Or better software.
As is seen by the now abysmal performance of Iris on Apple Silicon.
My M1 Pro with 10CPU cores and 16 GPU cores is about as fast with the (not really so) old V3.4.0.0.1.a as the M2Ultra with those newer versions despite the latter having 24 CPU cores and 60 GPU cores.
That’s right: 60 vs 16 GPU cores and still about the same performance due to lacking software optimisation.
unuseable. preview doesnt work anymore. source and processed is async and processed doesnt show any difference. its broken
Here are the benchmark results. Is it because my device configuration is low? If so, can you tell me what the minimum configuration requirements are?
Topaz Video AI v3.5.3
System Information
OS: Windows v10.22
CPU: 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz 15.709 GB
GPU: NVIDIA GeForce RTX 3050 Laptop GPU 3.8838 GB
GPU: Intel(R) UHD Graphics 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 1920x1080
Benchmark Results
Artemis 1X: 04.87 fps 2X: 03.41 fps 4X: 00.99 fps
Iris 1X: 04.05 fps 2X: 02.33 fps 4X: 00.67 fps
Proteus 1X: 03.45 fps 2X: 02.58 fps 4X: 00.84 fps
Gaia 1X: 01.08 fps 2X: 00.75 fps 4X: 00.50 fps
Nyx 1X: 01.36 fps
4X Slowmo Apollo: 05.27 fps APFast: 19.94 fps Chronos: 03.03 fps CHFast: 05.11 fps
Topaz Video AI v3.5.3
System Information
OS: Windows v10.22
CPU: 11th Gen Intel(R) Core(TM) i7-11800H @ 2.30GHz 15.709 GB
GPU: NVIDIA GeForce RTX 3050 Laptop GPU 3.8838 GB
GPU: Intel(R) UHD Graphics 0.125 GB
Processing Settings
device: 0 vram: 1 instances: 1
Input Resolution: 3840x2160
Benchmark Results
Artemis 1X: 00.84 fps 2X: 00.53 fps 4X: ERR fps
Iris 1X: 00.87 fps 2X: 00.41 fps 4X: 00.10 fps
Proteus 1X: 00.65 fps 2X: 00.41 fps 4X: ERR fps
Gaia 1X: 00.21 fps 2X: 00.12 fps 4X: 00.09 fps
Nyx 1X: 00.17 fps
You may compare your benchmark result with others.
This one is a bust Topaz, My it team is frantically uninstalling and reinstalling an older version (have not been successful yet) so we can work. We use this almost every day at our studio and we are dead in the water. I am running a machine that is top tier with a Ryzen 5950, Rtx 3090ti and 128gb of pc3600 running at 3600, and a gen4 NVME. The machine is solid and worked with every version until now. It is plain awful and is basically useless. Now If I hear someone say that this machine is not powerful enough, I call BULL! The machine is a little over a year and a half old and if your software outruns its headlights… you don’t understand your customers. I have been a paying customer for several years and have gone to bat for ou guys when people complain for no reason or they have a potato computer. But this is just sloppy.
Perhaps this software is only suitable for use by studios specializing in graphics. Not suitable for mass use.
Is there a reason that downloading updates for Video AI is 30-50x slower than downloading updates for Photo AI (especially given similar file sizes)? I know Photo AI sometimes does smaller patches (which Video AI should do also), but even in the case of a large download, Video AI is PAINFULLY slow to download.
On a similar topic, Photo AI has an option to download updates in the background and have them ready for install. Why doesn’t Video AI have this same function months later?
They used to… not any more.
You might want to request being a beta tester, and we can really use your perspective as you’re a professional user of the program, plus you get the benefit of testing the betas, which would be of immeasurable value as you will have stable builds to fall back on when encountering general release mishaps such as this one!
like the previous version its broken. preview doesnt work correct. source and preview picture is always async. its impossible to finetune parameter if the video is always showing the wrong scenes
I’d really like to hear the answer for why Video AI update downloads are ridiculously slow. Regardless if they are using a CDN like Akamai or their own servers, I would expect download speed parity across programs.
Wish I could tell you, ironically, I don’t update through the program I download through my browser, as the update process freezes on the “deleting shortcuts” uninstallation dialogue on my PC, so I’m forced to use window’s uninstall troubleshooter in order to uninstall the prog! 
I’d check that machine first with HWINFO64 and make sure that you are not getting any WHEA errors.
If you are you’d need to play with and check your bios settings.
I had your kind of problem a few years back and it took 2 days of head scratching before I worked out it was one simple bios setting causing me problems, this was with a 3950X and a ASUS Board.
On a side note I’m now on a 5950X, RTX 4070, 64GB PC3600 with a ProArt X570-creator motherboard and that ProArt was well worth the spend over the Gigabyte I had.
I’m able to run a the 4070 at 2790mhz (Card spec is 2475Mhz)
The 5950X has a all core boost at 4500MHz and averages around 67c for most video work except when encoding cli svt-av1 when it will hit 85c max.
Feel free to DM me if you need any advice.
Thank you for your reply.
About your comment relating to load any video, I very much would like to see load any video to the export queue added to the resurrected live preview. Since now if you close out TVA you have to re-render to do a comparison. However for me I think I may roll back to the previous version until the live preview returns as I rely heavily on that feature.
I don’t understand this statement. ffmpeg works perfectly fine when being controlled by other applications. ffmpeg is under the hood of many video editors.
The next processing attempt ended up with a red cross error. This makes version 4.0.0 unuseable for me. 
Hi… I have a question. When Themis is selected with another model (for example Nyx), what is applied first, nyx (or the other models) or themis?
Im getting this too now, just red-cross stops after a few hours of processing AND removes the directory with the exported images, everything gone! please ! this is a high prio fix.
Which leads to this issue also: If I deinterlace into the double frame rate, the export frame number needs to be doubled internally when rendering, this is a bug. which prevents me from running this in stages.
The RTX 30350 is not really a powerful GPU. And then, if you render to an SSD instead of a HD, saving straight to the storage wouldn’t have any significant impact.
