Ok, I have reproduced the problem. I must say that the error manifests itself not only on the Nvidia Studio 457.30 driver, but also on the Nvidia Game Ready 460.79 (these are the latest drivers)

Unfortunately, I don't see any indication for an error within Blender.

Is this a bug in the Nvidia drivers? I installed blender version 2.90.1 and the error did not appear there yet.

@Maxim (Konus) it looks like that is the case. If you suspect otherwise, we will need exact steps to reproduce the problem and a minimal project file that demonstrates the issue.

@Maxim (Konus) This problem has been cropping up for me quite a lot, even after switching gpus and doing a fresh install. The only "fix" i have found is to just dual boot Linux onto my machine and run Blender from Ubuntu(No anomalous behaviour after 3 months). Although your mileage may vary.

Out of curiosity are you sporting an AMD or Intel CPU?

@Tamuka Tagwireyi (tamuka120)

I have an AMD Ryzen 5 1600X processor.

I was able to partially solve the problem (however, instead of 10-15 errors per day, 1-2 still happen) by cleverly installing Nvidia drivers. The fact is that there are two types of drivers on the Nvidia website - "Standart" and "DHC". And by default, when we go to the Nvidia website and download the drivers, we are shoved with a "DCH" package without a choice. But if you go to the tricky link https://www.nvidia.com/download/Find.aspx?lang=en# - then in the "Windows Driver Type:" field there is a choice. If we choose "Standart", then a packet of a smaller size is downloaded than "DCH". And this driver ("Standart") is more resistant to our bug described in this bug report. Of course, you need to do a complete reinstallation of the drivers with the cleaning of old settings (there is such an option when installing Nvidia drivers). However, such a trick does not completely solve the problem and the error in nvoglv64.dll still sometimes happens.

@Maxim (Konus) Based on your comment the problem doesn't seem to be resolved. Unfortunately without clear steps how to reproduce the issue, we are unable to investigate, identify and fix the problem, if the issue is in Blender.

Does it make any difference if your start Blender by double clicking on blender_debug_gpu_glitchworkaround.cmd?

@Robert Guetzkow (rjg)

I tested a blender working through blender_debug_gpu_glitchworkaround.cmd - the problem persists. I have two crashes with an error in the nvoglv64.dll module.

The problem cannot be reproduced in the sequence of steps. There are 847,000 triangles in my scene. Sometimes we can work for hours without error, but sometimes touching any element of the interface causes an immediate crash.

A few more considerations. I originally exported the scene from 3d max. Sometimes there is a feeling that the crash occurs after a while, when I switch between different types of renders, sometimes the crash occurs when editing an object, but most often at the moment of switching something in the interface. I'm asking myself, is it possible that the geometry in the scene is wrong? Or are the video memory (3 gigabytes) and RAM (16 gigabytes) full? Perhaps, with the intensive work of the video subsystem and exchange with the disk, some complex accumulation of errors occurs as a result of a memory leak? Or is it a subtle moment in the interaction of the video driver in the case when the blender is simply difficult to digest? My intuition doesn't tell me anything yet. In simple scenes like two or three boxes, I cannot reproduce the error. But if this happens, I will write a comment additionally.

I can only guess as well. I would have to reproduce the problem on my system to identify the issue, since the stack trace only points to the graphics driver. That would require a project file or how to create one and exact steps on how to reproduce the crash.

You can validate the objects in your file by copying the following script into the Text Editor in the Scripting workspace and then executing it.

import bpy

for obj in bpy.data.objects:
    if obj.type == "MESH" and obj.data is not None:
        print(f"{obj.name}: {obj.data.validate(verbose=True)}")

Use Window > Toggle System Console to show the console and check if there is any objects where it prints "True". That object would be invalid.

Running out of memory would also be a possible cause, but that would likely not fail this consistently in the same place (the graphics driver). You could open the Task Manager and keep an eye on the memory usage if you suspect that's the case.

This could also be a bug in Blender that result in undefined behavior that just happens to cause a crash when using the graphics driver. It could also be a bug in the graphics driver that occurs after a none obvious series of steps.

@Robert Guetzkow (rjg)

I ran the script and received several "True" messages. I will fix the scene and erroneous moments, and then test for failures.

If fixing the scene doesn't help, I'll figure out how to reproduce the error on your computer.

I too have been having random crashes over 5-10x a day. Same issue due to nvoglv64.dll. Typically happens when I switch tasks and then click back into blender to select an object. But also occurs during regular work in blender. I've DDU and changed to oldest driver I could download (456) also tried with graphic drivers in between.

The same issue regardless of project file but occurs more often on the larger the file sizes.

Current GPU is one Gigabyte RTX 3090 Eagle