@Zanthexter, I will admit that I did not do extensive testing before disagreeing that "BI5 broke Nvidia hardware acceleration." I had only messed around with
one camera a few days earlier for the purpose of collecting data on which methods of hardware acceleration worked and which did not. During that test, I saw GPU utilization at expected levels.
So just now I did more testing to see if I also have reduced Nvidia decoding capacity, and it looks like maybe I do, but only just barely.
I still run the same BI server hardware as I did during
my initial tests of the Nvidia hardware acceleration feature in BI4. Back then I was able to run a particular set of 11 cameras with Nvidia decoding, and that was the limit of my GT 1030 card. So just now I tried enabling Nvidia decoding for the same 11 cameras in BI 5.2.3.2, Nvidia driver 432.00. Two of those cameras have since been upgraded from 4MP to 8MP, so for this test I substituted them for two different 4MP cams. The result is that I was able to get only 10 cameras running on Nvidia, down from 11. I can substitute the 11th cam (4MP) for a 2MP cam, in which case I can get an 11th cam going. Therefore, I am seeing possibly a small amount of reduced capacity. Or it could just be the result of changes in the camera frame rates that I have made since 2018, which my test just now did not account for. Certainly I am not seeing anything resembling a loss of 50+% of capacity as you guys have.
During these tests, I saw a change to how BI handles overloading the Nvidia decoder now. Instead of showing a "No Signal" screen for the cam, it logged "HW VA not compatible: -542398533" and automatically set HWVA to "no" for that camera. On occasion, I saw it also log "FORCE quitting pXServer thread" for the same camera. Whenever it did that, the camera would get stuck in the colored bars state (loading) until I restarted the
Blue Iris service.