Linux, 375.20, cuvid, hevc, 10-bits - failed to decode first GOP

[max_verem]

Hi,

I faced problem with hardware accelerated decoding of 10-bits HEVC after upgrading from 370.28 to 375.20.

Artifacts happens on a first GOP (first 50 frames), second GOP decoded fine.

decoded with 370.28: http://downloads.m1stereo.tv/13fb80594a826e9e2d9a17c1ce72702b/HAB_00204_cuvid-370.28.jpg - FINE

decoded with 375.20: http://downloads.m1stereo.tv/13fb80594a826e9e2d9a17c1ce72702b/HAB_00204_cuvid-375.20.jpg - CURRUPTED first GOP

decided with native ffmpeg’s decoder: http://downloads.m1stereo.tv/13fb80594a826e9e2d9a17c1ce72702b/HAB_00204_native_ffmpeg.jpg - FINE

source file used for as test source: http://downloads.m1stereo.tv/13fb80594a826e9e2d9a17c1ce72702b/HAB_00204.265

[philipl]

So, keep in mind that 10bit decode is not officially supported by nvidia, and the ability to turn it on is undocumented.

It’s quite possible there’s a bug in there somewhere - we really need to wait until the Video SDK 8.0 gets released and they officially designate a working driver version.

I’ve not seen this kind of corruption on any sample except yours. How was it encoded? Is it using any of the fancier hevc features?

[max_verem]

@philipl

So, keep in mind that 10bit decode is not officially supported by nvidia, and the ability to turn it on is undocumented.

It’s quite possible there’s a bug in there somewhere - we really need to wait until the Video SDK 8.0 gets released and they officially designate a working driver version.

I know, but i hoped no further regressions could be.

@philipl

I’ve not seen this kind of corruption on any sample except yours. How was it encoded? Is it using any of the fancier hevc features?

i will try to request a encoder setting which used to encode that videos.