When using the new Iris 2X FPS model on interlaced video, at a random duration through the preview, the application randomly stops processing and simply says “Error” with no further information available.
Once this “Error” happens, subsequent attempts to create new previews (e.g., to see if it’s a particular slider influencing the error) result in no preview display - the right pane stays stuck on “Generating preview” and eventually produces the message “Error” again. If I clear all previous previews, and switch the processor to the CPU, it succeeds, suggestive that this is possibly a GPU/GPU-driver issue.
The GPU is nothing special - it’s the Intel UHD Graphics 770 on-chip GPU. I haven’t had trouble with it before in 3.1.11, but of course I also wasn’t using Iris with the older version. I have not encountered the same problem using the Dione TV 2X model in 3.4.4, which makes me suspect it may be some interaction between Iris and the Intel GPU.
Logs: logsForSupport.tar.gz (1.3 MB)
System info: systemInfo-20230919.zip (139.0 KB)
I spoke too soon… I started exporting a video using the CPU as the processor, and after processing about 1 minute of video, it too produced the “Error” message (no further detail) and stopped. So possibly nothing to do with the GPU, other than maybe something triggers this particular behavior to happen sooner when using the GPU.
Confirmed this is still happening in 3.5.0.
Last night I tried using ffmpeg first to transcode this old mpeg2 stream to interlaced h.264 first, before attempting to run Iris 2 on it, but alas, it still fails. I’ve also taken the source file and had ffmpeg read the entire thing looking for errors, and not found any.
The relevant error message in the log from last night’s failure is the not-particularly-helpful “2023-09-20 22-37-08.916 Thread: 12972 Info EXITED: 7 -1073741819 1” after which literally tens of thousands of attempts (like almost 50K) are made to delete the temp file, all of which must have failed, because the file is still there this morning. What it left behind plays beautifully and looks great, for what it’s worth.
I’ve also attempted to run the job using entirely local drives vs a network drive, but the same thing happens. Additionally, I’ve tried disabling my antivirus and adding exclusions for the paths VAI uses and for the VAI process itself, and none of these have born fruit either.
If it helps I can upload the source video somewhere; it’s from a PBS program that is no longer aired which I purchased on DVD (but it’s had the life mpeg2 compressed out of it), so it isn’t likely to offend anyone to provide it as test material.
Thanks for this detailed report. When running Iris using onboard graphics I would definitely recommend lowering the “Max Memory Usage” setting in the app’s preferences to 50% or less.
Iris uses more memory than some of the previous models and will run into crashes on systems without dedicated GPUs.
I will try that; however, note that this also happens when setting the processor to the CPU.
Another issue which may be related: I’ve observed that when exporting, it can be a good long time before any data is actually written to disk exporting as H.264, and with H.265 sometimes data is never written to disk before processing fails. In my previous test to see if Dione would work, it got halfway through the source material and the temp file never got beyond 0 bytes. For h264, the first time the file grows in size at all it’s by about 96MB all at once, and only then will the live preview begin to show up.
I notice after dropping the maximum memory usage down to 40% (of 64GB) that it’s writing about 8MB at a time for h264 though, so these two problems may be inter-related?
Adjusting the memory all the way down to 40% doesn’t help unfortunately, not even with older models like Dione 2X Dehalo. Nothing much interesting in the log, the last few messages are:
2023-09-21 17-26-51.152 Thread: 24064 Info OUT: 7 frame= 9497 fps= 2.88702
2023-09-21 17-26-54.246 Thread: 24064 Info EXITED: 7 -1073741819 1