You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
J700 now has deep-sleep (suspend-to-RAM) reports at 0.049 W in an open series, speaker playback within 0.1 dB of macOS at full volume under speakerguardd v2, a working Neural Engine and hardware video encoders, and Linux running at EL2 through aura. Each result belongs to the development kernel its PR names. Deep sleep is not merged, suspend-to-idle standby power is unmeasured, the front port carries no display, and no public release install is qualified.
Refreshed November 8, 2026. Current public branches, PR states, follow-up reports and source were checked. No new hardware test was run for this update; the results below are those reported in the linked PRs, on the kernels they name. Source integration, individual subsystem tests and a complete laptop qualification are separate milestones.
What changed · November 8, 2026
Deep sleep: open linux #1208 adds suspend-to-RAM at EL2 and needs open aura #117 and aura #112. Its test kernels completed 30/30 deep sleeps over two boots and 16/16 systemd suspends on battery, and a 300 s sleep averaged 0.049 W (macOS on the same counter: 0.109 W). Merged suspend-to-idle gained USB and Bluetooth keyboard wake (linux #340); its standby power is still unmeasured.
Boot chain: aura replaced m1n1 as the Stage 1 and Stage 2 boot object (aura #16). With aura #77, linux #985, U-Boot #32 and aurora-installer #104, the Neo ran Linux at EL2 inside a 128 MiB CTXR window on November 7: 6 CPUs, 1,696 executable mappings and 0 violations.
Speakers: speakerguardd v2 qualified the J700 (speakerguardd #53, speakerguardd #104). With linux #1201 the Neo measured −0.06 dB(A) against macOS 26.6.2 at full volume and passed a 300 s full-volume protection ladder with a supervisor kill. The distribution pins of that release candidate are open.
GPU: G17P Mesa now lives on Mesa's default branch aurora-wip (mesa #29, mesa #87). On the Neo, native GL/GLES CTS went from 59 Fail to 0 over 188,060 names (mesa #30), and 1,650,061 applicable non-ray-tracing Vulkan cases gave 0 Fail (linux #423). linux #391 stopped advertising GPU soft faults, which bears on the M4 Pro fault in discussion #362.
Touch ID: retry limits, lockout and password re-arm now follow macOS (linux #904, linux #1134). A November 8 finger test on the Neo passed and completed an enrolment, on a development kernel. A fresh Fedora install found enrolment blocked by SELinux; fedora-aurora #125 fixes that, and its hardware acceptance is pending.
Thermal and installs: the CPU zones now control at 105 °C and 4.2 W (linux #1043), +8–9% sustained. GPU cooling is held in open linux #1065 after a trial with unvalidated 98 °C critical trips powered the Neo off. A fresh NixOS KDE install reached KDE (nixos-aurora #2); no public release install is qualified.
How to read this tracker
Marker
Meaning
🟢 Reported on hardware
A linked report identifies J700 and the tested feature. Its exact revisions and limits still apply.
🔵 Published code
Public implementation exists; this alone does not establish hardware operation.
🟠 Partial
Integration, qualification or functionality remains incomplete.
⚪ Not established
No qualifying J700 result was found in this audit; this does not mean unsupported.
🔴 Known gap
A public report identifies missing or failing functionality.
Results on another board, a private integration, an older implementation or an open branch do not automatically qualify the public J700 stack. “Merged” below refers to Aurora's repository, not mainline Linux or a distribution release.
Carries the J700 series merged through November 8; linux #1201 is in integration/aurora only. aurora-stable does not carry this work. The shared asahi.config builds the Neo drivers; the aurora_j700_defconfig from #306 is no longer in either tip. Deep sleep (linux #1208) is open.
Boot objects (aura, m1n1)
aura main has been the shipping Stage 1 and Stage 2 since aura #16, which carries m1n1 aurora-wip through its October 28 consolidation (m1n1 tip b6b459221ede). aurora-silicon/aura#29 added the Mach-O boot object and aura #77 Linux at EL2; the distributions pin aura 1.0.0b7
Its final Stage 2 (cda450d) booted the Neo on November 7, once as a one-shot trial. The PR lists idle power and restart or power-off without PSCI as unmeasured. Deep sleep needs open aura #117 and aura #112.
U-Boot J700
aurora-wipe1ef8b636fc9 carries U-Boot #1, the NVMe write-cache flush, U-Boot #8 (NVMe wedge during GRUB's disk search, HiDPI console font) and U-Boot #32 (CTXR window)
The wedge's cause is inferred, not proven on hardware. With the CTXR change, EFI runtime stays on, but GetTime is unsupported and efivarfs mounts read-only. A cold MTP handover to Linux (aurora-silicon/u-boot#37) is open.
Mesa
aurora-wip8ac88a263afc, now Mesa's default branch, carries G17P since mesa #29 and mesa #87; the agx3b to main PR closed
Distributions pin aurora-wip revisions; each CTS result below names its own kernel and Mesa pair.
Local firmware, unit calibration and device identity are required. They are not supplied by the kernel PRs; do not substitute another machine's inputs.
The 128 MiB padded Stage 1 had not run through the app when it merged. Alpha 17 failed at the Neo's Recovery handoff until aurora-installer #95. Omarchy's Neo DMG sealing (omarchy-mac #4) is still open.
Hardware status
🧩 Core platform
Hardware / SoC block
Public implementation
Hardware evidence / limits
Sources
Boot and kernel handoff
🔵 aura Stage 1 and 2, U-Boot and the EL2 CTXR path merged; #301 leaves the firmware-owned disp_cpu power state untouched
🟢 November 7, padded aura cda450d Stage 1, U-Boot 50d879ac and #985f6b1cbf: 6 CPUs at EL2, boot verify 1,696 mappings with 0 violations, 1,200 module load cycles, test_bpf 1061/1061 per CPU; KVM and s2idle passed. aurora-silicon/aura#29 earlier booted an installed Fedora 7.3.0-5 kernel to a login at EL1 (3 boots). 🟠 74 MiB of the window is already taken on the Neo's U-Boot and GRUB path (17 MiB on the M5); hotplug and kexec are refused under a window
🟢 #336 v1: 12/12 × 300 s s2idle with Wi-Fi and Bluetooth back every time and 0 DART lines; v2 smoke 3/3. Earlier: display hand-off fault absent in 5/5 boots (#221); 1,785 freed audio buffers intact over 1,572 cycles (#232); 54,392 capture cycles with 0 TLB timeouts (linux #237)
🔵 Driver and handoff merged; U-Boot flushes its write cache and no longer wedges in GRUB's disk search
🟢 Internal-storage boot; NVMe live after s2idle (#222); #320: 5/5 RTC s2idle cycles on each of 4 release kernels, 363–407 fsynced MiB intact per cycle. #1208 (open) stops and restarts ANS in every deep sleep, 1.4 GB/s after wake. 🟠 Inherited live ANS session; reset under I/O unqualified
🔵 WFI only, by design (linux #216); #298 adds an s2idle-eligible WFI entry; CPU power-off in system sleep only in open #1208
🟢 #216: plain WFI power-gates the E cores and second P core in 94–99.9% of samples; the CYC_OVRD power-down write is undefined on T8140. #298: 3/3 RTC sleeps accounted 19.8–20.4 s of 20 s. #1208 (open) powers CPUs off only inside a full deep sleep
🔵 One apple,t8140-pmu across both clusters (linux #865)
🟢 #865, one-shot Stage 2 trial: one PMU over CPUs 0–5 with 8 counters, and perf stat counts matched the expected per-iteration numbers on P and E cores. 🟠 #731 (bounded nested stage-2 TLB flush) is open
🟢 GNOME 2408×1506@60, 2–525 nits, 20/20 DPMS; the fade settles in 264–279 ms. #902 (7.3.0-aurora-dcpcc1): both DCPs start together, [drm] Initialized apple 7.782 → 7.589 s, and an hour passed with no DCP crash. #1016 gives GNOME a panel identity so brightness can be saved. 🟠 Recovery from a crashed DCP is only in open #1208
🔵 G17 driver in both branches with shared capacity, ray-query allowances and timeout recovery (linux #377), admission (#423, #1146) and fault retirement (#496); G17P Mesa on aurora-wip
🟢 #423 (module 18b408b7, aurora-silicon/mesa#62, Linux 7.3): Vulkan CTS 681,296 pass, 0 fail, 0 crash in 1,650,061 non-RT cases; ray query 33,542 pass, 0 fail; ray-tracing pipeline 13,558 pass, 0 fail, 2 crashes. mesa #30: native GL/GLES 188,060 names, 59 Fail → 0. Chromium WebGL at 60 FPS through zink, Gallium and ANGLE Vulkan (aurora-silicon/mesa#40). 🟠 Under sustained ray-tracing stress 10–13 queues still stuck at peak after #496; past the 58-context queue pool new clients lose their device until open mesa #184; #362's fragment-end read is unchecked on a J700; no game or Steam qualification
🔵 Rear-port H17P path, unconditional since linux #293; fast-replug recovery (linux #335); macOS's HDR brightness hints (linux #796)
🟢 #293 R69: Acer QG270 S3 at 1920×1080 180 Hz SDR, both cable orientations, clamshell. #335 (v5 kernel): 14/14 fast replugs, 5 slow, cold plug, 2 DPMS and 2 s2idle with the dock lit. #902: a hand replug re-detected 34 modes and s2idle 1/1 brought both displays back. #796: four PQ and SDR mode sets on an MSI MPG 325CQRXF accepted. 🟠 The dock often drops Type-C about 50 ms after the first 180 Hz mode set (about 5 s per plug); two sinks untested; HDR content, DSC, VRR and DP audio unqualified (#783 reverted DP audio by default). 🔴 Front port is USB 2.0 only; linux #334 only notifies the user
🟢 speakerguardd #53 (7.3.0-aurora-sgv2k5, SELinux enforcing): volume 0.3, 0.6 and 1.0 for 60, 60 and 300 s with 0 health faults; coil peak 92.6 °C, gain cut at most −4.45 dB; floor 22.7–29.9 ms after SIGSTOP. speakerguardd #104 RC6 with #1201: −0.06 dB(A) against macOS at full volume (3 microphones, spread 0.10, −5.7 dB before #1201); 300 s full-volume ladder with a supervisor kill, 0 amplifier faults, 0 dropouts. asahi-audio #7 (merged) had measured 0 clipped samples. 🟠 The RC6b distribution pins merged on Nov 8 (fedora-aurora #189, aurora-silicon/alarm-aurora#167, aurora-silicon/nixos-aurora#171); the RC7 pins are open; headphone jack: start volume only
🔵 AOP audio, run-state counting (linux #239), UCM on the AOP card and lid-closed hiding (#354, #537)
🟢 #237: 100/100 high-quality captures carry fresh data. #239: 85,313 capture cycles in 31 min with 0 EIO. #354: 4 attended lid cycles (3 into s2idle), the mic jack off within 0.4–1.4 s of closing and the PipeWire source restored each time
🔵 T8140 capture series plus #220, #229 and #319; frame gain and exposure (#689); macOS temporal-filter and mains settings (#774); camera-denoise filter
🟢 1080p NV12 at 30.01 fps, 30 fps at 0 lux; capture across s2idle (#229, #319). #774 (fleet14d): 6/6 fresh-boot first streams of 150 frames. aurora-silicon/camera-denoise#4: allowing 16.9–19.7 fps in 3–6 lux raised SNR by 1.9–3.5 dB. 🟠 One first-stream freeze after 114 frames (fleet14c) is unexplained; firmware starts once per boot; about 30–50 mW extra idle draw
🟢 34 runs / 1,740 exact frames in the qualification below. #758: 32 avd-bench rows (H.264, HEVC, VP9, AV1 at 1080p and 4K, 1–4 streams) and 8 FFmpeg clips bit-exact; HEVC through FFmpeg 286 fps at 1080p and 88 at 4K. #1116: bit-exact, H.264 1080p about 740 fps. 🟠 #725 recorded a Neo hang under the earlier module after 40 decoder error and timeout pairs
🟢 #615 v15: Start and Process commands byte-identical to macOS; 300/300-frame H.264 and HEVC encodes up to 2408×1506; sleep mid-encode 1500/1500 frames; v4l2-compliance 48/48. linux #853: 98–99% of macOS. #835: 422 Proxy to 4444 XQ at 1080p and 2160p. #848: JPEG 4:2:0 1080p 816 fps (macOS 556). 🟠 HEVC at 4096×2304 is refused
🟢 #801 (fleet14g): relu and add exact over 200 runs, MobileNetV1 top-1 16/16 at 2.82 ms p50, 0 DART faults. aurora-silicon/aura#52: 50 load, run and close cycles with nothing refused. #1137: MobileNetV1 2.3–2.5 → 1.25–1.31 ms. 🟠 No application framework or model set beyond these tests is qualified
🔵 Port, PHY and DART support; fixed-hub controller kept across sleep (#236); SuperSpeed kept through a PD re-election (#353) and retrained after a boot-time PIPE failure (#654); macOS repeater and VL122 tuning (#1041); liquid detection (#1053)
🟢 Earlier connector, SSD and PD results stand; #236: device resume 2822 → 293–297 ms. #1041, as modules: tuning applied and read back, kept after a pm_test resume; the DT values await open aurora-silicon/aura#91. #1053: both ports report contaminant = none (older head). 🟠 USB3 at both connectors unqualified; a dock's second hub never enumerates on the USB 2.0 front port
🔵 Automatic one-shot bootstrap, now DT-selected (#641, #649), retained across s2idle
🟢 #204: both functions enumerate in about 1.4 s; the Wi-Fi function stays powered through s2idle (#222). #1208 (open) takes the link through L2 and retrains it after deep sleep. Removal, kexec and hibernation unsupported
🔵 Fullmac with WPA3-SAE, OWE and 6 GHz (linux #375), a world fallback (#376), PMF (linux #784) and directed scans (#802)
🟢 #375: SAE with required MFP and PSK on 2.4 and 5 GHz; hidden 6 GHz SAE and OWE networks join, OWE at about 400 Mbit/s down. #376: CA, world fallback, then AU with 24 6 GHz channels, no reload. #784: every tested WPA2, WPA3, PMF and 6 GHz row passes after the fix, and roaming between two PMF-required SAE access points passed both ways with 0% loss on the #783 candidate. #336: back after 12/12 × 300 s s2idle. Earlier: about 183/187 Mbit/s on channel 149, 3589/3595 idle pings in an hour (#266). 🟠 DFS joins, 80/160 MHz, AP and module reload unqualified; iwd has no OWE offload; #646's power policy has no hardware run
🔵 Modular transport, controller reset and suspend fixes; keyboard wake armed by default (linux #340)
🟢 Power and scan (17–34 devices); 10 s2idle cycles with 0 errors (#230); 3/3 forced resets recovered (#262). #340: KB400 key wake through the MT7932 MSI, lid open and docked clamshell, on an identical earlier head. 🔴 From #1032 (November 6) until linux #1055 (November 7), integration/aurora built no mt7932_bt_pcie driver; both tips carry the rule now. 🟠 Pairing, audio profiles and reconnect after resume not re-tested
🔵 J700 input merged; wake sources (#235); DockChannel sleep notices (#651); a libinput trackpad quirk in aurora-neo-platform-config
🟢 Input operator-confirmed (#33); Shift, click and touch each woke s2idle, and a 90 s lid close stayed asleep (#235). 🟠 Linux adopts U-Boot's running MTP session; a cold handover is open in U-Boot
🔵 Integrated; GPIO domain kept on (#234); sensor bring-up retried before a verify (#339); xART guard left off on the Neo (#361)
🟢 #42: boot, RNG, trusted keys. #234: hwrng, trusted keys and a Touch ID verify start after each of 10 × 70 s s2idle cycles. #339: 5/5 cycles of a 15 s s2idle then fprintd-verify, re-registration in 56–63 ms, the same 3 fingers listed. aurora-silicon/aurora-sep-userspace#10: LUKS2 unlock with an enclave-sealed key on the Neo (loop file, fleet24)
🔵 Authenticated sessions, post-registration recheck (linux #303), boot-registration recovery (linux #325), lockout and password re-arm (linux #904), macOS retry limits (linux #1134, #1143) and captures armed on lift (#1087)
🟢 #303: 3 enrolled prints kept through more than 100 s2idle cycles. #325: 104/104 cold boots registered. #904 (fleet20): an enrolled finger matched after three misses and a password re-arm, twice. #1134, 2026-11-08 on 7.3.0-aurora-fleet22-tid27 with libfprint aurora10: finger test passed with 7 re-arm events, lockout at 5 misses and a completed enrolment. 🟠 Enrolment on a fresh public install waits on fedora-aurora #125's acceptance; lid-closed hiding merged in aurora-silicon/aurora-sep-userspace#1; aurora-silicon/aura#29's Neo has no sensor
🔵 Merged; SMC fan and charge policy left to firmware, thermal pressure exposed (#670); sensors the J700 lacks dropped (#631)
🟢 RTC read/write and reboot persistence; rtc0 is SMC and rtc1 the PMU wake alarm (#222); 84 temperature channels in the release check. Some labels remain guesses
🔵 AOP/CT817 series merged; per-unit calibration required
🟠 Compile-tested only; calibration upload and light-response qualification remain outstanding. The catalogue lists it as working without a cited J700 report
🔵 Battery driver merged; UPower keeps the kernel's time estimates
🟢 Full/100% readings; the UPower change ran 70 s on a full battery. #1208 read 11.3 mA on battery from the SMC charge counter. Charging transitions and positive time estimates still need a board report
🔵 IPA and PMP limits; CPU zones control at 105 °C with 4.2 W sustainable power (linux #1043); GPU cooling in open linux #1065
🟢 #1043 (fleet16, AC, all six CPUs, zone values set through sysfs): int64 +9.4% over 300 s; a 900 s soak averaged 103.7 °C and peaked at 106.5 °C, with no sample at the 110 °C critical trip. #219: the DCS limit held a GL/Vulkan load at 97–98 °C. 🔴 #1065's first trial put 98 °C critical trips on unvalidated SMC zones and powered the Neo off about 7 s into boot; a log-only trial saw hotspot keys at 108–113 °C for 1–3 s in the boot burst. 🟠 #1043 reports no boot of its own DTB; battery-only and long GPU loads open
🔵 s2idle (linux #222) with wake from RTC, power button, lid, keyboard, trackpad and USB or Bluetooth keyboards (linux #340); deep sleep in open linux #1208
🟢 See the sleep table below. 🟠 s2idle standby power unmeasured; deep sleep is opt-in (mem_sleep_default=deep) and none of the distribution trees sets it; hibernation refused
🟢 nixos-aurora #2: a fresh NixOS KDE install (Stage 1 preserved) finished encryption in 2m17s and opened KDE; its passphrase reboot and Wi-Fi join were not repeated. A fresh Aurora Fedora install on the Neo on November 2 showed the Touch ID SELinux block (aurora-silicon/aurora-installer#81 names a developer installer build that night). 🟠 Alpha 17's app failed at the Neo's Recovery handoff until aurora-installer #95; no release qualification or repeated public-installer run is reported
Test kernels with this series plus debug code; the later ones also carried two MTP commits not in it
30/30 over 2 boots; 16/16 on battery; 12/12 with the USB changes; 8/8 after review fixes
Wakes from RTC, power button, keyboard, trackpad, USB and Bluetooth keyboards, lid and a Wi-Fi magic packet, still associated. Entry 0.25 s and wake 1.02 s median. 300 s sleep 0.049 W (macOS 0.109 W). Needs the two open aura PRs, apple_sep loaded and deep selected
eDP + 1080p180 back with 0 DCP errors; Touch ID, camera, audio and NVMe checks pass
Suspend-to-idle standby power was never measured; NVMe, GPU, Wi-Fi and ISP domains stay powered (the ISP alone adds an estimated 30–50 mW). The only standby figure is #1208's deep-sleep 0.049 W.
Limits: only 1920×1080 at 60, 120 and 180 Hz is reported; the MSI run in drm/apple: send the external HDR brightness hints macOS sends (H17P) #796 does not state its mode. Two sinks and plugging in during a modeset are untested; VRR is refused. The catalogue's "up to 4K 60 Hz" has no J700 report behind it in this audit.
Physical J700/T8140, kernel 7.1.12-aurora-neo-existing-wifi-20261001-r2+, matching AVD module and added AVD/DART/power-domain wiring, with a temporary RAM overlay. Firmware was replaced and the idle driver normally unloaded/reloaded; this was not a new whole-system cold-boot qualification.
Test
Hardware runs
Exact hardware/software frame matches
H.264 360p/1080p, normal download and vaGetImage
4
120
HEVC, VP9 and AV1 360p, 8/10-bit, both download paths
12
360
H.264/HEVC/VP9 720p, repeated full decode and 1s/3s seeks
18
1,260
Total
34
1,740
Every run required hardware/backend markers and byte comparison against software decoding. Both guarded groups finished idle without timeout, wedge, decoder error or kernel fault. Other SoCs, concurrent clients, suspend/resume and application-level playback are outside this qualification; #758 and #1116 later add bit-exact multi-stream and 4K runs on 7.3 kernels.
Evidence archive retained locally, not public: neo-avd-aurora-evidence.tar.gz, SHA-256 dbd939da3b20e8c06d85d6e12538cadd559a15a18d665c287597678d672faa04. Public results and attribution are in #154.
The decoder foundation and #45 fixes credit sofus, Aaron (aquarat), and iconidentify. Neo bring-up credits DJ, Ryan Murray and Ace. Original author/co-author history and license notices are retained; Aurora firmware includes DJ's changes and preserves the inherited MIT notices.
Radio qualification and remaining work
The tested Wi-Fi path is fullmac, not the proposed mt76/softmac path. Local firmware and the unit's own calibration and identity were used; the PRs carry no blobs, credentials or unit data.
Since October 16, WPA3-SAE (including H2E), OWE, PMF on PSK and PSK-SHA256, 6 GHz with VLP and LPI country data, directed scans for hidden networks and a world regulatory fallback were added and tested on the Neo (#375, #376, #659, #784, #802). The earlier coexistence run (5 GHz 54/54 and 2.4 GHz 53/54 gateway replies with Bluetooth cycling, 18/18 HTTPS each) and the AirPods/AAC result are historical, not reruns of the current head. Roaming between two PMF-required SAE access points passed both ways on the #783 candidate kernel. Removal, kexec, hibernation, DFS joins, AP/P2P, Bluetooth pairing, SCO, LE Audio and long-duration stability are not qualified.
Audit scope and public references
This refresh covers public PR states, bodies and comments, current branches and relevant source. Earlier hardware results keep their original revisions.
Catalogue support data: it marks almost every J700 feature as working, including ambient light and a 4K external display, with no per-feature report cited; this tracker follows the PRs instead.
Reply with one feature or failure per report, tied to exact board and software revisions.
Board: J700
SoC: T8140
Model / configuration: MacBook Neo (with or without Touch ID)
Feature or SoC block:
Kernel repository / branch / commit:
aura / U-Boot / Mesa commits, where relevant:
Firmware source / revision / hash:
Steps to reproduce:
Expected result:
Observed result and repeat count:
Public log or artifact URL:
Artifact SHA-256 and command to repeat the check:
Known limits / recovery steps:
Keep credentials, per-unit calibration and device identities out of public reports. A maintainer can update the table when the linked evidence establishes the result for this board.
One hardware thread for MacBook Neo in the A18 family. Keep reports tied to J700/T8140 and exact software revisions.
Current source cross-check: in integration/aurora4fa3cc72d7d1, asahi.config builds MT7932 Wi-Fi and Bluetooth, AVE, ProRes, JPEG, the ANE and the CTXR execution window; the J700 DT enables AVE, JPEG, the ANE, the SEP and the Stockholm NFC node, and all six CPU zones control at 105 °C. drivers/bluetooth/Makefile builds mt7932_bt_pcie in both tips. aurora_j700_defconfig is in neither tip. Original merge ancestry is recorded separately below.
Audit baseline: linux integration/aurora4fa3cc72d7d1 and aurora-wip2991c10caab6; public source and PR reports checked November 8, 2026. No new physical test was performed for this update. Working-tree integration does not imply an aurora-stable release.
PR state, source integration and hardware qualification are separate. A closed PR is not necessarily merged. Public reports describe their original test kernels, including contributor branches. Repository administrators can edit this opening post.
hardwareBoard-specific hardware evidence and SoC indexessoc:T8140Hardware discussions for T8140 (A18 Pro)
1 participant
Heading
Bold
Italic
Quote
Code
Link
Numbered list
Unordered list
Task list
Attach files
Mention
Reference
Menu
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
MacBook Neo · A18 family
🧭 Hardware directory › A18 category › J700 · T8140
T8140J700Important
J700 now has deep-sleep (suspend-to-RAM) reports at 0.049 W in an open series, speaker playback within 0.1 dB of macOS at full volume under speakerguardd v2, a working Neural Engine and hardware video encoders, and Linux running at EL2 through aura. Each result belongs to the development kernel its PR names. Deep sleep is not merged, suspend-to-idle standby power is unmeasured, the front port carries no display, and no public release install is qualified.
Refreshed November 8, 2026. Current public branches, PR states, follow-up reports and source were checked. No new hardware test was run for this update; the results below are those reported in the linked PRs, on the kernels they name. Source integration, individual subsystem tests and a complete laptop qualification are separate milestones.
What changed · November 8, 2026
aurora-wip(mesa #29, mesa #87). On the Neo, native GL/GLES CTS went from 59 Fail to 0 over 188,060 names (mesa #30), and 1,650,061 applicable non-ray-tracing Vulkan cases gave 0 Fail (linux #423). linux #391 stopped advertising GPU soft faults, which bears on the M4 Pro fault in discussion #362.How to read this tracker
Results on another board, a private integration, an older implementation or an open branch do not automatically qualify the public J700 stack. “Merged” below refers to Aurora's repository, not mainline Linux or a distribution release.
Source and integration status
linux/integration/aurora@4fa3cc72d7d1, now the default branch.aurora-wip(2991c10caab6) is cross-merged with it (#1032)integration/auroraonly.aurora-stabledoes not carry this work. The sharedasahi.configbuilds the Neo drivers; theaurora_j700_defconfigfrom #306 is no longer in either tip. Deep sleep (linux #1208) is open.mainhas been the shipping Stage 1 and Stage 2 since aura #16, which carries m1n1aurora-wipthrough its October 28 consolidation (m1n1 tipb6b459221ede). aurora-silicon/aura#29 added the Mach-O boot object and aura #77 Linux at EL2; the distributions pin aura 1.0.0b7cda450d) booted the Neo on November 7, once as a one-shot trial. The PR lists idle power and restart or power-off without PSCI as unmeasured. Deep sleep needs open aura #117 and aura #112.aurora-wipe1ef8b636fc9carries U-Boot #1, the NVMe write-cache flush, U-Boot #8 (NVMe wedge during GRUB's disk search, HiDPI console font) and U-Boot #32 (CTXR window)GetTimeis unsupported and efivarfs mounts read-only. A cold MTP handover to Linux (aurora-silicon/u-boot#37) is open.aurora-wip8ac88a263afc, now Mesa's default branch, carries G17P since mesa #29 and mesa #87; theagx3btomainPR closedaurora-wiprevisions; each CTS result below names its own kernel and Mesa pair.avd-fwat9cbde5c7d993avd-fw-v5-t0.binand install it separately. This is the firmware used for the Neo decode qualification.Hardware status
🧩 Core platform
disp_cpupower state untouchedcda450dStage 1, U-Boot50d879acand #985f6b1cbf: 6 CPUs at EL2, boot verify 1,696 mappings with 0 violations, 1,200 module load cycles,test_bpf1061/1061 per CPU; KVM and s2idle passed. aurora-silicon/aura#29 earlier booted an installed Fedora 7.3.0-5 kernel to a login at EL1 (3 boots). 🟠 74 MiB of the window is already taken on the Neo's U-Boot and GRUB path (17 MiB on the M5); hotplug and kexec are refused under a windowpolicy->curin stepCYC_OVRDpower-down write is undefined on T8140. #298: 3/3 RTC sleeps accounted 19.8–20.4 s of 20 s. #1208 (open) powers CPUs off only inside a full deep sleepapple,t8140-pmuacross both clusters (linux #865)🎨 Display and media
7.3.0-aurora-dcpcc1): both DCPs start together,[drm] Initialized apple7.782 → 7.589 s, and an hour passed with no DCP crash. #1016 gives GNOME a panel identity so brightness can be saved. 🟠 Recovery from a crashed DCP is only in open #1208aurora-wip18b408b7, aurora-silicon/mesa#62, Linux 7.3): Vulkan CTS 681,296 pass, 0 fail, 0 crash in 1,650,061 non-RT cases; ray query 33,542 pass, 0 fail; ray-tracing pipeline 13,558 pass, 0 fail, 2 crashes. mesa #30: native GL/GLES 188,060 names, 59 Fail → 0. Chromium WebGL at 60 FPS through zink, Gallium and ANGLE Vulkan (aurora-silicon/mesa#40). 🟠 Under sustained ray-tracing stress 10–13 queues still stuck at peak after #496; past the 58-context queue pool new clients lose their device until open mesa #184; #362's fragment-end read is unchecked on a J700; no game or Steam qualificationv2branch) owns the J700 speaker path (aurora-silicon/speakerguardd#83; aurora-silicon/asahi-audio#22 retires its graph there); in-kernel protection ownership (#922); full-scale drive (linux #1201,integration/auroraonly)7.3.0-aurora-sgv2k5, SELinux enforcing): volume 0.3, 0.6 and 1.0 for 60, 60 and 300 s with 0 health faults; coil peak 92.6 °C, gain cut at most −4.45 dB; floor 22.7–29.9 ms after SIGSTOP. speakerguardd #104 RC6 with #1201: −0.06 dB(A) against macOS at full volume (3 microphones, spread 0.10, −5.7 dB before #1201); 300 s full-volume ladder with a supervisor kill, 0 amplifier faults, 0 dropouts. asahi-audio #7 (merged) had measured 0 clipped samples. 🟠 The RC6b distribution pins merged on Nov 8 (fedora-aurora #189, aurora-silicon/alarm-aurora#167, aurora-silicon/nixos-aurora#171); the RC7 pins are open; headphone jack: start volume only🔌 I/O and security
pm_testresume; the DT values await open aurora-silicon/aura#91. #1053: both ports reportcontaminant=none(older head). 🟠 USB3 at both connectors unqualified; a dock's second hub never enumerates on the USB 2.0 front portintegration/aurorabuilt nomt7932_bt_pciedriver; both tips carry the rule now. 🟠 Pairing, audio profiles and reconnect after resume not re-testedfprintd-verify, re-registration in 56–63 ms, the same 3 fingers listed. aurora-silicon/aurora-sep-userspace#10: LUKS2 unlock with an enclave-sealed key on the Neo (loop file, fleet24)7.3.0-aurora-fleet22-tid27with libfprint aurora10: finger test passed with 7 re-arm events, lockout at 5 misses and a completed enrolment. 🟠 Enrolment on a fresh public install waits on fedora-aurora #125's acceptance; lid-closed hiding merged in aurora-silicon/aurora-sep-userspace#1; aurora-silicon/aura#29's Neo has no sensor🔋 Power and system readiness
rtc0is SMC andrtc1the PMU wake alarm (#222); 84 temperature channels in the release check. Some labels remain guessesmem_sleep_default=deep) and none of the distribution trees sets it; hibernation refusedSleep reports
apple_seploaded anddeepselected7.1.12-aurora-wakefilter41awip-bt-gf22b055cef6b1cbfwith auracda450daurora-wipac895820e+ polish seriesSuspend-to-idle standby power was never measured; NVMe, GPU, Wi-Fi and ISP domains stay powered (the ISP alone adds an estimated 30–50 mW). The only standby figure is #1208's deep-sleep 0.049 W.
External display (DP Alt Mode)
displayportlink.Neo video-decode qualification
Physical J700/T8140, kernel
7.1.12-aurora-neo-existing-wifi-20261001-r2+, matching AVD module and added AVD/DART/power-domain wiring, with a temporary RAM overlay. Firmware was replaced and the idle driver normally unloaded/reloaded; this was not a new whole-system cold-boot qualification.vaGetImageEvery run required hardware/backend markers and byte comparison against software decoding. Both guarded groups finished idle without timeout, wedge, decoder error or kernel fault. Other SoCs, concurrent clients, suspend/resume and application-level playback are outside this qualification; #758 and #1116 later add bit-exact multi-stream and 4K runs on 7.3 kernels.
avd-fw9cbde5c7d993844bf086318974958b11076600a0. Installavd-fw-v5-t0.binat/lib/firmware/apple/avd-fw-v5-t0.bin; SHA-256d9f9e699dd63df51a3f3a44eee5141657d0a55d91d87039158e5edfe511ed63e.5b5046cbda689, with FFmpeg n9.0.1 and libva 2.24.neo-avd-aurora-evidence.tar.gz, SHA-256dbd939da3b20e8c06d85d6e12538cadd559a15a18d665c287597678d672faa04. Public results and attribution are in #154.The decoder foundation and #45 fixes credit sofus, Aaron (aquarat), and iconidentify. Neo bring-up credits DJ, Ryan Murray and Ace. Original author/co-author history and license notices are retained; Aurora firmware includes DJ's changes and preserves the inherited MIT notices.
Radio qualification and remaining work
The tested Wi-Fi path is fullmac, not the proposed mt76/softmac path. Local firmware and the unit's own calibration and identity were used; the PRs carry no blobs, credentials or unit data.
Since October 16, WPA3-SAE (including H2E), OWE, PMF on PSK and PSK-SHA256, 6 GHz with VLP and LPI country data, directed scans for hidden networks and a world regulatory fallback were added and tested on the Neo (#375, #376, #659, #784, #802). The earlier coexistence run (5 GHz 54/54 and 2.4 GHz 53/54 gateway replies with Bluetooth cycling, 18/18 HTTPS each) and the AirPods/AAC result are historical, not reruns of the current head. Roaming between two PMF-required SAE access points passed both ways on the #783 candidate kernel. Removal, kexec, hibernation, DFS joins, AP/P2P, Bluetooth pairing, SCO, LE Audio and long-duration stability are not qualified.
Audit scope and public references
This refresh covers public PR states, bodies and comments, current branches and relevant source. Earlier hardware results keep their original revisions.
🧪 Add a reproducible hardware report
Reply with one feature or failure per report, tied to exact board and software revisions.
Keep credentials, per-unit calibration and device identities out of public reports. A maintainer can update the table when the linked evidence establishes the result for this board.
One hardware thread for MacBook Neo in the A18 family. Keep reports tied to J700/T8140 and exact software revisions.
Current source cross-check: in
integration/aurora4fa3cc72d7d1,asahi.configbuilds MT7932 Wi-Fi and Bluetooth, AVE, ProRes, JPEG, the ANE and the CTXR execution window; the J700 DT enables AVE, JPEG, the ANE, the SEP and the Stockholm NFC node, and all six CPU zones control at 105 °C.drivers/bluetooth/Makefilebuildsmt7932_bt_pciein both tips.aurora_j700_defconfigis in neither tip. Original merge ancestry is recorded separately below.Audit baseline: linux
integration/aurora4fa3cc72d7d1 andaurora-wip2991c10caab6; public source and PR reports checked November 8, 2026. No new physical test was performed for this update. Working-tree integration does not imply anaurora-stablerelease.PR state and integration audit · November 8, 2026
mainmainbranch ·06aec67a8ac8mainmainbranch ·2acede7fe106mainmainbranch ·3616f9c8207fmain2224af0ab355main01429815d7d8mainmainbranch ·e17f7cd1b0edmainmainbranch ·f3ab139ed9b3mainmainbranch ·5e0408abb26emainmainbranch ·60c6165c9d8cmainmainbranch ·dd6a0ed240d9aurora-wip53d1304629a6aurora-wipintegration/aurorahistory andaurora-wip·4f3a64e7baa9aurora-wipintegration/aurorahistory andaurora-wip·d41b89bdd071aurora-wipintegration/aurorahistory andaurora-wip·3fecb279810daurora-wipintegration/aurorahistory andaurora-wip·18837942fa8daurora-wipintegration/aurorahistory andaurora-wip·c679c8a75a89aurora-wipintegration/aurorahistory andaurora-wip·259ee1d76704aurora-wipintegration/aurorahistory andaurora-wip·cf6198d365e9aurora-wipintegration/aurorahistory andaurora-wip·d588f760941eaurora-wipintegration/aurorahistory andaurora-wip·a490adab6fc3aurora-wipintegration/aurorahistory andaurora-wip·97075ac83279aurora-wipintegration/aurorahistory andaurora-wip·b133ffc174d0aurora-wipintegration/aurorahistory andaurora-wip·5ca66cef5a33aurora-wipintegration/aurorahistory andaurora-wip·d18a1305826aaurora-wipintegration/aurorahistory andaurora-wip·2fa9722cc154integration/auroraintegration/aurorahistory andaurora-wip·9ee605b6a8fbintegration/auroraintegration/aurorahistory andaurora-wip·6210b28b0036integration/auroraintegration/aurorahistory andaurora-wip·50a912fe81fcintegration/auroraintegration/aurorahistory andaurora-wip·9cbdd0734d9eaurora-wipintegration/aurorahistory andaurora-wip·95dee7ac031aaurora-wipintegration/aurorahistory andaurora-wip·11b76691030baurora-wipintegration/aurorahistory andaurora-wip·c869c99968b0aurora-wipintegration/aurorahistory andaurora-wip·52f0a6bee13baurora-wipintegration/aurorahistory andaurora-wip·d433ad1d6622aurora-wipintegration/aurorahistory andaurora-wip·27c8c52af421integration/auroraintegration/aurorahistory andaurora-wip·c1d69005617eintegration/auroraintegration/aurorahistory andaurora-wip·7399ba907087integration/auroraintegration/aurorahistory andaurora-wip·e5a1f1d71192integration/auroraintegration/aurorahistory andaurora-wip·bf8f10d4f9d7integration/auroraintegration/aurorahistory andaurora-wip·a8bbaa8e1f6eintegration/auroraintegration/aurorahistory andaurora-wip·f9ce365c1049integration/auroradf7f7f1ced3dintegration/auroraintegration/aurorahistory andaurora-wip·89f56150435bintegration/auroraintegration/aurorahistory andaurora-wip·1eac5facc8fdintegration/auroraintegration/aurorahistory ·d4d267839caaintegration/aurora8ad2161ad26baurora-wipaurora-wipbranch ·88e0e48e0d6faurora-wipaurora-wipbranch ·1d62a16e4f03aurora-wipaurora-wipbranch ·473e325b4f50aurora-wip2f09e08cda3emainmainbranch ·4aa66f5357d8feat/neo58c4d631b015v23d6dca349dfev28c1e08211975aurora-wipaurora-wiphistory and stablemain·15204343ccd6aurora-wipaurora-wiphistory and stablemain·38b39feacc5eaurora-wipaurora-wiphistory ·62ee47094877PR state, source integration and hardware qualification are separate. A closed PR is not necessarily merged. Public reports describe their original test kernels, including contributor branches. Repository administrators can edit this opening post.
All reactions