Subject: RAW decoding fails intermittently in batch mode — appears to be GPU VRAM related
David Zuma
Hello,
I'm running PTGui Pro 13.9 on Ubuntu 22.04 (headless server, batch processing launched via command line with -batch using a virtual display Xvfb on :99). GPU is an NVIDIA RTX 5060 (8 GB VRAM), with 128 GB system RAM.
Workflow: I process batches of ~25 panoramas at once, each made of 30 HDR source images (Olympus E-M1 Mark II ORF files, 10 positions × 3 exposures).
Problem: When processing a large batch, after a certain number of panoramas (variable, around the 4th–10th), one panorama fails with this error:
"The RAW files of this camera (E-M1MarkII) are currently not supported by the RAW converter in PTGui. To use these images, convert them to 16-bit TIFF files..."
The key points that make this confusing:
- The ORF files are perfectly valid — ExifTool reads them without any warning, and the same file opens and stitches without any problem when I load it manually in the PTGui GUI.
- It fails on a different image each time — it's not a specific corrupt file.
- It only happens in batch mode with many projects, never when processing a single panorama manually.
- When monitoring with nvidia-smi, I can see PTGui reaching peaks of almost 6 GB of VRAM near the end of each panorama. On an 8 GB card, if anything else is using VRAM, the peak has no room and the RAW decode seems to fail.
I found a related thread ("Crashing in batch mode") where you recommended disabling GPU acceleration. I tried limiting MaxVideoRamUsage (tested 5389 and 4000 MB) but failures still occur on random images — lowering the limit just changes which panorama fails.
Questions:
- Is this a known limitation of the GPU RAW decoder when VRAM is tight during batch processing?
- When the RAW decoder fails to allocate GPU memory, is it expected that it reports the error as "camera not supported" rather than an out-of-memory error? The misleading message made this very hard to diagnose.
- Is fully disabling OpenCL (EnableOpenCL=0) the recommended approach for unattended batch processing on an 8 GB card, or would more VRAM (16–24 GB) reliably solve it?
Thank you for any guidance.
PTGui Support
This error is indeed from the raw converter in PTGui, but the raw
conveersion doesn't use the GPU at all.
It's weird indeed if the problem occurs randomly. Could you make one
project available for download for which the error is triggered? Let me
try and see if I can reproduce the error.
Kind regards,
Joost Nieuwenhuijse
www.ptgui.com
On 6/18/26 21:03, David Zuma wrote:
> Hello,
>
> I'm running PTGui Pro 13.9 on Ubuntu 22.04 (headless server, batch
> processing launched via command line with -batch using a virtual display
> Xvfb on :99). GPU is an NVIDIA RTX 5060 (8 GB VRAM), with 128 GB system RAM.
>
> exposures).
>
> error:
>
> "The RAW files of this camera (E-M1MarkII) are currently not
> supported by the RAW converter in PTGui. To use these images,
> convert them to 16-bit TIFF files..."
>
> The key points that make this confusing:
>
> any warning, and the *same file opens and stitches without any
> problem when I load it manually* in the PTGui GUI.
> * It fails on a *different image each time* — it's not a specific
> corrupt file.
> * It only happens in *batch mode with many projects*, never when
> if anything else is using VRAM, the peak has no room and the RAW
> decode seems to fail.
>
> I found a related thread ("Crashing in batch mode") where you
> recommended disabling GPU acceleration. I tried limiting
> MaxVideoRamUsage (tested 5389 and 4000 MB) but failures still occur on
> random images — lowering the limit just changes which panorama fails.
>
>
> 1. Is this a known limitation of the GPU RAW decoder when VRAM is tight
> during batch processing?
> 2. When the RAW decoder fails to allocate GPU memory, is it expected
> out-of-memory error? The misleading message made this very hard to
> diagnose.
> (16–24 GB) reliably solve it?
>
> Thank you for any guidance.
>
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/
> ptgui/0be4a111-14d5-4fbd-abe5-dca6fc42783fn%40googlegroups.com <https://
> groups.google.com/d/msgid/ptgui/0be4a111-14d5-4fbd-abe5-
> dca6fc42783fn%40googlegroups.com?utm_medium=email&utm_source=footer>.
David Zuma
Subject: Re: RAW decoding fails intermittently in batch mode
Hi Joost,
Thank you for the quick reply.
That's very interesting — I assumed it was GPU-related because I found a clear correlation: with OpenCL disabled (EnableOpenCL=0), the error never occurs. With GPU acceleration enabled, the error appears intermittently in batch mode (on a different image each time, after processing a few panoramas). The exact same project, with the same RAW files, processes without any error once OpenCL is off.
I know you say the RAW converter doesn't use the GPU, so I can't explain why disabling it fixes the problem — but the correlation is very consistent in my setup. Perhaps it's an indirect effect (timing/concurrency), since without the GPU everything runs slower and more sequentially.
Some context on my setup, in case it's relevant:
- Ubuntu 22.04, headless, running via command line with -batch and a virtual display (Xvfb on :99)
- NVIDIA RTX 5060 (8 GB), driver 595.80
- The RAW files (Olympus E-M1 Mark II ORF) are on an external SSD
- Batches of ~25 projects launched at once, each with 30 source images
- The files are valid: ExifTool reads them fine, and the same file that fails in batch stitches perfectly when opened manually in the GUI
I'll prepare one of the projects that triggers the error and send you a download link so you can try to reproduce it. I'll include the ORF files and the .pts.
Kind regards,
David
David Zuma
I'd be happy to send you one of the projects that triggers the error so you can try to reproduce it (I'll include the ORF files and the .pts). Since I'd rather not make the project public, could you let me know which email address I can send the download link to privately?
Kind regards,
David
PTGui Support
Could you share your files with sup...@ptgui.com?
> I'd be happy to send you one of the projects that triggers the error so
> you can try to reproduce it (I'll include the ORF files and the .pts).
> which email address I can send the download link to privately?*
> Kind regards,
> David
>
>
> El viernes, 19 de junio de 2026 a las 17:36:33 UTC+2, PTGui Support
> escribió:
>
> Hi David,
>
> This error is indeed from the raw converter in PTGui, but the raw
> conveersion doesn't use the GPU at all.
>
> It's weird indeed if the problem occurs randomly. Could you make one
> project available for download for which the error is triggered? Let me
> try and see if I can reproduce the error.
>
> Kind regards,
>
> Joost Nieuwenhuijse
> > ptgui/0be4a111-14d5-4fbd-abe5-dca6fc42783fn%40googlegroups.com
> <http://40googlegroups.com> <https://
> > groups.google.com/d/msgid/ptgui/0be4a111-14d5-4fbd-abe5- <http://
> groups.google.com/d/msgid/ptgui/0be4a111-14d5-4fbd-abe5->
> > dca6fc42783fn%40googlegroups.com?
> utm_medium=email&utm_source=footer <http://40googlegroups.com?
> utm_medium=email&utm_source=footer>>.
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/
> groups.google.com/d/msgid/ptgui/91b5a743-8974-4cec-9638-
> cc6d4c2b536cn%40googlegroups.com?utm_medium=email&utm_source=footer>.
PTGui Support
That's really weird. The entire raw decoding runs on the cpu. It could
be heat or power draw but then I would expect to see other issues too.
Is the computer running stable otherwise? Do you get corrupted output
sometimes, or hangs or crashes?
On 6/22/26 09:18, David Zuma wrote:
>
> *Subject: Re: RAW decoding fails intermittently in batch mode*
> Hi Joost,
>
> Thank you for the quick reply.
>
> That's very interesting — I assumed it was GPU-related because I found a
> never occurs.* With GPU acceleration enabled, the error appears
> processing a few panoramas). The exact same project, with the same RAW
> files, processes without any error once OpenCL is off.
>
> I know you say the RAW converter doesn't use the GPU, so I can't
> concurrency), since without the GPU everything runs slower and more
> sequentially.
>
> Some context on my setup, in case it's relevant:
>
> * The RAW files (Olympus E-M1 Mark II ORF) are on an external SSD
> * Batches of ~25 projects launched at once, each with 30 source images
> * The files are valid: ExifTool reads them fine, and the same file
>
> I'll prepare one of the projects that triggers the error and send you a
> download link so you can try to reproduce it. I'll include the ORF files
> and the .pts.
>
> Kind regards,
> David
>
> El viernes, 19 de junio de 2026 a las 17:36:33 UTC+2, PTGui Support
> escribió:
>
> Hi David,
>
> This error is indeed from the raw converter in PTGui, but the raw
> conveersion doesn't use the GPU at all.
>
> It's weird indeed if the problem occurs randomly. Could you make one
> project available for download for which the error is triggered? Let me
> try and see if I can reproduce the error.
>
> Kind regards,
>
> Joost Nieuwenhuijse
> > ptgui/0be4a111-14d5-4fbd-abe5-dca6fc42783fn%40googlegroups.com
> <http://40googlegroups.com> <https://
> > groups.google.com/d/msgid/ptgui/0be4a111-14d5-4fbd-abe5- <http://
> groups.google.com/d/msgid/ptgui/0be4a111-14d5-4fbd-abe5->
> > dca6fc42783fn%40googlegroups.com?
> utm_medium=email&utm_source=footer <http://40googlegroups.com?
> utm_medium=email&utm_source=footer>>.
>
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/
> groups.google.com/d/msgid/ptgui/1214a303-1593-49d6-a99c-
> e5635b755c3fn%40googlegroups.com?utm_medium=email&utm_source=footer>.
David Zuma
Hi Joost,
I've been doing more systematic testing on the intermittent RAW decoding error and I now have much clearer data that I think will help reproduce it. Summary of what I found today:
Setup
- PTGui Pro 13.9 on Ubuntu 22.04, headless (Xvfb), driven in batch mode (
PTGui -batch *.pts) - Olympus E-M1 Mark II, ORF files, 3 camera bodies
- Large batch: ~4,600 ORF files, ~130 panoramas in a single batch run
- 128 GB RAM, NVIDIA RTX 3060 12GB (peak VRAM usage observed ~8.5 GB)
The error
On random panoramas the batch reports:
"Error loading file ... .ORF — The RAW files of this camera (E-M1MarkII) are currently not supported by the RAW converter in PTGui. To use these images, convert them to 16-bit TIFF in a RAW converter..."
Key findings
- It is intermittent. The same files that fail in one run process fine in another. It is not specific files — it's random which ones fail.
- It is not a filename/duplicate issue. Earlier I suspected duplicate filenames (multiple camera bodies producing the same filenames, e.g.
HDS20399, leading to(2)duplicates). I renamed every file to a unique scheme (SERIAL--originalname.ORF) so there are no duplicates and no collisions. The error still occurs on perfectly clean, uniquely-named files (e.g.BHUA32584--HDS30394.ORF,BHUA32584--HDS30492.ORF). - It happens with both GPU and CPU. This is the important new finding. I disabled OpenCL completely (
EnableOpenCL=0andEnableOpenCLForPreviews=0) and ran the full ~4-hour batch on CPU only. The error still appeared on random files. So it is independent of GPU/OpenCL — consistent with your note that the RAW converter doesn't use the GPU. - It is not a memory issue. It fails with plenty of free RAM and VRAM (VRAM peak ~8.5 GB of 12 GB available).
- The failure rate is low but constant — roughly 2–4 panoramas out of ~130 in a large batch, seemingly at random.
My read: this looks like an intermittent race/contention condition in the RAW decoder when many files are processed in sequence in a long batch, rather than anything specific to the files, the names, the GPU, or memory.
I have a fully reproducible case: a batch of clean, uniquely-named ORF files where a few fail randomly each run. I'd rather not post the actual project files publicly — could you share a private address where I can send you a reproducible project (the .pts files plus the ORFs) so you can investigate? The clean filenames make it a clean test case.
As a workaround I'm now pre-converting the ORF files to TIFF before stitching, which avoids the RAW decoder entirely, but I'd love to help you track down the underlying bug.
Thanks for your help.
You received this message because you are subscribed to a topic in the Google Groups "PTGui Support" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ptgui/mT022Gytr6I/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ptgui+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ptgui/040239b6-7a2a-42ee-a48f-9e9b5aafaa3f%40ptgui.com.
David Zuma
I wanted to share an observation in case it's useful to you, not as a firm conclusion on our part: we had been experiencing intermittent RAW decoding errors in long batches (messages like "Unexpected end of file" / "Input/output error" during decoding). Investigating on our end, we found those errors were concentrated on photos from one of our cameras, an OM System OM-5 (released in 2022), while photos from our other camera (Olympus E-M1 Mark II) never had any issue.
In our case, the root cause turned out to be an outdated version of the decoding library we used in pre-processing (it didn't yet support the OM-5). After updating it, those files decode correctly.
Since the symptoms we saw in PTGui when stitching those ORF files directly were similar, we were left wondering whether PTGui's internal RAW decoder might be affected in a similar way by the OM-5's specific format, especially in long batches.