[Question]: AMF initialization issue for VCE encoding in NoMachine (nxserver/nxnode) on Radeon HD 8330E (Kabini, VCE 2.0) #567
Description
Description
I am trying to use AMF for hardware H.264 encoding in NoMachine (processes nxserver / nxnode) on a system with an AMD Radeon HD 8330E (integrated in AMD GX-424CC SoC, GCN 1.1 "Kabini", VCE 2.0) in HP T620 terminal.
GPU identification:
root@kszwaba-HP-t620-Quad-Core-TC:/home/kszwaba# dmesg | grep -iE 'vce|uvd|amdgpu|radeon'
[ 0.459555] smpboot: CPU0: AMD GX-415GA SOC with Radeon(tm) HD Graphics (family: 0x16, model: 0x0, stepping: 0x1)
[ 5.829795] [drm] amdgpu kernel modesetting enabled.
[ 5.829806] [drm] amdgpu version: 6.12.12
[ 5.830117] amdgpu: Virtual CRAT table created for CPU
[ 5.830168] amdgpu: Topology: Add CPU node
[ 5.872846] amdgpu: PeerDirect support was initialized successfully
[ 5.873620] amdgpu 0000:00:01.0: amdgpu: detected ip block number 0 <cik_common>
[ 5.873628] amdgpu 0000:00:01.0: amdgpu: detected ip block number 1 <gmc_v7_0>
[ 5.873633] amdgpu 0000:00:01.0: amdgpu: detected ip block number 2 <cik_ih>
[ 5.873638] amdgpu 0000:00:01.0: amdgpu: detected ip block number 3 <gfx_v7_0>
[ 5.873642] amdgpu 0000:00:01.0: amdgpu: detected ip block number 4 <cik_sdma>
[ 5.873646] amdgpu 0000:00:01.0: amdgpu: detected ip block number 5 <kv_dpm>
[ 5.873650] amdgpu 0000:00:01.0: amdgpu: detected ip block number 6 <dce_v8_0>
[ 5.873654] amdgpu 0000:00:01.0: amdgpu: detected ip block number 7 <uvd_v4_2>
[ 5.873658] amdgpu 0000:00:01.0: amdgpu: detected ip block number 8 <vce_v2_0>
[ 5.924848] amdgpu 0000:00:01.0: amdgpu: Fetched VBIOS from ROM BAR
[ 5.924858] amdgpu: ATOM BIOS: 1950150001
[ 5.925014] kfd kfd: amdgpu: KABINI not supported in kfd
[ 5.925282] amdgpu 0000:00:01.0: vgaarb: deactivate vga console
[ 5.925291] amdgpu 0000:00:01.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported
[ 5.926938] amdgpu 0000:00:01.0: amdgpu: VRAM: 1024M 0x0000000F00000000 - 0x0000000F3FFFFFFF (1024M used)
[ 5.926946] amdgpu 0000:00:01.0: amdgpu: GART: 1024M 0x0000000000000000 - 0x000000003FFFFFFF
[ 5.927192] [drm] amdgpu: 1024M of VRAM memory ready
[ 5.927198] [drm] amdgpu: 3429M of GTT memory ready.
[ 5.929771] [drm] amdgpu: dpm initialized
[ 5.929870] [drm] AMDGPU Display Connectors
[ 5.930400] [drm] Found UVD firmware Version: 1.64 Family ID: 9
[ 5.931017] [drm] Found VCE firmware Version: 50.10 Binary ID: 2
[ 5.957746] [drm] UVD initialized successfully.
[ 6.067080] [drm] VCE initialized successfully.
[ 6.068386] amdgpu 0000:00:01.0: amdgpu: SE 1, SH per SE 1, CU per SH 2, active_cu_number 2
[ 6.069357] amdgpu: legacy kernel without apple_gmux_detect()
[ 6.069363] amdgpu 0000:00:01.0: amdgpu: Runtime PM not available
[ 6.071017] [drm] Initialized amdgpu 3.63.0 20150101 for 0000:00:01.0 on minor 0
[ 6.170497] fbcon: amdgpudrmfb (fb0) is primary device
[ 6.170521] amdgpu 0000:00:01.0: [drm] fb0: amdgpudrmfb frame buffer device
[ 7.100670] [drm:amdgpu_device_ip_set_powergating_state [amdgpu]] *ERROR* set_powergating_state of IP block <vce_v2_0> failed -22
The problem is that AMF initialization for VCE encoding appears to fail silently — NoMachine falls back to software encoding.
I have tested both the newest and oldest available AMDGPU PRO drivers (with AMF) from the official AMD repositories:
Newest driver tested
Source: https://repo.radeon.com/amdgpu-install/latest/ubuntu/jammy/
ii amdgpu-core 1:6.4.60402-2187269.22.04
ii amdgpu-dkms 1:6.12.12.60402-2187269.22.04
ii amdgpu-dkms-firmware 1:6.12.12.60402-2187269.22.04
ii amdgpu-install 6.4.60402-2187269.22.04
ii amdgpu-pro-core 25.10-2187634.22.04
ii amf-amdgpu-pro:amd64 1.4.37-2187634.22.04
ii libamdenc-amdgpu-pro:amd64 25.10-2187634.22.04
ii libdrm-amdgpu-amdgpu1:amd64 1:2.4.124.60402-2187269.22.04
ii libdrm-amdgpu-common 1.0.0.60402-2187269.22.04
ii libdrm-amdgpu1:amd64 2.4.113-2~ubuntu0.22.04.1
ii libdrm2-amdgpu:amd64 1:2.4.124.60402-2187269.22.04
ii xserver-xorg-video-amdgpu 22.0.0-1ubuntu0.2
Oldest driver tested
Source: https://repo.radeon.com/amdgpu/23.20.00.48/ubuntu/
ii amdgpu-core 1:5.7.00.48.50700-1666569.22.04
ii amdgpu-dkms 1:6.2.4.50700-1666569.22.04
ii amdgpu-dkms-firmware 1:6.2.4.50700-1666569.22.04
ii amdgpu-install 5.7.00.48.50700-1666569.22.04
ii amdgpu-pro-core 23.20.00.48-1666589.22.04
ii amf-amdgpu-pro:amd64 1.4.31-1666589.22.04
ii libamdenc-amdgpu-pro:amd64 1.0-1666589.22.04
ii libdrm-amdgpu-amdgpu1:amd64 1:2.4.115.50700-1666569.22.04
ii libdrm-amdgpu-common 1.0.0.50700-1666569.22.04
ii libdrm-amdgpu1:amd64 2.4.113-2~ubuntu0.22.04.1
ii libdrm2-amdgpu:amd64 1:2.4.115.50700-1666569.22.04
ii xserver-xorg-video-amdgpu 22.0.0-1ubuntu0.2
ii vulkan-amdgpu-pro:amd64 23.20.00.48-1666589.22.04
I also attempted to test with suggested by NoMachine driver version amdgpu-install_22.20.50200-1_all.deb but some required packages are no longer available.
Observations / strace output
nxnode opens /opt/amdgpu-pro/lib/x86_64-linux-gnu/libamfrt64.so but encoding still falls back to software.
Example from lsof:
nxnode.bin 9216 mem REG 8,2 8359976 /opt/amdgpu-pro/lib/x86_64-linux-gnu/libamfrt64.so.1.4.31
From strace:
openat(AT_FDCWD, "/sys/class/drm/renderD128/device/vendor", O_RDONLY) = ...
/sys/class/drm/renderD128/device/vendor -> 0x1002 (AMD)
System details
- OS: Linux Mint 21.3 (based on Ubuntu 22.04.4 LTS)
- Kernel: 5.15.0-144-generic
- GPU: AMD Radeon HD 8330E (VCE 2.0, GCN 1.1 "Kabini")
- CPU: AMD GX-424CC SoC
- AMF library: from amdgpu-pro package (see versions above)
- Application: NoMachine 9.1.24 (nxserver/nxnode)
- DISPLAY: X11 session, MATE desktop
Question
Does AMF currently support VCE-based hardware encoding on Radeon HD 8330E / Kabini (VCE 2.0) with the above driver versions?
If yes, is there a known requirement or workaround to ensure that AMF initializes properly in a headless/non-GUI process like nxnode?