1,926 Reply by flowreen91 (edited by flowreen91 21-09-2024 11:00:09)

  • Thanks: 99
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:

I will put back v1 models and test performance difference now between v1 and v2 properly.

Nice! Please check so we'll conclude if we can fully remove/ignore the v2 models.
To test the performance also try to set target fps to something high
and add in mpv config:

# D3D11 renderer (default) is required for the HDR playback
gpu-api=d3d11
# disable VSync so MPV will be able to output more frames than your monitor supports
d3d11-sync-interval=0
video-sync=audio

to disable the fps limit. (probably you will have to resize to a smaller than 4K resolution) xD

"RIFE v2 models which handle paddings internally and reduce memory transactions on heterogeneous devices."
Every time i tested on my setup, v2 was a bit faster.
The developer explanation of the difference between them can be found here:
https://github.com/AmusementClub/vs-mlrt/issues/66

Example 1:
https://gyazo.com/366e7ecfd2d4180d45062b9668449c23
This combination of 1 rife thread + v2 gives me 200 fps.
Same combination of 1 rife thread + v1 gives me 180 fps. sad

Nvidia App Overlay displays current FPS, GPU%, CPU% while you are watching that video maybe that helps.

1,927 Reply by Blackfyre (edited by Blackfyre 15-09-2024 19:36:23)

  • Thanks: 153
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

That's more advanced testing than what I do ^ I'm just doing simple testing with samples I am familiar with.

The way I am testing is using MSI Afterburner to monitor GPU usage while using v1 and v2

Initial quick test, I can confirm 4.16 Lite and 4.16 Lite v2 are both performing similar (GPU usage between 82% and 90% when using full 4K), so SVP Devs are maybe right, and my ORIGINAL post before which I edited was most likely correct too. (Edit: check comment below, more powerful Rife 4.18 showed performance difference better)

Because I was testing with v1 and v2 after the update and was seeing similar results, but then I discovered the screen shake problem, which made me change vsmlrt.py and that caused performance changes because it wasn't working properly? (not sure, need more testing).

So the only problem that needs to be fixed is the screen shake issue? (Edit again, nope, v2 performs better still, so might be time SVP add it as an option).

The fixed files posted on the last page work fine.

For reference, there was a time when the performance below would have been 100% gpu usage on v1 and 82% to 90% on v2, that seems to be no longer the case, it's not as big a gap now? (Edit, again there is a performance difference, as seen below).

https://i.imgur.com/gLb3ePD.png

1,928 Reply by Blackfyre (edited by Blackfyre 15-09-2024 11:37:25)

  • Thanks: 153
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

v2 shows a performance difference when using 4.18 vs 4.18 v2 while watching 4K not full screen (Rings of Power season 2).

So there is certainly better performance for v2 model here @flowreen91 achieving the same results while using around 7% to 10% less GPU power, which is a big difference.

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

1,929 Reply by Blackfyre (edited by Blackfyre 15-09-2024 12:04:27)

  • Thanks: 153
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

flowreen91 wrote:

Nice! Please check so we'll conclude if we can fully remove/ignore the v2 models.

So from the simple tests I done above, there is a clear difference between v1 and v2 as shown when testing with model 4.18 v1 and v2 and that is at 4K (3840 x 1600 resolution).

If someone can do the more advanced testing that you mentioned which lowers the resolution more and pushes for very high frame rates, the difference should be very clear between them when comparing frame rates as opposed to GPU usage like me.

Testing should be done with 4.18 v1 and v2

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

flowreen91 wrote:
Blackfyre wrote:

I will put back v1 models and test performance difference now between v1 and v2 properly.

Nice! Please check so we'll conclude if we can fully remove/ignore the v2 models.
To test the performance also try to set target fps to 478 (mpv or RIFE crashes at 479)
and add in mpv config:

# D3D11 renderer (default) is required for the HDR playback
gpu-api=d3d11
# disable VSync so MPV will be able to output more frames than your monitor supports
d3d11-sync-interval=0
video-sync=audio

to disable the fps limit. (probably you will have to resize to a smaller than 4K resolution) xD

"RIFE v2 models which handle paddings internally and reduce memory transactions on heterogeneous devices."
Every time i tested on my setup, v2 was a bit faster.
The developer explanation of the difference between them can be found here:
https://github.com/AmusementClub/vs-mlrt/issues/66

Example 1:
https://gyazo.com/366e7ecfd2d4180d45062b9668449c23
This combination of 1 rife thread + v2 gives me 200 fps.
Same combination of 1 rife thread + v1 gives me 180 fps. sad

Nvidia App Overlay displays current FPS, GPU%, CPU% while you are watching that video maybe that helps.

Thanks for the link. It reminded me of the name of the Akarin plugin which makes V1 seek performance similar to V2. V2 is still better if you have a GPU with limited memory bandwidth. Your testing shows that V2 still a little faster overall.

1,931 Reply by oriento

  • Thanks: 15
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

to test the performance you can use the transcoding feature

1,932 Reply by oriento

  • Thanks: 15
  • Thanks for the post: 3

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

new release

4.25beta - 21M - 2024.09.19 | [Google Drive](https://drive.google.com/file/d/10RhXyy … share_link) | I am trying using more flow blocks, so the scale_list will change accordingly.

1,933 Reply by dawkinscm (edited by dawkinscm 19-09-2024 07:13:12)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

oriento wrote:

new release

4.25beta - 21M - 2024.09.19 | [Google Drive](https://drive.google.com/file/d/10RhXyy … share_link) | I am trying using more flow blocks, so the scale_list will change accordingly.

I've been testing for a bit. So far it might be an improvement and a possible replacement candidate for v4.18 smile As with the last tested model it works best with a lower SC threshold. But when I do that, Hugo is then negatively impacted.

1,934 Reply by scb

  • scb
  • Member
  • Offline
  • Thanks: 8

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

4.25 seems good on performance!

1,935 Reply by oriento (edited by oriento 22-09-2024 07:36:14)

  • Thanks: 15

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

scb wrote:

4.25 seems good on performance!

yes, now it's 15% less fps than 4.22 lite v2 !

1,936 Reply by Xenocyde

  • Thanks: 47

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Tested 4.25 with my usual interference pattern clip and it seems to have the fewest artifacts thus far, certainly fewer than 4.18. Not sure about GPU load.

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

The v4.25 model is not complete and yet it is still a little better than v4.18. The next version will be the complete model and hopefully be even better smile

1,938 Reply by Drakko01

  • Thanks: 49

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Totally agree, I  test the new model v4.25 and very pleased with the results, i'll use as my main. Like dawkinscm hope that the next model continuing improving.

1,939 Reply by flowreen91 (edited by flowreen91 21-09-2024 10:57:35)

  • Thanks: 99

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Nice! A new SVP update!

1,940 Reply by Chainik

  • Thanks: 1622
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

just a bug fix for a rare situation RIFE + HDR + SVP's tone mapping

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Chainik wrote:

just a bug fix for a rare situation RIFE + HDR + SVP's tone mapping

Any fixes coming for SVP/NVOF scene detection? IC works well and is working even better with the latest models.But it would be nice to have a choice.

1,942 Reply by scb

  • scb
  • Member
  • Offline
  • Thanks: 8

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Is there any guidance on which type of scene detection to use and when?

Is it an advantage using NVOF if you own an NVIDIA card?

1,943 Reply by John86

  • Thanks: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

oriento wrote:

new release

4.25beta - 21M - 2024.09.19 | [Google Drive](https://drive.google.com/file/d/10RhXyy … share_link) | I am trying using more flow blocks, so the scale_list will change accordingly.

how to install it? Previously I just dragged the .onnx file into the models/rife folder

1,944 Reply by Drakko01

  • Thanks: 49

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

John86 wrote:
oriento wrote:

new release

4.25beta - 21M - 2024.09.19 | [Google Drive](https://drive.google.com/file/d/10RhXyy … share_link) | I am trying using more flow blocks, so the scale_list will change accordingly.

how to install it? Previously I just dragged the .onnx file into the models/rife folder

Same as any otb other model, Drop in his respective folder, reopen if you have open Spv and selected in list

1,945 Reply by flowreen91 (edited by flowreen91 21-09-2024 16:53:10)

  • Thanks: 99
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

John86 wrote:

how to install it? Previously I just dragged the .onnx file into the models/rife folder

Ignore the google drive link.
You go to models page and get rife that contains the .onnx files here:
https://github.com/AmusementClub/vs-mlr … nal-models

  • Thanks: 2
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

oriento wrote:
scb wrote:

4.25 seems good on performance!

yes, now it's the same performance as 4.22 lite v2 !

almost...
4.22 lite v2 vs 4.25 v2 on 2080ti, 1920x1080@23.976 anime, at x3 (->71.928) plus Anime4K, on 3440x1440 (effective video resolution after upscaling is 2560x1440):
https://files.catbox.moe/q9my6b.png
https://files.catbox.moe/2nim5m.png

1,947 Reply by oriento

  • Thanks: 15
  • Thanks for the post: 1

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

another update (no changelog)
4.26 - 21M - 2024.09.21 | Google Drive 百度网盘

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Link not there above, where is everyone getting the Google Drive links for beta and alpha versions? I can't see on github a link to them.

1,949 Reply by dawkinscm (edited by dawkinscm 21-09-2024 22:48:31)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

oriento wrote:

another update (no changelog)
4.26 - 21M - 2024.09.21 | Google Drive 百度网盘

v4.26 is supposed to be the completion of v4.25 training from what I remember so it should hopefully be even better.

Blackfyre wrote:

Link not there above, where is everyone getting the Google Drive links for beta and alpha versions? I can't see on github a link to them.

That's because it is not there. The Google Drive link is the Rife model. But it needs to be converted into and onnx engine file before we can use it in SVP. Hopefully it will get done before Monday. If not then it will almost certainly be done by then.

1,950 Reply by oriento (edited by oriento 22-09-2024 07:12:16)

  • Thanks: 15

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

the onnx file is available now

dawkinscm wrote:
oriento wrote:

the onnx file is available now

Yep. But it does not work properly with SVP. It might look like it is working but the engine parse crashes. Looking at the new vsmlrt script there are a lot of changes and I'm not sure which ones are relevant. Especially since SVP moved some functionality into helpers.py.

i suppose you must change 32 into 64

multiple_frac = 32 / Fraction(scale)