[Cin] How to access a second Intel iGPU in a workstation with additional discrete Nvidia GPU?

Tue Sep 10 17:34:26 CEST 2024

Den 31.08.2024 00:58, skrev Terje J. Hanssen:
> I am stuck how to configure QSV access to an enabled iGPU onboard an 
> Intel Core i7-6700K (Skylake). This is a legacy, rebuilt workstation 
> with additional, discrete Nvidia GeForce GTX 960 GPU on a MSI-Z170-A 
> Pro mobo.  Any idea and code modification below?
>
> Details as follows:
>
> inxi -CGMSz
> System:
>   Kernel: 6.10.5-1-default arch: x86_64 bits: 64
>   Desktop: GNOME v: 46.3.1 Distro: openSUSE Tumbleweed-Slowroll 20240803
> Machine:
>   Type: Desktop Mobo: MSI model: Z170-A PRO (MS-7971) v: 1.0
>     serial: <superuser required> UEFI: American Megatrends v: 1.K0
>     date: 07/10/2018
> CPU:
>   Info: quad core model: Intel Core i7-6700K bits: 64 type: MT MCP cache:
>     L2: 1024 KiB
>   Speed (MHz): avg: 800 min/max: 800/4200 cores: 1: 800 2: 800 3: 800 
> 4: 800
>     5: 800 6: 800 7: 800 8: 800
> Graphics:
>   Device-1: Intel HD Graphics 530 driver: i915 v: kernel
>   Device-2: NVIDIA GM206 [GeForce GTX 960] driver: nvidia v: 550.100
>   Display: wayland server: X.org v: 1.21.1.12 with: Xwayland v: 24.1.1
>     compositor: gnome-shell driver: X: loaded: modesetting unloaded: 
> fbdev,vesa
>     gpu: nvidia,nvidia-nvswitch resolution: 1920x1080~60Hz
>   API: OpenGL v: 4.6.0 vendor: nvidia v: 550.100 renderer: NVIDIA GeForce
>     GTX 960/PCIe/SSE2
>   API: Vulkan v: 1.3.290 drivers: N/A surfaces: xcb,xlib,wayland
>   API: EGL Message: EGL data requires eglinfo. Check --recommends.
>
>
> lspci -k | grep -EA3 'VGA|3D|Display'
> 00:02.0 Display controller: Intel Corporation HD Graphics 530 (rev 06)
>     DeviceName:  Onboard IGD
>     Subsystem: Micro-Star International Co., Ltd. [MSI] Device 7971
>     Kernel driver in use: i915
> --
> 01:00.0 VGA compatible controller: NVIDIA Corporation GM206 [GeForce 
> GTX 960] (rev a1)
>     Subsystem: Micro-Star International Co., Ltd. [MSI] Device 3201
>     Kernel driver in use: nvidia
>     Kernel modules: nouveau, nvidia_drm, nvidia
>
> =====================================================================
>
> Trying to test QSV via FFmpeg benchmark similar like this Vulkan code
> ffmpeg -init_hw_device "vulkan=vk:0" -hwaccel vulkan 
> -hwaccel_output_format vulkan -i INPUT -f null - -benchmark
> https://trac.ffmpeg.org/wiki/HWAccelIntro#Vulkan
>
> cause an output error which reports: "device type qsv needed".
> Intel Skylake with HD 530 graphics gen 9, should be capable to decode 
> and encode MPEG-2/H.264/H.265 according to
> https://trac.ffmpeg.org/wiki/Hardware/QuickSync#HardwareSupport
>
> ffmpeg -hide_banner -hwaccel qsv -hwaccel_output_format qsv -i 
> h264_8bit_yuv420p.mp4 -f null - -benchmark
> [AVHWDeviceContext @ 0x557526873500] Error creating a MFX session: -9.
> Device creation failed: -1313558101.
> [vist#0:0/h264 @ 0x55752685c6c0] [dec:h264_qsv @ 0x55752685b300] No 
> device available for decoder: device type qsv needed for codec h264_qsv.
> [vist#0:0/h264 @ 0x55752685c6c0] [dec:h264_qsv @ 0x55752685b300] 
> Hardware device setup failed for decoder: Unknown error occurred
> [vost#0:0/wrapped_avframe @ 0x55752685a740] Error initializing a 
> simple filtergraph
> Error opening output file -.
> Error opening output files: Unknown error occurred
> bench: maxrss=67452KiB
>
>
> I have tried to define and add a qsv device according to the last 
> section here
> https://trac.ffmpeg.org/wiki/Hardware/QuickSync#Transcode
>
> |-qsv_device| is an qsv customized option can be used to specify a 
> hardware device and avoid the default device initialization failure 
> when multiple devices usable (eg: an Intel integrated GPU and an 
> AMD/Nvidia discrete graphics card). One example on Linux (more details 
> please see https://trac.ffmpeg.org/ticket/7649 
> <https://trac.ffmpeg.org/ticket/7649>)
>
> ffmpeg -hwaccel qsv -qsv_device /dev/dri/renderD128 -c:v h264_qsv -i 
> input.mp4 -c:v h264_qsv output.mp4
>
> and also had a look at this answer
> https://stackoverflow.com/questions/44747381/i-cant-use-hw-encoder-of-qsv-by-ffmpeg#47145215 
>

..... snip

In the meantime I have also queried this QSV issue on the ffmpeg-user 
mailing list, where I got some hints that helped me to solve it.
So to not leave this as an open issue here, I repost my reply to that list:

Yes, now I have verified that "libmfx" runtime was not installed by 
default on the Skylake workstation  (possibly because I enabled its iGPU 
in BIOS afterwards).
My ffmpeg had "libvpl" enabled, and "libmfx-gen1_2" was installed by 
default. When

When I add-installed "libmfx1", QSV via FFmpeg worked as it should:

ffmpeg -hide_banner -hwaccel qsv -hwaccel_output_format qsv -i 
h264_8bit_yuv420p.mp4 -f null -

1) On  Skylake i7-6700K / HD-530 iGPU:
frame= 1780 fps=882 q=-0.0 Lsize=N/A time=00:01:11.20 bitrate=N/A 
speed=35.3x

2) On KabyLake i7-8550U / UHD-620 iGPU:
frame= 1780 fps=795 q=-0.0 Lsize=N/A time=00:01:11.20 bitrate=N/A 
speed=31.8x

3) On AlderLake i7-12700KF + DG2 A750 GPU:
frame= 1780 fps=833 q=-0.0 Lsize=N/A time=00:01:11.20 bitrate=N/A 
speed=33.3x

What surprise me here, is that the current most powerful workstation 3), 
is not faster here than the legacy 1) and 2)

References:
https://github.com/intel/libvpl?tab=readme-ov-file#architecture
https://github.com/Intel-Media-SDK/MediaSDK?tab=readme-ov-file#id10
https://wiki.archlinux.org/title/FFmpeg#Intel_QuickSync_(QSV)

And as orientation for possibly other QSV interested in reading more of 
this thread at ffmpeg-user, here is the url:
https://ffmpeg.org/pipermail/ffmpeg-user/2024-September/058677.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20240910/b21d08d3/attachment.htm>