Figured I’d make a standalone post as I’d really like to see this stuff get fixed in the Linux version but I can easily see Linux getting tossed by the wayside with the v4.0 release and all of those issues. All these issues are on 3.5.3 The Linux version is barely usable, but for reasons that should be easily fixed…
Non-working models with erratic memory usage
I’ve posted this in the 3.5.3 Beta thread. Over half of the models simply don’t work because of initial spikes In VRAM usage. This may be survivable on a GPU with 24gb of VRAM. I am using a 4070 and these spikes blow past the 12gb VRAM on the card. All of these models specifically have this issue…
-Gaia CG
-Gaia HQ
-Artemis Alias or Moire
-Theia detail
-Theia fidelity
-Apollo
-Apollo Fast
-Chronos Fast
…And this is what happens, mind you this is only a 640x480 clip!
Compared to the few models that behave as they should…
Videos failing or being deleted at the very end.
This one I can’t quite figure out and it’s driving me nuts, I’ve started cutting videos into smaller clips before running them through Topaz since half a video gone is less waste than a full video that might take 3 or 4 hours. I’m pretty much sticking to Iris w/Chronos for everything since most of the others do not work. Running that combo gets an export all the way to the finish 90% of the time but 50% of the time the video has a red X error or just disappears! I’ve tried every combination of “Include Live Preview” and “use temp directory” possible but it does not help.
My thought is that something is going wrong when the temporary file gets copied to the finalized video. I see nothing useful in the logs other than “trying to delete video”. What really bugs me out about this is that this morning I watched 2 out of 6 videos finish with a green checkmark and then just disappear from the output folder! So now I’m wondering if there’s a bug where TVAI just deletes some clips before finishing the copy.
Running this on Windows upto 3.2.x I have always taking input videos over an SMB share on a 10gb network and output to the single OS NVME in the system, I’ve never seen this happen. I tried copying all the inputs to one SSD and outputting to the OS NVME, Input and output on the same NVME…nothing has helped.
The 2 odd bugs I noticed that may or may not be causing this…
All of those were unchecked when I saved, no matter what, “Use Temporary Directory” shows checked next time I open preferences but TVAI is not using the temp directory, it’s putting the partial video in the output folder as it should.
The 2nd bug is that nothing stops Live Preview from being generated. If it’s unchecked in settings and in the video output, there will still be the same preview.
Max Memory Usage does not (seem) to work
This could be related to the first issue. Reducing the memory usage in settings does not in any way reduce the memory TVAI would use. As long as the model is running right I see no reason to reduce the memory usage on a card with 12gb VRAM, but if there’s no control over the memory usage something must be wrong right?
Very limited GPU selection
I had 2 ARC A770s for about 2 weeks before I got rid of them. In fairness, they didn’t work in Windows either. Most models got a few seconds into an export and died. I don’t remember all the details but ARCs were supposed to work. At best 1 ARC almost worked, 2 together never worked at all. I also tried a 1080 and a 1080ti…didn’t work either, no idea why 10 series does not work in Linux, I ran TVAI on 2 1080ti’s for 2 years in WIndows without an issue. IIRC the 1080s didn’t even work in Windows versions new enough to have the Iris model so I have no idea what happened there. The 4070 is the only thing I’ve had remotely working in Linux, I didn’t want to spring for a more expensive card knowing it probably won’t work. I suppose it’s not worth mentioning that AMD probably does not work at all since that much was stated by Topaz. So from what I know at best Linux is limited to 20-40series Nvidia GPUs
Multiple GPUs?
I have no idea if multiple GPUs work in Linux. I’m hearing multiple GPUs is broken in Mac and I couldn’t get 2 Intel ARCs to work in Windows either. This is kind of shocking since I ran 2 1080ti’s in SLI with Topaz in Windows from V2.x until V3.2.x with no issues, and I ran those cards so hard it was like a 600w space heater. The cooling water cooked off so fast I had to fill it every other day. Nobody is really reporting on Linux and I’m not dropping another $600 on a GPU to find out the hard way it doesn’t work so can somebody tell me?
It would be nice if Topaz could chime in on some of this. This forum is really the only form of customer support offered for TVAI. Linux might be in beta but I still paid for it lol.
