Shenzhen RK PVT — Remote Test, Round 2 (prd5)
2026-10-26 · Wilber Zheng · times are UTC+7 · every test ran remotely over SecureLink · round 1: 2026-10-21, dbg455
Summary: the PVT board now runs the production-channel image prd5. Every functional, video, soak, service and interface test in the suite passed (34 of 34 results across 32 cases, plus 23 of 23 interface checks), and no board hardware fault was found. The render benchmark now runs and gives numbers, but no pass/fail verdict, because there is no baseline to compare with. Three issues remain: the board rebooted during the CPU/memory benchmark, NTP cannot sync on a China network, and the troubleshoot wizard stalls on a China network (a player bug). The wizard case was removed from the Shenzhen suite.
1. Environment and device
Device under test
| Item | Value |
|---|---|
| Device | RK3576 ProPlayer Gen5 PVT board (the same board as round 1), reflashed on site with prd5 and paired again. Board: Giada DN84 (gaidatech-evb-giada-dn84) |
| Image | prd5 · channel prd · git e061f212a (develop-rockchip) · bake 20261025T165343Z-3132212908-e061f212a · built 2026-10-25. Local copy: pp5-rk3576-os6-prd5-proplayer-gen5-evt-firstboot-welcome-e061f212a.img |
| Kernel / firmware | 6.1.99-rockchip-standard · firmware version 2642298.5 |
| Player / utils | 5.7.2-P3-RK / 4.0.1 |
| Hardware | 8-core CPU (governor performance, big cores 2208 MHz) · RAM 3.99 GB · eMMC 58.3 GB (TY2964, made 10/2025) · zram 7.6 GB · 43–47 °C at idle |
| Identity | The reflash gave the board a new identity. UUID 999e526f8112967647a2d5799e96f9fb513b · backend _id 6adebfeac9f7827064730c74 · serial a92c55baf63fbb9b · MAC 84:fc:14:24:d1:08. The old screen record was unpaired |
| Portal | Screen name rk_pvt_shenzhen · account testing10 (team FsnNPHeasJZGW5mgq) · folder "Test Bench — wilber-msa2-ubuntu (Wilber's desk — remote)" · time zone Asia/Shanghai |
| Content | Playlist 2Bws8X3cnD3Q23tfg, the same as before the reflash. Video player type: default |
Network
| Item | Value |
|---|---|
| Site network | Wi-Fi on 5 GHz (5300 MHz), signal −54 dBm, 258 Mbit/s. Ethernet not connected. The site reported a Wi-Fi outage for part of the afternoon |
| SecureLink | 100.65.184.207 · region APAC · connected |
| Reachable from the board | optisigns.com answers in about 3 s. www.google.com never answers: no response within 25 s, because the packets are dropped, not refused |
| Time | Time zone Asia/Shanghai is correct. NTP is not synchronized: the only configured servers are time1–4.google.com. The hardware clock still reads 2026-05-03 |
Peripherals
| Peripheral | State |
|---|---|
| Monitor | Xiaomi Mi Monitor on HDMI-A-1, 1920×1080 @ 100 Hz (preferred mode 60 Hz). Switched to 60 Hz only during S7b, then back to 100 Hz |
| USB | USB 2.1 hub, Compx 2.4G keyboard/mouse receiver, AIC8800D80 Wi-Fi/BT, USB 3.2 hub |
| Not present | SD card, capture card, smart plug, serial. Cases that need them are not in this suite |
Test side
| Item | Value |
|---|---|
| Test kit | The same local remote-board suite as round 1 (the harness repo is not modified). It runs stages in order with a link check before and after each stage, reruns each failure once, and restores the portal name, folder, content, video player type and player after each stage |
| Image cases | OS-IMG-001/002/003 ran against the root filesystem extracted from the downloaded prd5 image. Its BUILD_ID matches the board |
| Kit changes this round | The rig now finds the board's portal record by its hardware serial before each run, so a reflash or re-pair needs no hand edit · OS-CFG-001 reads what the bake actually found for dev-only items on a prd image (dbg behaviour unchanged) · an unset video player type counts as "default" · S7b switches the output to 60 Hz for the stage and runs a 5-minute version of gate-lite (one pass per item, preheat ≤ 60 s) · S7b's benchmark pages were missing from the kit's harness copy and are now present · SYS-FN-055 removed from the Shenzhen suite |
| Timeline | 10:37 first full run (stopped after S5 on a restore-check bug in the kit) · 12:03 full rerun after the kit fixes · 13:38 S7a, board rebooted · 13:44–14:20 SYS-FN-055 retests with screenshots · 14:49 S7b and S8 · 15:44 S7b at 60 Hz · 15:58 done |
2. Results by stage
| Stage | What | Pass | Fail | Skip | Time | Data | Notes |
|---|---|---|---|---|---|---|---|
| S1 | Read-only OS health (12) | 12 | 0 | 0 | 6 min | 0.7 MB | Includes the three image cases against the prd5 image |
| S2 | Device management via backend (7) | 7 | 0 | 0 | 5 min | 1.9 MB | SYS-AN-001 (proof of play) failed in the 10:37 run because the screen had just been renamed; it passes here |
| S3 | Player kill and respawn (1) | 1 | 0 | 0 | 1 min | 0.3 MB | |
| S4 | Content push and web (3) | 3 | 0 | 0 | 12 min | 5.1 MB | SYS-FN-055 failed in every run on this network (the troubleshoot wizard stalls before Content) and is now removed |
| S5 | mpv video (5 cells) | 5 | 0 | 0 | 3 min | 0.5 MB | |
| S5b | Playback soak (4, pl-gen5-golden) | 4 | 0 | 0 | 49 min | 129 MB | Numbers in §3 |
| S6 | Services that may be blocked in CN (2) | 2 | 0 | 0 | 15 min | 5.1 MB | Power BI soak and S3 upload pass |
| S7a | CPU / memory (hwbench) | 0 | board rebooted | — | 2 min | — | The board reset about one minute after the watchdog daemon was parked and came back by itself; cause not found. Not rerun: a reboot that fails to come back needs someone on site |
| S7b | Render benchmark (5-min, 60 Hz) | — | 0 | — | 14 min | 1.2 MB | Numbers but no verdict: no baseline to compare with. Two passes, no failed item |
| S8 | Interface probe (23) | 23 | 0 | 0 | 1 min | 0.2 MB | Read-only sysfs/proc checks |
Data used on the PVT in this round: about 145 MB. After every stage the kit confirmed the screen name, folder, content, video player type and player state were back as before. The monitor was confirmed back at 100 Hz after S7b.
3. Performance
Render benchmark (S7b: 1920×1080 @ 60 Hz, two passes)
| Metric | Pass 1 | Pass 2 | Notes |
|---|---|---|---|
| Canary (rig health) | 59.6 fps, p95 16.7 ms | 60.0 fps, p95 16.8 ms | healthy |
| rAF mean / p95 frame | 60.0 fps / 16.7 ms | 60.0 fps / 16.8 ms | No dropped frames |
| MotionMark Multiply | 282.4 | 280.8 | |
| Aquarium, 500 fish (1280×720) | 47.4 fps | 48.0 fps |
WebGL renderer: ANGLE on Mesa, Mali-G52 r1 (Panfrost). System during the run: load at most 2.49, minimum available memory 2.85 GB, no swap, max temperature 46.2 °C, renderer CPU peak 130%.
Playback soak (S5b, pl-gen5-golden)
OS-RES-001, 20 minutes: player memory 0.99–1.50 GB (median 1.27 GB), minimum MemAvailable 2.26 GB, no swap used, CPU mean 23%, max temperature 47.2 °C, no OOM kill, no restart. SYS-REL-008: player memory median 1274 MB in the first half and 1253 MB in the second (×0.98, peak 1512 MB), no mpv frame drops.
CPU / memory (S7a)
No numbers this round: the board rebooted during the run. Round 1 numbers on dbg455: CoreMark 10582 iter/s, STREAM triad 10192 MB/s.