Skip to content
View in the app

A better way to browse. Learn more.

Armbian Community Forums

A full-screen app on your home screen with push notifications, badges and more.

To install this app on iOS and iPadOS
  1. Tap the Share icon in Safari
  2. Scroll the menu and tap Add to Home Screen.
  3. Tap Add in the top-right corner.
To install this app on Android
  1. Tap the 3-dot menu (⋮) in the top-right corner of the browser.
  2. Tap Add to Home screen or Install app.
  3. Confirm by tapping Install.

Orange Pi 5 Ultra: Armbian 26.11 / Linux 6.18.60 test report and maintainer interest

Featured Replies

Hi everyone,

I have physical access to an Orange Pi 5 Ultra with 16 GB RAM and would like to help with testing and maintenance of this board. Here is my first round of hardware testing on the current/mainline image. I am posting the working features, reproducible problems and remaining gaps separately. This is an initial test report, not a release sign-off.

Image and test environment

  • Board: Orange Pi 5 Ultra, RK3588, 16 GB RAM.

  • Image: Armbian community 26.11.0-trunk.120, Ubuntu 26.04 Resolute, GNOME/Wayland, booted from microSD.

  • Kernel: 6.18.60-current-rockchip64. Image build commit: c24ea08f8759a37aaa83a101774582c37c18eda6.

  • Kernel, DTB and BSP packages: 26.11.0-trunk.120. Mesa: 26.0.8-1ubuntu0.3.

  • Storage: 64 GB SDXC card using SDR104 at approximately 148.5 MHz; Samsung SSD 980 500 GB on PCIe 3.0 x4; Toshiba MQ04UBF100 1 TB USB hard disk detected at SuperSpeed 5000M.

  • Ethernet: RTL8125B using r8169, linked to a 1 GbE LAN peer. Cooling fan present. PSU model/specification has not yet been recorded.

  • Tests were run on 5–6 November 2026, UTC+8, using remote shell/Python automation and local hands-on cable/desktop checks.

A baseline package list and boot log were saved before testing. The desktop updater subsequently updated userspace packages, including libc6 from 2.43-2ubuntu2.3 to 2.43-2ubuntu2.4. The recorded kernel/DTB packages and Mesa version remained as listed above. This should therefore not be treated as a completely untouched-image test for every later result. Additional test utilities were unpacked into a separate test directory.

I do not yet have an armbianmonitor -u upload URL from this session. Redacted baseline boot, hardware and package information is included with the test evidence; a formal support report is still incomplete.

Completed checks

  • Boot and desktop: SD boot, GNOME/Wayland desktop, HDMI output, USB keyboard and mouse were usable. This covers this desktop image only, not a separate minimal/CLI image.

  • GPU: accelerated OpenGL renderer reported Mali-G610 MC4 (Panfrost). glmark2 output validation passed. Vulkan enumeration with the Panfrost ICD and a 240-frame Wayland vkcube run completed successfully. Three short glmark2 scenes also completed; this was not the full default performance benchmark.

  • RAM: memtester completed one full pass over 2 GiB with exit status 0 in 836.51 seconds. The highest sampled temperature during that test was 53.6°C. This is not exhaustive coverage of all 16 GB.

  • CPU policy: three cpufreq policies were present, with observed peaks of 1.8 / 2.4 / 2.4 GHz during the memory test. The desktop tuning profile and ext4 commit=15 setting were present. A full CPU/VM soak result is still missing, as noted below.

  • RTC: HYM8563 time reads worked. A future, minute-aligned alarm triggered the RTC hardware interrupt at the requested RTC timestamp; its IRQ counter increased from 0 to 1. This was an awake-system alarm test, not a suspend wake-up or battery-backed power-off retention test.

  • Ethernet: TCP receive to the board was approximately 940 Mb/s and transmit from the board approximately 764 Mb/s. A bidirectional run also completed. IPv4, IPv6 and 1500-byte-MTU ping checks passed. The peer was limited to 1 GbE, so 2.5 GbE throughput was not validated.

Storage checks

fio 3.41 was used for bounded tests. SD and USB file tests used 1 MiB blocks, direct I/O, libaio queue depth 16 and CRC32C write/read verification. The NVMe test used direct, read-only access with a 4 GiB working set for 15 seconds. Rates below are decimal MB/s.

Device / scope

Write

Read

Result

SD, 2 GiB file

24.75 MB/s

65.81 MB/s

CRC32C verification passed

USB hard disk, 1 GiB file, first USB 3 port

59.08 MB/s

138.65 MB/s

CRC32C verification passed

NVMe, read-only

Not tested

1403.87 MB/s

Read test passed; health queries completed

The SD and USB 4 KiB random-read tests also completed. No new storage-related kernel errors were found during this test interval. These tests do not cover the entire capacity of the media. NVMe write verification and the second USB 3 port remain untested.

Open finding 1: onboard Wi-Fi interface is missing

At baseline, Ethernet and Bluetooth were present but no WLAN interface or wireless PHY appeared. The installed bcmdhd-sdio-dkms package was built for the running kernel, while the SDIO bus had no enumerated device. The live device tree contained:

mmc@fe2d0000 {
    status = "disabled";
};

Observed package/driver state:

bcmdhd-sdio-dkms: 101.10.591.52.27-10
bcmdhd-sdio/101.10.591.52.27-10, 6.18.60-current-rockchip64, aarch64: installed
iw dev / iw phy: no wireless interface or PHY listed
/sys/bus/sdio/devices/: empty

The vendor board DTS contains SDIO enablement and a Wi-Fi power sequence. Missing board-level SDIO/power-sequence description is a suspect; no proposed fix or overlay has been runtime-tested in this session.

Open finding 2: HDMI input self-loop did not lock

With no independent HDMI source available, I connected this board's HDMI OUT to its HDMI IN. A fullscreen colour-bar pattern was generated, and the receiver was given the v4l2-ctl HDMI EDID preset. DRM state confirmed a 1920×1080 at 60 Hz RGB 8 bpc transmitter mode, with a 148.5 MHz TMDS clock.

The receiver reported power_present=1, but v4l2-ctl --query-dv-timings failed with “No locks available”. The kernel logged:

hdmirx_phy_register_write wait cr write done failed
hdmirx_controller_init wait timer base lock failed
hdmirx_query_dv_timings: signal is not locked

The HDMI RX HDMI/DMA IRQ counters stayed at zero while the 5V-detect GPIO IRQ counter increased. Unloading/reloading synopsys_hdmirx and reapplying the test EDID reproduced the same result. The original receiver EDID and monitor connection were restored afterward.

This is a reproducible self-loop observation on this setup, not proof that every external HDMI source will fail. Testing with an independent source and checking a clean cold boot remain necessary. Relevant earlier changes include #10172 and the 6.18 patch adjustment #10180.

Incomplete and intentionally omitted checks

  • A 15-minute all-CPU plus 4 GiB VM verification run was started, but no final completion record has been collected. The last collected progress line was around 58 seconds. I am not counting this soak as passed.

  • Later, new SSH connections were closed before authentication while the board still answered LAN pings. The cause has not been established, and I am not attributing it to the load test or kernel without further evidence.

  • Bluetooth firmware loading, discovery and three rfkill recovery cycles were observed; pairing and data transfer were not validated. Further Bluetooth testing was intentionally skipped.

  • Actual audio playback/capture, the second USB 3 port, repeated warm/cold boot recovery, RTC power-off retention, other images/kernel branches and additional header/camera interfaces remain outside this completed test set.

Maintenance interest

I would like to contribute release-image testing and reproducible board reports for the Orange Pi 5 Ultra. In particular, #10673 explicitly noted missing Ultra/Linux 6.18 runtime coverage, and #11157 reported no board hardware testing for the CPU policy changes. The observations above provide some additional runtime evidence, with the limitations stated.

I have saved the scripts, package inventories and logs, and can provide specific redacted excerpts. Guidance on the highest-priority follow-up tests and the board-maintainer application process would be appreciated.

KingFalse

orangepi5-ultra-6.18-first-round-evidence.zip

  • Author

Follow-up to my earlier current 6.18 GNOME report: this round tests vendor 6.1 on Debian Minimal.

Board: Orange Pi 5 Ultra, 16 GB. Image: Armbian_26.11.0-trunk.127_Orangepi5-ultra_trixie_vendor_6.1.172_minimal.img.xz. Kernel: 6.1.172-vendor-rk35xx; build commit: e5a400a. This is a nightly/rolling build, not a release sign-off.

Tests ran on 10-11 November 2026 (UTC+8). Commands and log summaries used automation and AI assistance. A person confirmed display, audio, cable connections, pairing-code matching, and the power cycle.

Check

Result

Boot

SD boot/root expansion, software reboot and an approximately 30-second power-off/cold boot passed.

Networking

Ethernet RX/TX 935/762 Mb/s; 5 GHz Wi-Fi RX/TX 341/377 Mb/s. Ethernet IPv4/IPv6/MTU checks and Wi-Fi IPv4 pings passed.

SD / USB storage

2 GiB / 1 GiB CRC32C file checks passed. Write/read: 23.86/63.56 and 54.21/119.93 MiB/s.

NVMe

Read-only 1570.44 MiB/s; zero write bytes. Existing data preserved.

RAM / CPU

2 GiB memtester pass; 15-minute 8-CPU + 4 GiB VM verification passed. Peak core 76.7 C.

RTC / cooling

Future alarm IRQ 0 to 1 at the requested time; automatic fan PWM 0 to 255 observed.

Display / audio

HDMI console and 1080p60 RGB888 output worked; left/right 48 kHz stereo was confirmed locally.

The original 349-package set remained unchanged during core testing; additional test utilities ran from a separate directory. The Bluetooth supplement then added only rfkill=2.41.5-0+deb13u1 and bluez=5.82-1.1+rpt2, giving 351 packages. No other package was upgraded or removed. The workaround did not change the kernel, bootloader, DTB or firmware file.

On this Minimal image, Bluetooth initially failed before firmware loading. The enabled ap6611s-bluetooth.service could not execute /usr/sbin/rfkill (status 203/EXEC), and BlueZ management userspace was absent. After adding those dependencies, three logical power/scan cycles passed. SSP pairing with a Xiaomi Pad 6 Pro succeeded, with the matching code confirmed locally. Services started automatically after a whole-device software reboot; Paired, Bonded and Trusted remained true. L2CAP echo returned 10/10 replies both before and after reboot. Codec-capability -22 warnings remain; no HCI transport timeout was captured. These are post-workaround results, not an out-of-box Bluetooth pass. Bluetooth audio, OBEX, HID and LE GATT applications were not tested.

HDMI-IN same-board loopback did not lock under load or at idle. The transmitter was 1920x1080p60 RGB888 at 148.5 MHz. Input power was detected, and the 256-byte receiver EDID had valid checksums. QUERY_DV_TIMINGS returned ENOLCK; the kernel logged vertical-sync threshold events. An independent external HDMI source was unavailable, so the root cause remains unassigned.

On the second USB 3 port, the disk did not enumerate and its power LED stayed off. I skipped further checks of this port. Another controller reported device mode; hardware damage is unconfirmed. The Ethernet link was only 1 Gb/s, and the memory test did not cover all 16 GB. RTC battery-backed retention, desktops/3D/video decoding, installation/upgrade paths and other expansion interfaces remain untested.

The attached ZIP contains selected redacted evidence, exact command templates and the result matrix. Feedback on the missing Bluetooth dependencies and the HDMI-IN reproduction would be appreciated.

KingFalse

orangepi5-ultra-vendor127-validation-evidence.zip

  • Author

Follow-up: Ubuntu Resolute Minimal .128 validation (11 November 2026)

I have completed another bounded validation run on my Orange Pi 5 Ultra, complementing the earlier Debian results.

Image: Armbian_26.11.0-trunk.128_Orangepi5-ultra_resolute_vendor_6.1.172_minimal.img.xz
Ubuntu 26.04 Resolute Minimal; kernel 6.1.172-vendor-rk35xx; build commit e5a400a.
Hardware: 16 GB RAM, approximately 59.5 GiB SD card, root filesystem on /dev/mmcblk1p1.

The imported first-boot preset was consumed successfully. Ethernet, Wi-Fi and automatic NTP synchronization worked. Initial IPv6 NTP queries timed out, then IPv4 fallback synchronized the clock around 52 seconds after boot.

Area

Result

Memory

PASS: one locked 2 GiB memtester pass; all 18 checks completed successfully.

CPU / VM

PASS: approximately 15 minutes, eight CPU workers plus one 4 GiB VM worker with verification; peak 76.7 C.

SD storage

PASS: 2 GiB CRC32C file verification; write 23.74 MiB/s, verification read 64.23 MiB/s.

NVMe

PASS: unmounted, raw read-only test at 1,568.35 MiB/s; no host-written sector increase.

Network

Functional PASS: 12/12 checks; Ethernet and Wi-Fi observations detailed below.

RTC

PASS: HYM8563 future-minute alarm fired, incremented its interrupt counter, and was cleared.

HDMI / keyboard

PASS: visible console refresh and USB keyboard Enter response confirmed manually.

HDMI audio

PASS: 48 kHz S16_LE stereo playback; left and right channels confirmed audible manually.

Boot persistence

Core PASS: two warm reboots and one cold boot after approximately 30 seconds without power.

Stock Bluetooth

FAIL: missing rfkill / BlueZ; details below.

Ethernet negotiated 1 Gb/s with my router. Separate TCP tests measured approximately 937 Mb/s board RX and 771 Mb/s board TX. Wi-Fi measured approximately 326 Mb/s RX and 373 Mb/s TX. RX means router to board; TX means board to router. These are receiver-reported decimal Mbit/s, not MiB/s. The peer cannot validate 2.5GbE performance.

Wi-Fi accumulated 232 additional TX drops during testing, with retransmissions. The iw RX packet count was also inconsistent with kernel interface counters. Similar observations already existed in my Debian .127 evidence; this does not establish a new Ubuntu regression.

HDMI controller state was 2560x1440p59.95. Reboot checks confirmed changed boot IDs, the expected SD root, working network, synchronized NTP and an unchanged installed-package set. Test utilities and their libraries came from 61 DEBs in the configured Ubuntu/Armbian repositories, extracted into a separate directory. No test packages were installed into the system.

The stock image lacks rfkill and BlueZ clients. ap6611s-bluetooth.service fails at ExecStartPre with 203/EXEC before firmware loading. I did not patch this image, and Bluetooth pairing remains untested. This Ubuntu image was uploaded at 03:24 UTC on 11 November, before dependency-fix PR #11331 merged at 05:26 UTC:
https://github.com/armbian/build/pull/11331

This confirms the packaging problem in the older image; it does not validate the merged fix on Ubuntu. I am keeping this result with the existing issue rather than opening a duplicate.

Power supply rating: 5 V / 5 A. Two ML307 modules and one 2454 radar remained GPIO powered, with unchanged wiring. Their serial functions and power margin were not tested. No USB storage device was connected this round.

Not covered: USB storage/other ports, HDMI input, GPU rendering, codecs, NPU inference, camera/MIPI, eMMC/SPI or NVMe boot, analog audio, mouse interaction, RTC battery retention, full-capacity RAM/media testing, or extended stability.

Command execution, evidence collection and log review were AI-assisted. I performed the physical actions and confirmed the display, keyboard and audio observations. This is bounded single-board testing, not full release certification.

Selected redacted evidence archive: orangepi5-ultra-ubuntu128-validation-evidence.zip

KingFalse

orangepi5-ultra-ubuntu128-validation-evidence.zip

  • Author

Follow-up to my earlier Orange Pi 5 Ultra reports: I tested the newer Debian 13 vendor Minimal image on 13 November 2026 (Asia/Taipei), focusing on Bluetooth startup and reboot recovery.

Official image: Armbian_26.11.0-trunk.130_Orangepi5-ultra_trixie_vendor_6.1.172_minimal.img.xz; kernel 6.1.172-vendor-rk35xx. Booted from the native TF/microSD slot. I retained my first-boot preset, including Tailscale state restoration. The original image file was not independently hashed this round. No packages were installed, upgraded or repaired during testing, and the beginning/end package inventories matched; this comparison starts after first-boot setup.

  • Bluetooth: rfkill 2.41.5-0+deb13u1 and bluez 5.82-1.1 were preinstalled. Both ap6611s-bluetooth.service and bluetooth.service were active before Bluetooth client queries. The earlier missing-rfkill 203/EXEC failure was not observed. Pairing with Xiaomi Pad 6 Pro initially timed out during confirmation (AuthenticationCanceled); after rediscovery, pairing succeeded. This was not a first-attempt success.

  • Boot recovery: 5/5 warm reboots and 1/1 cold boot passed; the cold test included approximately 30 seconds disconnected from power. Each boot automatically restored the controller, Paired/Bonded state and Tailscale connectivity. Initial checks and another check after 15 seconds found no HCI initialization timeout or failed systemd units. No recovery commands were needed. These limited repetitions cannot rule out occasional failures or establish an upper bound on startup time.

  • Ethernet: negotiated 1 Gb/s full duplex; gateway pings 10/10 and WAN pings 5/5. DNS resolution and the APT endpoint response (HTTP 302) worked. This checks endpoint reachability, not a complete package download or installation.

  • Wi-Fi: the initial dual-interface, same-subnet setup had packet loss. Header capture showed five requests leaving WLAN but their five replies entering Ethernet. With a temporary router-neighbor fixture scoped to this board, WLAN sent and received 5/5 packets with zero loss. Dynamic neighbor learning was restored afterward. This conditional short test does not establish sustained throughput or long-term wireless stability.

Remaining observations: /etc/udev/rules.d/60-media.rules is absent; MPP, RGA and DMA heap device nodes remain root:root 0600, without permission changes for testing. Codec-capability -22 warnings remain in the logs; these were not classified as HCI initialization failures. Other kernel warnings also remain. Related work: PR #11331 and PR #11353.

Desktop environments, hardware video decoding, Bluetooth audio/application data transfer, the full peripheral matrix and RTC retention were not tested in this round. Commands and log summaries used automation and AI assistance; I confirmed device pairing and the physical power cycle. A selected, sanitized evidence summary is attached as orangepi5-ultra-debian130-validation-evidence.zip. This is a scoped regression report, not formal release sign-off.

orangepi5-ultra-debian130-validation-evidence.zip

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

Account

Navigation

Search

Search

Configure browser push notifications

Chrome (Android)
  1. Tap the lock icon next to the address bar.
  2. Tap Permissions → Notifications.
  3. Adjust your preference.
Chrome (Desktop)
  1. Click the padlock icon in the address bar.
  2. Select Site settings.
  3. Find Notifications and adjust your preference.