On RTX A4000 with 1080p/60fps input to 2160p/60fps output I was getting ~0.28spf with around 4GB VRAM use but 100% CPU use (AMD Ryzen5 5600X 6-Core @ 100W PBO). Not sure how well multiple streams would go given the high CPU use (in my case).
Hoping they can sort this 3.1.x line of releases with Proteus manual as that’s how most of my upscaling is done.
5.2fps for 1080p to 2160p with my 3090
I noticed that the RTX 3xxx version downloads and uses ONNX models, not the Tensor ones, is it intended?
At least for prob-v3-fnet-fp16-288x288-4x, Chronos and Chronos Fast 288x288 models, cause for 2x model, VEAI got prob-v3-fgnet-fp16-288x288-2x-rt806-8500 file. 
In the last few releases, I’ve discovered that the relative settings have become much more sensitive than they were. - Try keeping your relative offsets down to small numbers.
Also, don’t touch the decompression settings. It is now doing the job very well, pushing it up tends to create more problems than it solves.
Mac Studio - Doing some tests now for full output but even in the previews… WOW it’s so much quicker!!
Absolutely faster than 2.6 when processing video but I’m finding the UI to be a lot slower to navigate.
same stupid slow rendering. and now im in a stripped down windows 10 version that should speed up no matter what. not even apptempting output.
Thanks Johnnycyan, that worked!
Just to add 2 more card numbers for this update with the same file as I just tested on the 3060 Ti. My 3060(12Gb) is giving me identical numbers to the 3060 Ti. I’m fine with that. My RX 6600 is now completely left in the dust, only 11 fp/s.
Now the not so good news in some ways. running multiple instances on the 3060’s cuts the time almost in half. The way I’m looking at that is that I am still getting the same total throughput as before but as the individual files get processed faster that’s less chance of losing one or more files at once.
Thanks it works !
Found the issue, it is currently applying no parameters at all when in manual mode. The manual parameters should be fixed in next beta.
Test with Proteus Auto with the WinAll Version for both :
1050ti GTX 720p 29.97 → 1080p = 0.81 seconds per frame (1.33 spf before)
A4500 RTX 720p 29.97 → 1080p= 0.26 to 0.27 seconds per frame (don’t remember before)
Change the output to H265 or anything other than H264 if outputting higher than 1080p on windows. The encoders do not support 4K output, seems like the error message is broken. Will fix that.
Can you send me your system specs or a log file.
I don’t know what happened but that was unexpected with the grain settings hahaha will have to deactivate it for once… 
It is supposed to download and use rt806 models. Can you send me the logs.
I’m using hevc_nvenc however so that doesn’t seem to be the issue
i’m noticing that on the A4500 RTX, that the playback of the preview is a bit faster than the normal speed, while the rendered file played outside the software play at a normal speed !
which version must be installed on a 4500RTX ? the winAll version or the 30x0x RTX version ?
Will fix the too fast playback.
You should use the RTX 30 version, it should be much faster than the all version for the card.
sure have the logs.
logsForSupport.tar.zip (44.3 KB)
and dxsiag log as well
DxDiag.txt (79.4 KB)
