Thank you for your respond.
I checked the folder you referred to. But the files under “\ AppData \ Roaming \ Topaz Labs LLC \ Video Enhance AI \ models” all are has -fp16- value.
None of the has -fp32- value.
Please see the attached picture
And today I tried something new;
I changed Windows Power options to “High Performance”. (was Balanced)
Then unhide the Processor Power Management > “Processor Performance Boost Mode” option with the registry hack under Power Options > Advanced Settings.
Disabled “Processor Performance Boost Mode” (was aggressive on default)
Run the software. PC now crashed after 20-30 mins instead of 3mins.
I returned BIOS to default settings = crashed again in 30 mins.
I wrote to support service. They asked me to send the logs for investigating.
Are there any new model that comes close to CG HQ for low res/quality clips pre-cleaned with Neat Video? With older versions, CG HQ was miles ahead.
I find that blocking needs to be broken up by neat video, in order to be recongnised and removed by CG HQ. The program cannot deal with full blocking squares!
Ok! (-fp32-) is downloaded from the Internet on its own, if not prohibited in the settings. At the same time, the more the memory level regulator is set in the program settings, the higher the block values are for the modules for the used model. This can be seen in real time when coding, how files are uploaded to this folder. Files of models with blocks (480x384 … 576х448 …) (- fp32-) can be loaded.
Also, I installed the card in the motherboard with AMD 20.12.1 drivers and Windows 20H2 19042.508 with updates to 2020.12. program 2.1.1 kept crash. Then I removed them (- fp32-) and it works now. But program 2.1.1 also didn’t work with files (-fp16-) on a system with the latest updates and the latest driver. Today I will check it sequentially, after which it will crash again if I install the latest drivers and updates. Or maybe the direct x (d3dx9_43.dll) files that I additionally installed for some programs. Also, the Reduse Machine Load setting strongly influenced the operation of the system itself, and sometimes if you turn it off, the program immediately reported a lack of memory with a 1 GB AMD card, and if you turn on this function, the program worked. And also, with AMD R7 260X 1GB cards, the 2.1.1 program works with any files and updates.
Will we ever see Gaia CG restored to the quality found in 1.6.1? I assumed that since this is no longer listed in the issues that it was fixed in 1.9.0, but after switching back to 1.6.1, I can see that it is still the superior Gaia GC model.
no I don’t have any such thing. nor any registry hack xd. At Microsoft, they said that it releases the update gradually. So they have to test all machines progressively. On the pro version, you need stability. So it takes longer to receive the update.
If you have longish filenames they get cut off everywhere, maybe someplace near the top (or if you tooltip the source & target filenames) you can show the whole thing?
I don’t like the recent UI change where when adding several videos, the default is now that when processing begins only one video will be processed. It should default to “select all” with a “uncheck” or “select none” option.
The big flaw of Artemis is that if it works in 2x it gives a result (good enough), if it works in 4x, it changes result, It seems like a postprocessing 4x, and it loses fine details compared to 2x !! Absurd! (Look at texture of a cloth or skin) In 6x the same effect, even worse.
This is not normal, it does not happen with Gaia hq, it does not happen in photographs with gigapixels and other competing programs!
I don’t think Gaia CG had a version number at the time. By the time it reached VEAI 1.6.1, it was named Gaia CG 1.0.1, and is not compatible with recent VEAI versions.
I also got this, but not in the list, but when you iterate over different versions of the models for the test, and at some point the version remains the same and does not change. This can be seen at the bottom of the screen by the name of the output file, it does not write the model that was chosen last. This is the same error and the list of tasks turns out.
This is visible and very strong! 4X blurs very small details and makes this choice meaningless. It’s easier to choose 2X and interpolate to 4X, the result will be better than doing 4X right away.
Today I installed new driver 21-4-1 for AMD card, and it works better than the previous one. There is no general system failure and the mouse does not freeze when the graphics card is loaded. But the program still keeps crashing on my RX 560 2GB video card. When the models (-fp32-) are used, it crashes almost immediately, and if the models (-fp16-) then the system lasts for 30-40 minutes, then the program simply closes completely and the driver error about exceeding the response and waiting for the program. With the AMD 20-12-1 driver, the program works stably only with models (-fp16-), which on this driver work twice as slow, but stably. Perhaps the whole point is in delays and mismatch of expectations of timings and some kind of synchronization with interrupts of the system and the program, because the program crashes hard. Perhaps these are features of the structure of processors, and some unaccounted for differences in the program code, because there are no such failures with other programs. On older AMD video cards of a different model, this program works with any driver, which is very strange …
