Video Enhance AI v1.2.2 - Stability improvement & old devices support

Is there a reason why the program doesn’t turbo your GPU while rendering? My Titan RTX is running at 1350Mhz while trialing the product yet is capable of going to a 2000Mhz boost and it’s also not using all 24 GB of RAM. Is there a reason for this? I’m using HW manager to watch the process take place.

1 Like

It´s a driver problem. Uninstall them with display driver uninstaller (ddu) and install them again.

It´s common on rtx cards, bad generation.

Is there any way to download 1.0.2?

Thanks in advance.

It’s a fresh install. Turbos just fine in gaming.

The NVIDIA driver by default sets the cards to P2 power state for CUDA operations. You have to manually tell the driver to use P0 for CUDA with nvidiaProfileInspector. That may or may not help with your turbo situation, but it’s the first place to start diagnosing.

Hi,

Tried that to no avail. Strange. Are you getting a turbo on your Nvidia GPU? It’s also using a fraction of the GPU and nowhere near the 24GB of available RAM

Hi. Love the product. A couple things that (if possible) might be helpful to users.

  1. Is it possible to integrate an update system? Or at least just tell us when an update has rolled out within the software?
  2. Would it be possible to establish a published benchmark system? I don’t particularly have a top-tier GPU and I’d like to see which GPUs perform best with your software. More importantly, I’d like to determine the point of diminishing returns. Sure, perhaps everyone could go out and buy a RTX 2080 Ti… but they’re really expensive and it’d be nice for individuals to determine personalized cost/benefit ratios.
  3. This might be really difficult, but there may be external influences that require us to close the program… perhaps to apply a critical security update to our operating system. It’d be nice to integrate a resume function in the software. Sometimes my renders take an entire week, and if my computer crashes or the software crashes… I have to start over. I die a little bit if my 7-day render crashes at day 6.
  4. In the seek function (to assign where to begin and end a render from the source) sometimes clicking the timeline makes leaps that are too large. Alternately, the frame-by-frame method can be tedious and slow if you’re too far away from your target. It’d be nice to (roughly) get to where we need by clicking the timeline, but then utilizing a play/pause feature (at 1x speed) to get closer, then finally use the frame-by-frame feature to get exactly where we need to begin and end our clip.

Thank you and keep up the great work.

4 Likes

I get boost with the 1080ti. Peak during gaming is usually 1925-2000mhz and running Topaz I’m varying anywhere from 1700 to 1950mhz with it mostly in the 1800’s. Be sure to click apply changes a few times after changing the setting as nPI can be flakey sometimes. Also, reboot after getting the setting applied. It must be a registry level change that only takes effect when the driver initializes because it only works after a reboot. You also need to reapply the setting after every driver upgrade and reboot again.

1 Like

Unfortunately no. Not from the official side, i think.

Please add lossless video output option!! We really need this!!

5 Likes

Export to PNG or TIFF image, then you can use ffmpeg join the images back to video with whatever video output option.

Here is my test of the new Version v1.2.2.

Source:
MPEG4 Video (H264) 1920x1080, 23.976fps, 7988kbps
Genre → Anime

Processing:
Gaia-CG: P, CG, HQ

Output Video:
4K UHD (3840x2160)
Crop to fill frame
8 bit TIF

Test:
Output in 8 Bit TIF, 16 Bit TIFF and mpeg4. Each with AI Ressources Demand 20%, 100%, 200% and 500%.

Result:
I’m still facing Blue/Black stains through the whole Video on some frames. TIF or TIFF doesn’t make any difference for me except the file size. The output in mpeg4 gives me some shaky lines, which I can handle with the lossless output. Thanks for that. Also the only big difference with the AI Ressources Demand Setting is with 20%. All higher than 20% don’t have big differences to each other which you will see in the following pics.
But even with 20% I can’t remove the Blue/Black stains. There are some Frames with much worse stains and others you can hardly see.

Here is the Before and after Pic with 20% and 100% AI Resources Demand:

Here is the Before and after Pic with 100% and 200% AI Resources Demand:

So I kindly ask you to fix this issue.

Thanks alot and stay healthy!

1 Like

What is the source of your video ? did you tried with an official source like a bluray one ?

Yep.
Doesn’t matter which source. I’m facing the same issue.

By the way… Faced this issue also on a self recorded video with the Smartphone. And as I read in the other Thread, other people too.

1 Like

Anyone else having the issue where it cuts off the last 20-50 frames of the video and makes them black instead?

I have yet to find a solution for it since it only happens on some videos. It works fine on most of them. It’s really annoying, cus it ruins videos that are meant to be looped. The issue has been there since 1.0.2.

It looks like earlier versions performed much better denoising and sharpening.
why is not possible to include older filters in newer versions?

4 Likes

Würde ein Grafikkartentausch von einer NVIDIA 1070 zur 2080 Super oder 2080Ti geschwindigkeitsmäßig was bringen?

Warte lieber auf die nächste Generation, von 1070 auf 2080 super oder so lohnt sich meiner Meinung nach nicht.

Okay I’m having this issue:

While converting 480p to 4K im getting the exact same frames/sec on the following gpus

1080ti
2070
Titan RTX

So is there a bottleneck somewhere in the software? There’s no reason an RTX card should be getting the same performance as last gen card if this uses tensor as the base code. Also is there a loss in fidelity using the 1080ti vs Titan RTX?

As far as I know VEAI uses CUDA for calculations and doesn’t use any RTX specific features. The 1080ti is a very good CUDA performer. Quite a bit better than the 2070 and only a bit worse than the Titan.

But I suspect GPU performance is only part of the equation for VEAI.