Very poor performances

On the Mac studio M2 Ultra this task is much faster, running at somewhat under 6 fps:

2 Likes

If you don’t expect wonders, I think especially Iris V1 can in fact enhance really bad quality videos. quite a bit:


2 Likes

Cliff :heart_eyes:

1 Like

Thanks, but it’s just personal family videos, I can wait until I change my computer.

That’s an amazing difference, very good !

That’s 30x better then on my old Mac… It will be time for me to upgrade my computer in 2024, but I would directly switch to a M3 pro or M3max processor, as it also embeds new features like hardware raytracing.

I’ve been upscaling to 4k with Iris on two different M1 Macs and the performance is terrible. Both are running 1.5 fps and primarily using the two Efficiency cores instead of the Performance cores. And the system settings are practically unusable. Are there better setting for Mac and are they planning to optimize for Apple silicon considering the number of people in the film and TV industry using Macs?

Please help!
Z

1 Like

awful perfomance. first i was told “ow its your old horrible computer” agreed i had a core i7 of the 3rd gen, 24 gbs of DDR3 ram and a radeon graphics card with 8 GBS of ram DDR3 and pci express 3.0
nowi have a gaming laptop with a 12th gen i7 core with 16 threads (old one was with 8 threads) 16 gbs of DDR6 ram (if i am not mistaken) and a nvidia RTX 4050 (mobile version) GPU with 6 gbs of RAM, games runs several times better and faster in my laptop. yet video AI keeps giving me with proteus from low res to 1080p processes at 6 frames/second (almost the same as in my old museum computer) so eighter my old museum computer was awesome or my new gaming laptop sucks (totally not everything is insanely fast done not just games running at 1080p over 60 fps) so its feels topaz just lied to me convincing me i needed a new pc because mine was so old probably bill gates learned to code on the same hardware (totally not that old but you get my point)

1 Like

Iris does have terrible performance on Apple Silicon since a few months now.

I have old model files from 08/23 for Iris V1 for upscaling SD/HD to FHD that are much faster. Unfortunately not the whole set and unfortunately I also don’t have a copy of the fast Iris V2 version from Sept. 2023.

1 Like

Video AI doesn’t seem to use any of my 8 high performance cores, and just maxes out the efficiency cores, making everything else on the system extremely slow.

A 1 hr HD clip (upscaled from SD) is going 1.5fps for 20 hr ETA.
This really limits the app for everybody in the film and TV industry who almost all use the latest Macs.

1 Like

Are you using Iris V1 or V2?
Do you know how to fiddle with Apps / show Package Contents and replace something in there?

1 Like

Here is my benchmark, very poor. However, it seems that some other people here have also poor performances on video AI with recent and high performances computer, and that video AI seems not managing correctly all the capacity of some computers.

Topaz Video AI  v4.0.4
System Information
OS: Mac v12.0701
CPU: Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz  32 GB
GPU: AMD Radeon R9 M395X  4 GB
Processing Settings
device: 0 vram: 0.75 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	00.94 fps 	2X: 	00.62 fps 	4X: 	00.23 fps 	
Iris		1X: 	00.70 fps 	2X: 	00.36 fps 	4X: 	00.12 fps 	
Proteus		1X: 	00.78 fps 	2X: 	00.51 fps 	4X: 	00.23 fps 	
Gaia		1X: 	00.26 fps 	2X: 	00.18 fps 	4X: 	00.14 fps 	
Nyx		1X: 	00.38 fps 	2X: 	00.32 fps 	
4X Slowmo		Apollo: 	00.79 fps 	APFast: 	03.53 fps 	Chronos: 	00.25 fps 	CHFast: 	00.47 fps 	

I’m set it to:
Upscale to 4k (from HD - 29.97fps)
Iris - Face / LQ - MQ
Medium Quality (input condition)
ProRes mov export

And yes, I’m a long time audio/video pro Mac user, but stop short of Terminal.
Thanks for any ideas!
Z

Unfortunately I don’t have a copy of the fast Iris MQ models files, just for the LQ ones (and even there not the fastest version but still much faster than what TVAI downloads now).

1 Like

awful my perfomance and its quite new hardware its not “my old computer for a museum” now

Topaz Video AI  v4.0.4
System Information
OS: Windows v11.23
CPU: 12th Gen Intel(R) Core(TM) i7-12650H  15.708 GB
GPU: NVIDIA GeForce RTX 4050 Laptop GPU  5.7969 GB
GPU: Intel(R) UHD Graphics  0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis		1X: 	05.75 fps 	2X: 	04.08 fps 	4X: 	01.38 fps 	
Iris		1X: 	06.47 fps 	2X: 	03.14 fps 	4X: 	01.08 fps 	
Proteus		1X: 	05.27 fps 	2X: 	03.70 fps 	4X: 	01.51 fps 	
Gaia		1X: 	02.16 fps 	2X: 	01.45 fps 	4X: 	00.99 fps 	
Nyx		1X: 	02.57 fps 	2X: 	02.27 fps 	
4X Slowmo		Apollo: 	09.46 fps 	APFast: 	29.26 fps 	Chronos: 	05.09 fps 	CHFast: 	07.98 fps 	

I was using MQ because the source footage is already HD, but would LQ be almost as good? I’m new to Topaz Video so I don’t know the difference between the two.

That is a very varying thing dependig on personal preferences if you look at comments here- so there’s no TRUE/FALSE answer here.

I myself like Iris LQ much more than V2 even for HQ material as it mostly doesn’t do strange things. Iris MQ while having more details often just creates details that are simply fake and has some other issues with “crawling” noise and unnatural repetitive patterns.

Your mileage may considerably vary so you have to test out for yourself. Oh, and better don’t use the “Auto” setting on Iris LQ as it IMO doesn’t work very well there (and is one of the reasons that some people say that Iris LQ smooths out the picture too much).

I use high values for fix compression on LQ input material and lower ones for higher quality vids, the improve detail settings between 40 and 60, only little sharpening (10-15) and very little noise reduction (as Iris already drastically reduces noise even if this set to zero).
And the most important setting: set Antialias/Deblur to a high positive value, e.g. 60 (except for when you get artifacts with very LQ input).
Also Recover orig. details should be set to a high value (80-85) to defeat the plastic/comic look.

But everyone has different preferences as it seems.

2 Likes