I’m currently shooting with a Canon EOS 90D (producing cRAW .CR3 files) and have been using Darktable (now at 3.8.1) successfully to handle my RAW images, including some that I’ve processed with Topaz’s DeNoise AI and Sharpen AI tools and exported to DNG. I’ve recently tried Topaz’s new Photo AI (now at 1.0.2) and find that my Darktable can’t import the DNGs it generates. As a result, I’ve switched my workflow to transfer Photo AI images as TIFF, which Darktable has no problem processing. Topaz support says that they have made changes to their DNG generation processing in the newer product, but I don’t know further specifics. I’ve uploaded a copy of a raw.pixls.us EOS 90D reference image as processed to DNG by Photo AI 1.0.2, in case this provides anything worth investigation from a Darktable perspective. The file is available at Nextcloud - it’s too large to attach to this post. Thanks for any interest.
Hi, welcome to the forum!
I think darktable’s DNG handling is pretty good, so long as the software generating the DNGs generates them to spec. Seems like the issue is on the Topaz side, and since they have paid support and developers, perhaps you should peruse that avenue.
Thanks. Fair point, and I’m not presuming a need for any reaction on behalf of Darktable. I’m OK using TIFF, which is working fine for me.
DNG is likely not giving you anything over TIFF in case, since Topaz isn’t outputting raw data, the DNG is likely just wrapped around TIFF data.
Perhaps the same thing as with this one Hasselblad L2D-20c DJI Mavic 3 main camera · Issue #10852 · darktable-org/darktable · GitHub
@kmilos may be able to give a better answer.
It is probably a Topaz AI problem. Other forums have also described problems with DNG files created by the new Topaz AI (for example a wrong size/quality of the preview embedded in DNG files).
I testet your file and could not open it in darktable either. I run your DNG file through the Adobe DNG Converter and the converted file was opened by darktable without problems.
The linear DNG files from DxO photolab also work without issues.
And there was a thread before where ON1 nonoise DNG files worked ok before , and since a ON1 update they didn’t work (well they would load but colors way off or something).
It’s just a rabbit hole .
But as long as the DNG files (linear or not ) use no compression or default Adobe compression, they open fine. And as long as the data inside is truly unaltered camera space data , it should process fine .
The only downside i have is that all the highlight recovery algorithms don’t work on linear DNG files (already demosaiced input )
I hadn’t thought to try that experiment, or even known that it was possible to pass a DNG as input to the Adobe converter. It looks like Topaz has adopted a different compression algorithm for use within its DNGs, and that Darktable doesn’t support that algorithm. Thanks for investigating.
It looks more like an error from Topaz.
Yes, a top secret one 
[SubIFD] Compression : Uncompressed
However, as @Peter pointed out, and rawspeed rightfully reported, the file is has an inconsistent tag and thus appears as corrupt:
[SubIFD] RowsPerStrip : 41904
which should, of course, be 4660 in this case.
Even the thumbnail has the incorrect
[IFD0] RowsPerStrip : 768
while the thumbnail height is 170…
Btw, using exiftool/exiv2 and reading the TIFF and DNG specs is not rocket science. I mean, if it’s only me who can do it, I’ll start charging, and then some. 
Yes, 41904 rows would encode quite the sensor, and uncompressed isn’t a compression algorithm except in a degenerate sense. I’m interested to note that my RawTherapee can open the image, though, so will infer that DT is more comprehensive in its input validation.
Precisely, though it’s down to the rawspeed library used by dt.