2,476 (edited by dawkinscm 23-12-2025 19:46:38)

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

oriento wrote:

a 5090 cannot do 120fps for 4k

Yes it can, yes it does, can actually go way beyond that and in 3D too. But maybe SVP is lying to me.

oriento wrote:

a 5090 cannot do 120fps for 4k

edit: i have tested right now, and rife 4.25v2 gives me 85fps at transcoding 4k with my 5090

Same here. But I'm not using Rife which is what I've been saying every post for the past week including the one below I posted just yesterday:

dawkinscm wrote:

Rife 26 heavy is stuck at around 80fps although I haven't pushed it recently. Don't worry about v2 models, especially with a 5090.

2,477

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

sorry, I misunderstood you, I thought you said him to configure 120hz at real time

2,478 (edited by dawkinscm 23-12-2025 21:16:27)

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

oriento wrote:

sorry, I misunderstood you, I thought you said him to configure 120hz at real time

Yes SVP real time at 120 is no problem without Rife. I know SVP without Rife can run at even 240fps because I have done it. But with Rife 25+ SVP tops out at around 80fps.

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

Why doesn't SVP update the AI Model? The latest I have on there is 4.18. I don't see 4.25.

2,480

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

kellykline wrote:

Why doesn't SVP update the AI Model? The latest I have on there is 4.18. I don't see 4.25.

Download and add manually those that you want to use.

It's about 2.5GB of RIFE and v2 models from 2023 and 2024 and most of them many of us never used.
There are almost no updates recently.
Each models needs to be computers before use for particular resolution (take up to coople of minutes each), this adds extra GBs.
Most users use only a few of them based of HW specification and video watched (Anime vs movies vs sport) and their liking.
This is not on-size-fits-all.

From my perspective, it doesn't make much sense to implement RIFE models management in SVP, see the reasons above.

2,481 (edited by dawkinscm 26-12-2025 21:14:21)

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

kellykline wrote:

Why doesn't SVP update the AI Model? The latest I have on there is 4.18. I don't see 4.25.

I no longer user RIFE but 4.18 is a good model to use because it has many of the benefits of RIFE without needed a beefy GPU like a 4080. I wouldn't recommend 4.25 for anything less than maybe a 4070.

2,482 (edited by dawkinscm 27-12-2025 04:48:23)

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

Recently I noticed that setting my TV display refresh rate to 120Hz helps in reducing the workload when running at 120fps. Obvious I know. But what it means is that SVP only needs to perform a gentle smoothing operation which is well within it's capabilities and nvof is not needed either. For confirmation I tried running without SVP at 120hz and the jerkiness came back. Then I tried mvp interpolation on my PC and same thing happened. So SVP is definitely working and tweaked a little I see all the smoothness of RIFE but none of the obvious artifacts or occlusion errors.

Using the SVP "Automatic" profile and performance slider moved all the way across to "Higher Quality" I got this running at 70% on my integrated GPU. But some of the processing was still leaking into my 50 series so maybe it would be running at around 95% without it. The easiest option would be to set the refresh rate to the highest matching SVP rate, rerun SVP "Assess System Performance". I've tweaked it further to increase smoothness and remove all artifacts (so far). I don't think this is if this is TV model dependent because I'm using VR so any extra TV processing is probably lost.

P.S. On PC, when matching refresh rates, using Vulkan is the smoothest providing RIFE like fluidity with basic SVP settings. But depending on the config, some pixelation may occur when using Nvidia cards.