nvidia sample programe BSOD with gpu-p

[jack2007]

Hi, I setup hyper-v gpu-p in my computer with esay-gpu-pv script. When I run nvidia sample programe in nvidia sdk, the windows host get BSOD. These sample programes are AppEncD3D9/AppEncD3D11/AppEncCuda in SDK package\Samples\AppEncode directory. By the way, the AppEncD3D12 sample is OK in gpu-p.

host&guest windows: win11 10.0.22631.2861 nvidia gpu driver: 546.33 nvidia sdk version: Video_Codec_SDK_11.1.5 gpu: geforce 3060

Best Wishes

[kittuov]

Hi I'm facing the same issue.

In this case it's a piece of code i wrote. And i was able to narrow it down to "end of nvenc session". When i try to close the nvenc session, this is happening. I couldn't find any logs in event viewer

It also happens when I try to force idr refresh in nvenc encoder.

[devusr1x]

Try using 517.40 and earlier (see https://github.com/LizardByte/Sunshine/issues/2141) and please let me know what happens.

[sschamp]

517.40 does not support RTX 4090 :( Lowest supported driver is 522.25, which already has NVIDIA Video Codec SDK v12.0 And thus crashes the Host

[sschamp]

Strange thing is, on 536.99 this just works on the VM:

ffmpeg.exe -y -hwaccel cuda -hwaccel_output_format cuda ^
-f lavfi -i testsrc -t 50 ^
-vf hwupload -fps_mode passthrough -c:a copy -c:v av1_nvenc -b:v 4M -f null - ^
-vf hwupload -fps_mode passthrough -c:a copy -c:v av1_nvenc -b:v 1M -f null - ^
-vf hwupload -fps_mode passthrough -c:a copy -c:v av1_nvenc -b:v 8M -f null - ^
-vf hwupload -fps_mode passthrough -c:a copy -c:v av1_nvenc -b:v 6M -f null - ^
-vf hwupload -fps_mode passthrough -c:a copy -c:v av1_nvenc -b:v 5M -f null -

[sschamp]

Adding -outputInVidMem 1 to the sample apps mentioned above, does not BSOD the host.

Somehow it seems that some implementations do:

    RegisterOutputResources();
...
    FlushEncoder();
    ReleaseOutputBuffers();

And others don't.

So rewrite your code to use the implementation of:

Video_Codec_SDK_12.2.72/Samples/NvCodec/NvEncoder/NvEncoderOutputInVidMemCuda.cpp or Video_Codec_SDK_12.2.72/Samples/NvCodec/NvEncoder/NvEncoderOutputInVidMemD3D11.cpp

[sschamp]

Try Server 2025 (vNext Preview), GPU-P runs more stable there.

[devusr1x]

@sschamp Can you please explain what do you mean by that ("more stable")? Did you manage to make it work without specific flags? What's your setup?

[sschamp]

@sschamp Can you please explain what do you mean by that ("more stable")? Did you manage to make it work without specific flags? What's your setup?

Since 555 drivers there are no more crashes. The main reason for this is that they can use WDDM 3.2 on Server 2025 (vNext).

But I have since switched to Windows 11 24H2 (Insider Preview Dev Channel), as that also has WDDM 3.2 and is more suited for my needs.

[devusr1x]

Interesting, thanks for the info.