote:

V2 models for some reason have jiggling, the more to the right and down, the worse the effect. Almost unnoticeable for 1080p, but very annoying for 1440p and 4K (both native and downscaled to 1440p, here's the example - https://cdn.discordapp.com/attachments/ … WGll8w.mp4). For this one, I'm running Game of Thrones S1 E1 at 4K on 3440x1440 (so SVP downscales to 2560x1440) at x2 framerate on 2080 Ti with 4.4v2, 4.6v2, 4.9v2 and 4.13_litev2. There's no such effect on V1 models (tried all same versions)

have you updated to the newest SVP version? the bug should be fixed with the newest version.

https://github.com/AmusementClub/vs-mlr … 1791986979

Go to forum: Using SVP

34 Reply by aloola

(2,112 replies, posted in Using SVP)

hvs wrote:

I hope someone can guide me in getting Soap Opera effect using RIFE in SVP.

currently, the best quality that almost gets rid of the soap effect is models 4.12 and 4.13.

4.0->4.6: about the same quality and same performance
4.7->4.9: about the same quality and same performance, the performance cost +20% than 4.0->4.6
4.10->4.13: about the same quality and same performance, the performance cost +20% than 4.7->4.9
4.12lite, 4.13lite: performance between 4.7->4.9 and 4.10->4.13

my recommended models are: 4.6, 4.9 and 4.13

Go to forum: Using SVP

35 Reply by aloola

(2,112 replies, posted in Using SVP)

V2: faster, should use this as default if you not encountering any problems. model V2 47+ has a bug preventing you from using performance boost off.(It means it will run trt command on every video with a different resolution.)
4.13v1lite: lite is just a name, you may think of it as 4.135 which runs faster than 4.13
ensemble: this one will cost half the performance of the normal model. I couldn't see much difference from this, just use a better model instead.

try to use model in performance order: 4.6 > 4.9 > 4.13 lite > 4.13 and decide which one you like the most

4.9 is a sweet pot for me which has low artifact and runs faster than 4.13 20%

Go to forum: Using SVP

36 Reply by aloola

(2,112 replies, posted in Using SVP)

flowreen91 wrote:
aloola wrote:
Chainik wrote:

sounds like a bad advice that others will follow blindly and get a worse results (as usual when you're following some advices w/o understanding them)

In my case it's improvement.

Is there any simple way to test the difference between any two configurations?
How can we confirm if any change is improvement or not?
Can you give us a basic example on a test that you do where we can see consistent results of numbers/data going up/down on multiple attempts?
Thanks!

simplest way to test is to use MSI Afterburner, and see the GPU power consumption graph.

in my case: 1080p24 + RIFE x3 + DXVA = 200W
1080p24 + RIFE x3 = 150W
lower GPU power consumption means it has more room for RIFE.

if anyone has a better way to test, let me know. smile

Go to forum: Using SVP

37 Reply by aloola

(2,112 replies, posted in Using SVP)

Chainik wrote:

sounds like a bad advice that others will follow blindly and get a worse results (as usual when you're following some advices w/o understanding them)

I agree, they need to test it themself first if they see performance improvement or not.

In my case it's improvement.

But in some cases, if the CPU is so weak, the performance would drop a lot.

well, it costs you nothing to try, and can easily roll back.

Go to forum: Using SVP

38 Reply by aloola

(2,112 replies, posted in Using SVP)

Insindro wrote:

How exactly would I be able to turn off gpu decoder, and use the software decoder if I'm using mpv?

if you're using default mpv, then no need to edit anything. The default is off.


if you're using SVP's mpv then go to C:\Program Files (x86)\SVP 4\mpv64
and edit the file mpv.conf

hwdec=auto-copy change to #hwdec=auto-copy. Just remove the "#" if you want to use it again.

Go to forum: Using SVP

39 Reply by aloola

(2,112 replies, posted in Using SVP)

dawkinscm wrote:
aloola wrote:

Turn off GPU decoder, and only use software decoder in player

Interesting that you say that. I don't know if you mean with mpv player but for SVP this player does work better with hardware decoding off.

with GPU hw decoder it will copy back frames from GPU/VRAM to CPU/RAM, making your RAM bandwidth exhausted.

in general, just turn off GPU hw decoder if you using Vapoursynth with 4K video.

Go to forum: Using SVP

40 Reply by aloola

(2,112 replies, posted in Using SVP)

RAGEdemon wrote:

Hi guys,

I am trying to get 4k content to run at 60fps. Sometimes I manage it, but most times, the best I can do is double the movie FPS. Ideally, 4k120fps would be awesome.

How can I try to do that? I have tried 4.4,4.6,4.9,4.12. 3/4 threads.

I am on a 4090.

I'd appreciate help, thank you smile

Turn off GPU decoder, and only use software decoder in player
Turn off HAGS
use model 4.6 v2

or pre-resize your video to 1440p

Ideally, 4k120fps would be awesome.

not possible for now, based on people's benchmarks here most you can do with 4090 is 24fps x3

Go to forum: Using SVP

41 Reply by aloola

(2,112 replies, posted in Using SVP)

cemaydnlar wrote:

Tried everything seems like v2 does not work on 3070 plus i am having frame drops on 4.13. They should somehow start to find a solution to make it use less gpu.

v2 4.7+ doesn't work with performance boost Disabled, it will crash trt if you combine these options.

Go to forum: Using SVP

42 Reply by aloola

(2,112 replies, posted in Using SVP)

aostang wrote:

The seek hanging has annoyed me for ages.

Using model V2 should fix the seeking.

Go to forum: Using SVP

43 Reply by aloola

(2,112 replies, posted in Using SVP)

it looks like TRT 9.2 fixed the crashed issues when seeking

TRT 9.2 has build-up time faster than TRT 8.xx
https://github.com/AmusementClub/vs-mlrt/discussions/65

need more testing though

Go to forum: Using SVP

44 Reply by aloola

(2,112 replies, posted in Using SVP)

reconhell wrote:

Why does my SVP screen does not show "engine" or "performance boost"?
Also I tried this and i have the files on location like in screenshots. but the dropdown box in SVP doesnt show any new option...

because you haven't installed RIFE trt. there are 2 RIFE versions in SVP

1. RIFE NCNN or generic, which can run on AMD/NVIDIA card but slow

2. RIFE TRT (most of ppl here using it) only run on NVIDIA cards and is faster a lot than RIFE NCNN

https://www.svp-team.com/wiki/RIFE_AI_interpolation





Blackfyre wrote:

@Chainik regarding "Scene Change Threshold", it's interesting how it works like artifact masking in the past. 15% provides the smoothest panning shots and the most artifacts. 6% shows the least artifacts, but less smoothness. 1% (set manually from all settings) has ZERO artifacts, but LEAST smoothness.

It depends on the content actually, animation should pick a high number while the real scenes should pick a lower number.

in my case, I disabled the scene change threshold and picked the RIFE version that has fewer artifacts.

Go to forum: Using SVP

45 Reply by aloola

(14 replies, posted in Using SVP)

try to add
Expert: add scores = 10
higher numbers = higher priority

Go to forum: Using SVP

46 Reply by aloola

(2,112 replies, posted in Using SVP)

dawkinscm wrote:
aloola wrote:

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

The aim of the 4.12 Lite version is to make it less GPU intensive of 4.9. So it should arguably be called 4.9 Lite. But I'm just glad they are releasing updates and trying things.

have you tested with SVP? The SVP script needs to be updated to work with 4.12 lite.
you can easily check this by going to the cache folder to see if there is any model cache name 4.12_lite

Go to forum: Using SVP

47 Reply by aloola

(2,112 replies, posted in Using SVP)

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

Go to forum: Using SVP

48 Reply by aloola

(2,112 replies, posted in Using SVP)

donnieyeen wrote:
aloola wrote:
donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

9.1 just won't work and it won't boost RIFE performance. so well let's wait for trt 10
https://github.com/AmusementClub/vs-mlrt/discussions/65

I've had 9.1.0 working fine on previous version before this last update of SVP. And I was talking about the time it takes before it generated  the engine cache files, its faster then 8.6.

I know it works but it has a critical bug, crashed on seek or crashed when entering fullscreen. at least I tested with mpv. haven't tested with mpc yet.

Go to forum: Using SVP

49 Reply by aloola

(2,112 replies, posted in Using SVP)

donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

9.1 just won't work and it won't boost RIFE performance. so well let's wait for trt 10
https://github.com/AmusementClub/vs-mlrt/discussions/65

Go to forum: Using SVP

50 Reply by aloola

(2,112 replies, posted in Using SVP)

Chainik wrote:

This log is with an old helpers.py

OK v2 4.6 works now! thanks!

damn, I'm pretty sure I updated the file before...

also, V2 doesn't need add/crop border, I think tweaking this will give us more performance boost.

EDIT: V2 4.9 doesn't work due to not enough GPU memory, hmm weird??
https://cdn.discordapp.com/attachments/ … 175352.log

edit2: with performance boost enabled it works.
edit3: would be nice if we have an option for workspace, the default value None might cause some GPUs to run out of memory, I would like 1024, 2048, 4096 value options wink

Go to forum: Using SVP