Skip to main content
morphinapg
Inspiring
June 7, 2023

I may have encountered a bug, but I can't be sure what's causing it. Need help diagnosing.

  • June 7, 2023
  • 5 replies
  • 344 views

I have had several crashes trying to encode a 4K video I edited in Premiere and am exporting with Media Encoder that is 11.5 hours and takes 45 hours to encode with the export settings I'm using. I have had 8 crashes so far, and most of them anywhere from 50 to 75% of the way through the encode, so you can understand my frustration here. 

 

First off, my specs:

CPU: 5950x

GPU: 3080Ti

RAM: 64GB DDR4

Source Footage is on two 8TB SATA SSDs striped together, roughly 1 GB/s read speeds as a result

Saving to a 12 TB SATA HDD

 

I am encoding using Voukoder, an export plugin that allows me to export using x265, which I prefer over the built in encoders, for certain specific reasons that I need for my exports. I know Adobe's first response may be to try to blame Voukoder, but I've had that plugin crash in the past. It produces its own log, and its crashes also don't look like what has been happening here, so I don't suspect it's Voukoder at fault here. 

 

I actually successfully encoded an HDR version of this video. I am currently working on a version of the video exported with an SDR grade, and this is specifically the one that has been causing problems, although it's possible I just got lucky with the first encode. Both sequences had a LUT applied through Lumetri, for grading both HDR and SDR, and use the same footage and other effects, so there shouldn't be any technical differences in what Premiere is doing behind the scenes with the sequences.

 

When I did the first few encodes, Premiere was version 23.2 (because of an issue I had with 23.3), while Media Encoder was the most recent 23.4. After the first 3 crashes, I wondered if maybe the difference in versions was causing the issue, so first I tried with both being 23.2, and then tried with both being updated to 23.4. Crashed both times. So it's clear the issue is not because of a mismatch between versions. If there is a bug in either app causing this, it was not fixed in any of the combinations I tried.

 

2 of the 8 crashes were Media Encoder itself, where it popped up with the dialog box saying it had crashed, and to send the error report. I did so, and added my email for followup, but I haven't received any responses about that. If an Adobe employee sees this, it would be helpful if you could look into those reports to help diagnose this.

 

6 of the 8 crashes however were BSOD. Half of those BSOD claimed "Memory Management" issues, and the other half claimed it was an issue with my video card driver. I thought perhaps there was an issue with an old GPU driver, which perhaps was interacting with my RAM in a corrupted way, so I updated to the latest Nvidia driver - 535.98. Unfortunately, no luck, crashed again.

My next guess was that perhaps there was something wrong with my RAM. I had upgraded my CPU to the 5950x in late April, and I use a large heatsink/fan on top of the CPU, which just barely touches the top of one of my RAM sticks. I was worried I had perhaps knocked one of my RAM sticks when reinstalling the HSF. I had considered this earlier, but wanted to try all other options before testing my RAM, because doing so requires taking off the HSF, and reapplying thermal paste, which I don't like doing.

 

But a couple days ago, I decided to go in and do just that. I took all of my RAM out, reattached my HSF, and then placed one stick of RAM in at a time, in slot 2 which is accessible even with the HSF in, and is the recommended slot for 1 stick of RAM. Windows Memory Diagnostic did not seem to be working, stalling at _% forever, so I downloaded memtest64, and ran it on every stick individually. All 4 sticks of RAM gave me no errors in this app. I thought maybe I did in fact just knock one of the sticks, and simply reseating them solved the issue, or perhaps one of the other RAM slots is at fault, rather than the stick itself. I placed in a second stick of RAM in the 4th slot and ran the same tests with both sticks in. Again, no errors. I ran the encode again, and about 36 hours later, once again, another crash.

 

I have now come to what I think might be my final attempt to rule out the possibility of either Adobe or Nvidia software being the thing causing this crash. I thought - what if it's my 12TB HDD failing? Perhaps the encode is hitting a certain sector at some point during the encode that causes some kind of corruption that freaks out one or more of either Media Encoder, Memory Management, or the GPU driver. So I just started my 9th attempt at this encode, telling it to save the exported file to the same SSD that the source footage is on. It will be up to 2 more days before I know this encode is successful or not, but even if it is successful, I can't 100% rule out that this issue was caused by either Adobe software, or Nvidia's driver.

 

If this is a software issue, it may be possible that it would be impossible to recreate without a similarly long 4K, CPU-based encode. I have no issues encoding smaller files.

 

If anybody has any more suggestions about what I can do to help diagnose this, it would be very helpful. As I said, my current encode will take up to 2 more days from when I posted this, if it is successful.

 

Here's a link to all of my minidumps from the recent BSOD crashes, in case they help:

https://drive.google.com/drive/folders/1NJ2EPV5VWuHWaYsw6ZtIxvja9OFCtRpd?usp=sharing

5 replies

morphinapg
Inspiring
March 25, 2024

Strangely enough, I think I solved what was probably causing this: For long encodes, I had the tendency to go into task manager and set Media Encoder process priority to low in order for my computer to be fully functional in every other task while the encode was happening. I believe the crashes were only happening specifically when the process priority was set to low. I figured it out because there was another GPU-heavy app I was using where I did the same thing, and encountered the same issue. I believe this is most likely an issue with the GPU driver expecting more real time access between the app and the GPU, and low process priority is causing it to perhaps go out of sync and cause unexpected behavior with the GPU or something. Either way, since then I've stopped using low process priority during encodes, they have all been successful. The same is true for my other GPU-heavy apps.

morphinapg
Inspiring
June 17, 2023

I can confirm now that using a different drive didn't solve my problem, I must have just gotten lucky on that encode. Nor did switching to the earilier 531 Nvidia driver, like I suggested in my last post (although that solved other performance issues I was having) On another encode, I had the same issue. Since these were smaller encodes, I tried it in Premiere as well, and just had the same issue there too. Unfortunately, these were BSOD crashes, so no crash reports sent to adobe, although I could grab the windows minidumps from those if they'd be helpful, although I expect they're probably similar to the ones I already posted.

Adobe Employee
June 14, 2023

Please send me a private message in this forum with your e-mail address. 

morphinapg
Inspiring
June 10, 2023

I didn't run System Compatibility Report, but I just checked and it says no issues. 

 

I wasn't able to test exporting directly from in Premiere. Like I said, this was an encode that would take nearly 2 days straight, and I was still working on transcribing subtitles while the project encoded.

 

As I mentioned above, I did add my email address to the crash comments.

 

I would like to update that my final test did finish, where I was exporting to a different drive. One other thing I did was disconnect an external SSD. I did this because while working on those subtitles, I kept randomly hearing the windows "device disconnected" sound a few times every night, even though I wasn't getting any notifications, so I thought maybe something had a weak connection. No idea if this could cause issues, but I thought I'd mention it, just in case.

 

However, after the encode finished, I noticed GPU performance issues in another program (Dreambooth, training Stable Diffusion models) that wasn't present in a previous driver. It turns out the last two Nvidia releases (532 and 535, the same drivers I had used for all of these encodes) have modified the way VRAM memory management works, and will start swapping to system RAM long before the VRAM gets fully utilized. Considering the two BSOD errors I was receiving were about memory management, and about the GPU driver, I think it might be worth investigating Nvidia's latest drivers and how they handle VRAM swapping to system ram, in case that can cause issues with Premiere or Media Encoder. It's possible that my encode finally finishing may have just been me getting lucky for once. 

Adobe Employee
June 10, 2023

Thanks for your report. Let's try to find out more:
* Did you try to run Sytem Compatibility Report to see if you were/are using the right drivers?
* Does this also happen if you try to export directly from within Premiere Pro?

* Please also add your e-mail address to the crash comments so that we can associate them with your issue.