Skip to content

j514m (MacBook Pro 14-inch, M3 Max, 14 CPU cores): test results #333

Open

Description

@techgyani

Mac and installed build

  • Model (/proc/device-tree/model): Apple MacBook Pro (14-inch, M3 Max, 14 CPU cores, Nov 2023)
  • /proc/device-tree/compatible: apple,j514m, apple,t6034, apple,arm-platform
  • CPUs online: 0-13 (14)
  • Installed Aurora tag: unavailable. Installed with omarchy-mac-installer 2.3.9 from the edge channel on 2026-11-02;
    no release-tag file on the system.
  • uname -r: 7.2.11-14-1-ARCH
  • Packages (pacman -Q): linux-aurora 7.2.11.aurora14-1, m1n1-aurora 1.6.1.aurora22-1,
    mesa-aurora 1:26.3.0_devel.20261031.231483.a1969111475-1, omarchy-mac-boot 20261027-1, hyprland 0.57.0-4,
    aurora-audio 4.1.aurora3-3, speakersafetyd 2.0.1-1
  • Boot chain (/proc/device-tree/chosen): m1n1 stage 1 v1.6.1-dirty, stage 2 bc79acc4-dirty;
    iBoot1 mBoot-20457.41.3, iBoot2 iBoot-10151.140.19.700.2
  • /etc/m1n1.conf: does not exist. No asahi,t6031-* or other GPU switches in /chosen.
  • Distribution: Arch Linux ARM (Omarchy). Local changes: none.

Both firmware versions and provisioning

asahi,os-fw-version: 14.7
asahi,system-fw-version: unknown

Provisioning: not relevant to these results.

Actions and results

Feature / action Result Details
Boot to desktop pass Hyprland on simpledrm (card0), internal panel 3024x1890@60
Wi-Fi pass wlan0 associated and has an IPv4 address
Keyboard, trackpad detected Apple MTP keyboard, Apple MTP multi-touch in /proc/bus/input/devices; not separately tested
Battery pass macsmc-battery: Charging, 96 %
Speakers pass 5 numbered pw-play tones to audio_effect.j514-convolver (stream target "MacBook Speakers", /proc/asound/card1/pcm1p/sub0/status RUNNING before and during each, all exited 0), 3-20 s apart: tones 2-5 heard. Tone 1 was not heard; the sink read 0.70 [MUTED] before the test and 0.80 unmuted after it, and the owner reports it was muted at the start
Built-in microphones pass pw-record from omarchy_asahi_mic, 6 s, s16 48 kHz mono: volumedetect mean -29.3 dB, max -6.0 dB; RMS per 0.5 s between -45.4 and -22.9 dB; the owner heard it played back
Bluetooth pass owner-reported; Magic Keyboard and Magic Trackpad connected (bluetoothctl devices Connected)
Keyboard backlight pass brightnessctl -d white:kbd_backlight set 100% -> 255/255, the owner saw it light up, restored to 0
Internal display brightness fail no device under /sys/class/backlight (simpledrm)
Camera capture pass FaceTime HD Camera (apple-isp, /dev/video1): 30 frames NV12 1920x1080 via ffmpeg v4l2, analysed with signalstats only (no image kept). Frame 0 YAVG 28.5; frames 5-29 YAVG 157.7-160.1, Y range 10-233, YDIF 2.1-2.4
Charging over USB-C PD pass with both monitors connected (typec ports 1 and 3), macsmc-battery read Charging, 96 %; after unplugging them, Discharging, power_now -10.768 W.
Suspend/resume (s2idle) fail systemctl suspend: PM: suspend entry (s2idle) 21:51:52, PM: suspend exit 21:52:26, suspend_stats success=1. After resume the OS kept running (Wi-Fi CONNECTED_GLOBAL, audio PCM running, Magic Keyboard reconnected; Magic Trackpad did not), but the internal panel stayed dark; no drm/backlight kernel line after resume. Session lock service started on resume; the lock screen could not be seen. Recovered only by a forced shutdown (macsmc-input: Triggering forced shutdown!). 1 of 1 attempts
Headphone jack pass ALSA Headphone Jack = on (Headset Mic Jack = off); the default sink switched to alsa_output.platform-sound.HiFi__Headphones__sink; pw-play of complete.oga twice, heard by the owner
GPU fail DT node gpu@40a400000: compatible = "apple,agx-t6034", status = "disabled"; apple,firmware-version, -abi and -compat all zero. No kernel line matching asahi, agx, g15 or dcp this boot
External displays fail Two monitors connected over USB-C showed nothing. While connected: two typec partners (ports 1 and 3) with ID header 0x5440043e, no DisplayPort alt mode registered, and DRM had only card0-Unknown-1. One was an LG 27UP850N on a USB-C to USB-C cable (with power delivery; its USB hub also carried an external SSD). The second was an identical LG 27UP850N, also on USB-C. Ports, as the owner described them: one monitor in the left USB-C port next to MagSafe (connected at boot), one in the right USB-C port (hotplugged); an external SSD in the other left port. Linux: LG partners on typec ports 1 and 3, the SSD on port 2. Coldplug (both connected at boot): fail. Hotplug (unplugged, then plugged back in): fail; afterwards one typec partner with ID header 0x5440043e (port 1), still only card0-Unknown-1, and macsmc-battery Charging. No DCP node exists to drive any output: the live device tree has no compatible string containing dcp, no apple DCP/DRM module is loaded, and the source T6034 trees (t6034.dtsi, t6034-j514m.dts, aurora-silicon/linux f1534999) have no DCP node (t6031* has 14 mentions). HDMI not tested
Hyprland CPU use (idle desktop) fail top -b -n 2 -d 3: Hyprland 413.6 %CPU in the second sample; see #190

For failures: frequency is every boot so far (one boot). Last known-good / first bad: none, first install.

Graphics and session details

  • Mesa: mesa-aurora 1:26.3.0_devel.20261031.231483.a1969111475-1
  • Active renderer: not determined; glxinfo and eglinfo are not installed.
  • GPU profile / experimental opt-in: none.
  • Compositor: Hyprland 0.57.0 built from branch v0.57.0 at commit 0aa6745f705f2c4b0a5004eb019ef76e5bbdb426 clean
  • Suspend: tested once, see Actions and results (panel stays dark after resume). Crashes: none observed.

Relevant logs and related reports

Kernel (journalctl -k -b), the only DRM lines:

[    0.421605] kernel: [drm] Initialized simpledrm 1.0.0 for 108d2f82880.framebuffer on minor 0
[    0.421738] kernel: simple-framebuffer 108d2f82880.framebuffer: [drm] fb0: simpledrmdrmfb frame buffer device

Hyprland log:

ERR from aquamarine ]: CDRMRenderer(drm): Can't create renderer, no matching devices found
ERR from aquamarine ]: drm: onReady: no renderer for gl formats
ERR from aquamarine ]: drm: initMgpu: no renderer
ERR from aquamarine ]: drm: Failed to initialize renderer state for /dev/dri/card0

GPU node as m1n1 hands it over (/proc/device-tree/soc/gpu@40a400000):

compatible = apple,agx-t6034
status = disabled
apple,firmware-version = 000000000000000000000000
apple,firmware-abi = 000000000000000000000000
apple,firmware-compat = 000000000000000000000000
reg-names = asc sgx
reg = asc 0x40a400000 size 0x6c000, sgx 0x408000000 size 0x3fdc000

The macOS ADT sgx node on this Mac (read from macOS 27.2) has compatible = "gpu,t6034",
gpu-num-perf-states = 13, and IODeviceMemory 0x408000000 (length 0x3fdc000) and 0x408d00000
(length 0x16c000).

Further values read from macOS 27.2 (26B123) on 2026-11-03:

  • chip-id: <34600000> (little-endian 0x6034), from
    ioreg -p IODeviceTree -n chosen -r -l -d1 -w0 | grep -E '"chip-id" = '
  • gpu-core-count = 30, from
    ioreg -r -c AGXAccelerator -d 1 -w0 | grep -oE '"gpu-core-count" ?= ?[0-9]+'
  • GPUConfigurationVariable: num_cores=40, num_mgpus=4, num_gps=16, num_frags=40, gpu_gen=15,
    gpu_var="C", usc_gen=3, core_mask_list=(509,509,493,493), from
    ioreg -r -c AGXAccelerator -d 1 -w0 | grep -oE '"GPUConfigurationVariable" = \{[^}]*\}'. As binary, the four
    masks are 0x1fd, 0x1fd, 0x1ed, 0x1ed, with 8, 8, 7 and 7 bits set (30 in all).
  • mcc regions of 0x2000000 each: 0x220000000, 0x222000000, 0x224000000, 0x226000000, 0x228000000,
    0x22a000000 (six), from ioreg -p IODeviceTree -n 'mcc@903C0000' -r -l -d1 -w0 (IODeviceMemory)
  • system_profiler SPDisplaysDataType: Chipset Model: Apple M3 Max, Total Number of Cores: 30

Related:

Untested features and limitations

  • Not tested, with reasons:
    • Touch ID: this kernel has no apple,sep node in the device tree and no SEP driver bound; fprintd and libfprint
      are not installed.
    • HDMI: no HDMI node in t6034.dtsi or t6034-j514m.dts (aurora-silicon/linux f1534999; the HDMI include
      t603x-j514s-j516s-hdmi.dtsi is used by the T6030 boards only), and no DCP node (see External displays).
    • SD card: the owner chose not to test.
    • Headset microphone: the headphones used were AirPods Max on a Lightning to 3.5 mm cable, and the jack reported
      Headset Mic Jack = off.
    • Thunderbolt or USB-C devices other than the two monitors and an external USB SSD: none tested.
  • One boot only.
  • GPU values not measured on this Mac: the identity registers (id-version, id-counts-1, id-counts-2,
    core-mask-0, core-mask-1) and the loaded GFX firmware's TEXT digest. No m1n1 path reads them on T6034, and I
    have not tried raw register reads. macOS's own view of the core mask, core_mask_list=(509,509,493,493), is
    listed above.

Questions for the M3 Max workstream

This Mac is available for read-only measurements, with macOS kept on the same disk for recovery. A MacBook Air M4
(macOS) is available as a proxy host.

  1. Is there, or would you accept, an m1n1 path that runs only the powered GPU identity read on a T6034 (power on,
    read, release, publish in /chosen, GPU node left disabled), like kboot_gpu: publish the T8122 powered identity on the 25G83 path m1n1#20 for T8122? I would boot it
    once and post the values here.
  2. Alternatively, for the m1n1 proxy from the Air: which J514M USB-C port does the proxy use, and which
    proxyclient script reads the identity registers safely?
  3. For the J516C, malik-na took InitData from the macOS driver under the m1n1 hypervisor
    (t6031/initdata-dump). Are those steps written down anywhere I can follow on this J514M?

Activity

  1. techgyani commented on Nov 8, 2026

    @techgyani
    Author

    Retest with the unreleased J514M display and GPU work: native display and GPU rendering work on this Mac. Brightness control does not.

    Machine and build

    Actions and results

    Action Result
    Boot with the new kernel and Mesa, old m1n1 pass: no change, as expected (not enabling: the boot loader did not set apple,t6034-handoff)
    Cold boot (full shutdown) with m1n1 from #32 pass
    Native internal display pass: Hyprland eDP-1 3024x1964@120.00000 (was Unknown-1 3024x1890@60 on simpledrm)
    120 Hz presentation fail on the build above (60.000 Hz); pass with .swap_timestamps added for the J514M (see below)
    GPU rendering pass: glxinfo -B renderer Apple M3 Max (G15C B1), Accelerated: yes, OpenGL 4.6 / ES 3.2
    Hyprland CPU at idle pass: 10 % over 20 s (was 190-731 % on llvmpipe, see #190); no initMgpu lines
    GPU load, glmark2 terrain 60 s windowed and 60 s off-screen pass: no GPU or DCP errors in the kernel log
    Brightness keys fail: no effect; /sys/class/backlight is empty
    Vulkan not tested: vulkaninfo not installed
    External displays, suspend, audio, camera, Bluetooth, Wi-Fi not retested on this build
    chosen.asahi,m3-defaults=0 and the other owner switches not tested

    One cold boot with this build so far.

    Graphics and session details

    Kernel log at boot, trimmed:

    apple-t6034-display: handoff: apple,t6034-handoff dcp 0x1 display 0x1 piodma 0x1; dcp firmware-compat 14.7.0 version 14.7.0
    apple-t6034-display: native display on by default on J514M; apple_t6034_display.enable=0 keeps the boot framebuffer
    apple-drm display-subsystem: bound 386c00000.dcp (ops dcp_comp_ops [appledrm])
    asahi-g15 40a400000.gpu: GPU firmware build: J514M RTKit 2419.140.12 (runtime qualification pending)
    asahi-g15 40a400000.gpu: M3: firmware identity accepted
    asahi-g15 40a400000.gpu: T6034 G15C: power acknowledged, version=0x7041100 counts=0x21040a core-mask=0x7b5ed7f5fd
    asahi-g15 40a400000.gpu: T6034 G15C: cores enabled per cluster: [8, 8, 7, 7]
    asahi-g15 40a400000.gpu: G15: firmware at READY, starting the backend
    

    The per-cluster cores match macOS on this Mac (core_mask_list=(509,509,493,493), quoted in the report above).

    DCP firmware syslog lines also seen (the display ran at 120 Hz regardless):

    IOMFB: Could not find Backlight service
    IOMFB updateFrequencies EDT ERROR: getClockFrequency(0) (400000000) < videoClock 402278400! Giving up on frequencies.
    IOMFB: videoClock is zero, defaulting to 402280000
    

    Refresh rate (edited 2026-11-08)

    The mode is 120 Hz everywhere it is reported (hyprctl monitors, the DRM mode: 804560 kHz / (3104 x 2160) = 120.0005 Hz, apple-dcp: J514M display started: 3024x1964@120, and wp_presentation feedback refresh=8333333 ns), but frames are presented at 60 Hz:

    wp_presentation timestamps, 1200 frames: mean 60.000 Hz, interval median 16.667 ms (min 16.603, max 16.736)
    es2gears (Wayland): 60.000 FPS; glxgears: 60.0 FPS
    600 swaps submitted at random times: all 600 completions on one phase of a 16.667 ms grid, none on the 8.333 ms slots between
    

    In drivers/gpu/drm/apple/iomfb_v14_7.c at 8eddf5aeb6ac, dcp_v14_board_j514m has .promotion = true but not .swap_timestamps, which dcp_v14_board_j516c sets. The field's comment in iomfb_v14_7_board.h says that with zero timestamps the firmware completes at most one swap per 16.7 ms, so a 120 Hz panel presents at 60 fps. #358 (open) makes the same change for the T6030 record after the same measurement on a J514S. I have not tested setting it on the J514M.

    120 Hz with .swap_timestamps on the J514M record (edited 2026-11-08)

    Rebuilt kernel 7.2.11.aurora15-1.3 = 8eddf5aeb6ac plus three local changes: .swap_timestamps = true, in dcp_v14_board_j514m, the driver commits of #406, and a J514M dcpext0 device-tree node; m1n1 = c6fef490 plus omacom/m1n1#29. External displays were not enabled (no /etc/m1n1.conf) and none were connected. Mesa unchanged. Cold boot, then:

    es2gears_wayland: 597 frames in 5.0 seconds = 119.233 FPS; 601 frames in 5.0 seconds = 120.128 FPS
    glxgears:         597 frames in 5.0 seconds = 119.324 FPS; 601 frames in 5.0 seconds = 120.002 FPS
    wp_presentation, 1200 frames: mean 119.461 Hz, interval median 8.333 ms (1189 in the 8.5 ms bin, 8 near 12.5 ms, 1 at 20.9 ms)
    600 swaps submitted at random times: 373 intervals at odd multiples of 8.333 ms (0 before the change)
    

    GPU load numbers

    glmark2 2023.01, terrain, off-screen, 20 s per size:

    Size FPS Frame time
    400x300 64 15.8 ms
    800x600 59 17.0 ms
    1600x1200 52 19.6 ms
    3024x1964 40 25.1 ms
    800x600, LIBGL_ALWAYS_SOFTWARE=1 (llvmpipe) 25 40.8 ms

    Lighter scenes off-screen: build 1540 FPS, clear 1696 FPS; glxgears about 2050 FPS.

    /sys/kernel/debug/asahi-g15/stats and pstate, two reads 5.003 s apart during an off-screen terrain run:

    pstate 8 freq_khz 1380000 source stats
    render jobs: +17695 (3537/s)
    stage ta_us + fs_us (span_us): +573820 us = 11.5 % of the window
    starved_us: +40218 us (0.8 %), prep_us: +102112 us (2.0 %)
    horizon hold[0]: +53580 (10709/s)
    

    This kernel has no g15_max_inflight parameter (#261 is open). #261 describes hold[0] on the two-in-flight cap on a J516C. I have not tested whether #261 changes these numbers here.

    Power during the 60 s off-screen run (SMC PSTR / PHPC): total system 19.0 W average, 30.4 W peak; heat pipe 9.7 W average, 15.0 W peak (idle: 11.8 W / 4.7 W). Fans stayed at 0 RPM. NAND, battery, charge regulator, Wi-Fi and NVMe temperatures stayed within about 2 °C. This install exposes no SoC or GPU die temperature: the SMC hwmon node has the same keys as upstream hwmon-common/hwmon-laptop/hwmon-fan-dual.dtsi.

    Untested and limitations

    • Vulkan, external displays, suspend, and the owner opt-out switches, as listed above.
    • One cold boot. Load runs were short (60 s at most). No sustained-load or thermal qualification.
    • No reference terrain result for an M3 Max under macOS or another kernel, so these numbers are not compared with one.
  2. techgyani commented on Nov 8, 2026

    @techgyani
    Author

    External displays on this J514M with #406: pass on all three USB-C ports, hotplug and coldplug. Unreleased build, same as the 120 Hz section of my results comment above: kernel 7.2.11.aurora15-1.3 = 8eddf5aeb6ac + #406's driver commits + a J514M dcpext0 device-tree node + .swap_timestamps for the J514M; m1n1 c6fef490 + omacom/m1n1#29; /etc/m1n1.conf: chosen.asahi,t6034-dispext=1. Mesa unchanged.

    Monitor: LG 27UP850N (EDID name LG ULTRAFINE), connected by a USB-C to USB-C cable. No device appeared under /sys/bus/thunderbolt/devices.

    Action Result
    Hotplug, right-side port pass: DP-3 3840x2160@60, Type-C DPTX PHY 2, picture shown
    Hotplug, rear left port pass: DP-1 3840x2160@60, PHY 0, picture shown
    Hotplug, front left port pass: DP-2 3840x2160@60, PHY 1, picture shown
    Unplug, then replug on another port pass: dcp_dptx_disconnect(port=0), USB devices removed, the next port came up
    Coldplug, right-side port (monitor connected at power-on) pass after login: DP-3 3840x2160@60. Nothing on the monitor during boot (boot screen and disk-unlock prompt on the internal panel only)
    Internal panel at the same time pass: eDP-1 stays 3024x1964@120
    10-bit output to the monitor not available: see below
    Two external monitors at once (second LG on the rear left port, first on the right) fail: second stays dark, replugging does not help (log below)
    Lid closed with the monitor attached pass: no suspend (logind Docked true, HandleLidSwitchDocked=ignore); eDP-1 leaves hyprctl monitors, DP-3 keeps running
    Thunderbolt displays, HDMI not tested

    Kernel log (trimmed): the first line is from boot, the rest from the hotplug on the right-side port:

    apple-t6034-display: enabled dcpext0 (/soc/dcp@306c00000) with PMP DISPEXT0 request, CPU power floor, display DART /soc/iommu@305304000 and its Type-C routes; started by the display driver on first use
    apple-dcp 306c00000.dcp: allocated Type-C DPTX PHY 2
    apple-dcp 306c00000.dcp: display attached: starting the external display processor
    apple-dcp 306c00000.dcp: dcpext RTKit session running
    apple-dcp 306c00000.dcp: external display processor running
    

    On the rear left port, the firmware's mode set:

    IOMFBStatus UnifiedPipeline::mode_set_gated(uint32_t, uint32_t): 3840x2160@60 Hz DSC=NO VRR=NO link: 1
    apple-dcp 306c00000.dcp: display mode "3840x2160": 60 537600 3840 3888 3920 4000 2160 2163 2168 2240 0x48 0x0 (timing 42, color 71): 0
    

    Other lines seen, with the display working:

    apple-dcp 306c00000.dcp: AFK[ep:24]: failed to extract init props: -2
    IOMFB load_ca_data: Unrecognized data version 0 (expected 1 or 2)
    IOMFB read_pmu_data_sync: pmu ram read error (e00002d8)
    IOMFB WARNING: set_timings() called after timings were started.
    

    Coldplug timing: external display processor ready; its firmware starts when a display is attached at 1.7 s, display attached at 15.1 s (after the disk unlock), external display processor running at 16.2 s.

    Second monitor, plugged into the rear left port while the first was on the right-side port:

    sn201202x 1-0c: failed to configure Type-C connection: -28
    apple-dcp 306c00000.dcp: display on USB-C Left Back waits for a free external display pipeline (each one drives another display); it lights up on its own when one of those displays is unplugged
    sn201202x 1-0c: cable revalidation failed after 3 attempts
    

    Only dcp@306c00000 (dcpext0) is in this device tree, and aquamarine reports found 2 CRTCs.

    Colour depth. Omarchy sets Hyprland colour management to edid on Apple Silicon (default/hypr/wide-gamut.lua), so Hyprland shows "Monitor DP-3: wide color gamut is enabled but the display is not in 10-bit mode." A bitdepth = 10 monitor rule left the output at XRGB8888. In drivers/gpu/drm/apple/plane.c, apple_plane_init_v14_7 registers only dcp_primary_formats_14_7 (XRGB8888, ARGB8888), and aquamarine logs Plane 35 has 2 formats and Plane 42 has 2 formats. Hyprland also logs atomic drm: failed to commit hdr metadata: no HDR_OUTPUT_METADATA prop support.

    Not tested: Thunderbolt displays, HDMI, suspend with a monitor attached, lid closed without a monitor, and DSC modes.

  3. techgyani commented on Nov 8, 2026

    @techgyani
    Author

    Retest with a second external display processor and panel brightness. Unreleased build, compiled from source:

    Action Result
    Built-in panel brightness, brightnessctl 20 % / 100 % / back pass: /sys/class/backlight/apple-panel-bl, max 500; read back 100 / 500 / 140; the panel visibly changes
    Brightness keys pass
    One LG 27UP850N, right-side port pass: DP-3 3840x2160@60 on dcpext0 (306c00000)
    Second LG 27UP850N, rear left port, first still attached pass: DP-1 3840x2160@60 on dcpext1 (44ac00000); built-in eDP-1 at 3024x1964@120 at the same time
    Lid closed with both monitors pass: no suspend (logind Docked), eDP-1 leaves hyprctl monitors, both monitors keep running
    Brightness of the external monitors fail: no DDC bus on the DP connectors (no ddc link under /sys/class/drm/card1-DP-*; ddcutil finds only the SoC SMBus adapters), so Omarchy's DDC path has nothing to use
    HDR on the external monitors not available: 8-bit only (XRGB8888, ARGB8888), and atomic drm: failed to commit hdr metadata: no HDR_OUTPUT_METADATA prop support

    Kernel log, second monitor (trimmed):

    apple-t6034-display: enabled dcpext1 (/soc/dcp@44ac00000) with PMP DISPEXT1 request, CPU power floor, display DART /soc/iommu@449304000 and its Type-C routes; started by the display driver on first use
    apple-display-crossbar 70304c000.mux: dpphy: T602X route 2, clocks disabled
    apple-dcp 44ac00000.dcp: allocated Type-C DPTX PHY 0
    apple-dcp 44ac00000.dcp: display attached: starting the external display processor
    apple-dcp 44ac00000.dcp: external display processor running
    IOMFBStatus UnifiedPipeline::mode_set_gated(uint32_t, uint32_t): 3840x2160@60 Hz DSC=NO VRR=NO link: 1
    

    Also from the dcpext1 firmware, with the picture working: late_init_ios(bool): Realtime Bandwidth setup failed, vid-clock-to-disp-clock-factor read failed, using default, read_pmu_data_sync: pmu ram read error (e00002d8). Aquamarine reports found 3 CRTCs.

    Built-in panel brightness, from the panel's DCP:

    UPPipeDCP_H13P.cpp:2564: brightness_changed: brightness value: 6553600
    UPPipeDCP_H13P.cpp:2564: brightness_changed: brightness value: 32768000
    UPPipeDCP_H13P.cpp:2564: brightness_changed: brightness value: 9175040
    

    Not tested: three or four external displays, HDMI, Thunderbolt displays, suspend with displays attached, lid closed without a monitor.

  4. webdevtodayjason commented on Nov 10, 2026

    @webdevtodayjason

    Second J514M outside the lab: GPU, native display, 120 Hz and brightness work on the default path

    Another apple,j514m / apple,t6034 here (14-inch M3 Max, 14 CPU cores, 30 GPU cores, 36 GB). Installed from the stable channel, then built the three J514M pieces from source with the stock PKGBUILDs from omarchy-pkgs. No /etc/m1n1.conf, no module parameters, no firmware override.

    Build

    Piece Version Source
    linux-aurora 7.2.11.aurora15-1.2 (7.2.11-15-1.2-ARCH) integration/aurora at 49a8c150, which has #408, plus the two-line patch below
    mesa-aurora, vulkan-aurora 26.3.0_devel.20261107 b7b03c5b3b8 integration/aurora, which has omacom/mesa#33. 129 of 129 Meson tests passed
    m1n1-aurora 1.6.1.aurora22-1.1 omacom/m1n1#32 head c6fef490
    hyprland / aquamarine 0.56.2-7 / 0.15.1-1.3 stable channel, unchanged

    Boot chain from /proc/device-tree/chosen:

    asahi,iboot1-version: mBoot-18000.120.36
    asahi,iboot2-version: iBoot-10151.140.19.700.2
    asahi,m1n1-stage1-version: v1.6.1-dirty
    asahi,m1n1-stage2-version: c6fef490-dirty
    asahi,os-fw-version: 14.7
    asahi,system-fw-version: 26.5
    

    Before (stock stable: aurora14, m1n1 bc79acc)

    • GPU node disabled, /dev/dri had only card0 on simple-framebuffer.
    • Hyprland on llvmpipe: 444% CPU in one top sample, about 1,150% in another. Load average 7.3.
    • Panel at 3024x1890@60. Fans near 2,900 and 3,100 rpm at idle.

    After

    Test Result Notes
    Boot to desktop, LUKS root pass Two reboots, one per kernel build. No cold boot from power off yet
    GPU probe pass asahi-g15 bound, renderD128 present
    Firmware identity pass, no override J514M RTKit 2419.140.12 (runtime qualification pending), M3: firmware identity accepted
    Core layout pass cores enabled per cluster: [8, 8, 7, 7]
    OpenGL pass Apple M3 Max (G15C B1), 4.6 compatibility profile, accelerated
    Vulkan pass (enumeration only) Apple M3 Max (G15C B1), Borealis, API 1.4.363
    Native internal display pass eDP-1 3024x1964@120 on apple-drm
    120 Hz presentation pass with patch vsynced glxgears under Xwayland: 120.07 fps over 5 s. Not measured without the patch
    Panel brightness pass with patch apple-panel-bl, max 500. Set and restore through brightnessctl works. No backlight device without the patch
    Hyprland idle CPU pass 3 to 6.5% of one core
    Fans at idle 0 rpm Both fans
    DART or GPU faults in dmesg none Checked within 10 minutes of each boot. No long session yet
    Wi-Fi pass 5 GHz, 160 MHz HE
    Audio not retested Speaker DSP sink and speakersafetyd are up. No listening test on this build

    The patch (J514M board record)

    Same change #416 makes for this board, cut down to the J514M lines. Happy to send it as its own PR if that is useful ahead of #416.

    --- a/drivers/gpu/drm/apple/iomfb_v14_7.c
    +++ b/drivers/gpu/drm/apple/iomfb_v14_7.c
    @@ -226,6 +226,9 @@ static const struct dcp_v14_board dcp_v14_board_j514m = {
     	.handoff = "apple,t6034-handoff",
     	.promotion = true,
     	.log_bringup = true,
    +	/* As on the J516C: stamped swaps for 120 Hz, and the panel level block. */
    +	.swap_timestamps = true,
    +	.panel_nits = true,
     };
    @@ -1740,7 +1743,7 @@ static bool dcp_v14_panel_nits_admit(const struct dcp_v14_board *board,
     		 machine_is(ctx, "apple,j516c") || machine_is(ctx, "apple,j613") ||
    -		 machine_is(ctx, "apple,j504"));
    +		 machine_is(ctx, "apple,j504") || machine_is(ctx, "apple,j514m"));

    Log lines that may help

    GPU:

    asahi-g15 40a400000.gpu: G15: probing (T6034 G15C)
    asahi-g15 40a400000.gpu: T6034 G15C bring-up: unqualified J516C reference: max_power_mw=116739 relative_power=[11, 23, 33, 44, 55, 71, 90] fast_die_mask=0x802a002b802a002b alarm_mask=0x902a102b802a002b; OPP powers are boot-loader stand-ins
    asahi-g15 40a400000.gpu: GPU firmware build: J514M RTKit 2419.140.12 (runtime qualification pending)
    asahi-g15 40a400000.gpu: M3: firmware identity accepted
    asahi-g15 40a400000.gpu: board data sha256 123d45ea418ffb20d6cc6bce0420c519117a9fbd17ea5e78f2cb45bce9754ae0
    asahi-g15 40a400000.gpu: T6034 G15C admitted, ASC=0x0
    asahi-g15 40a400000.gpu: T6034 G15C: power acknowledged, version=0x7041100 counts=0x21040a core-mask=0x775ed7f5fd
    asahi-g15 40a400000.gpu: T6034 G15C: cores enabled per cluster: [8, 8, 7, 7]
    asahi-g15 40a400000.gpu: M3: firmware accepted owned initdata and device controls
    asahi-g15 40a400000.gpu: G15: firmware at READY, starting the backend
    

    Display clock, for the open 400 MHz question on #408. The DCP firmware reports the clock is slightly under the panel's video clock, and the panel still runs at 120 Hz:

    apple-dcp 386c00000.dcp: display clock 0: 400000000 Hz
    apple-dcp 386c00000.dcp: RTKit: syslog message: PPipeDCP_H13P.cpp:13844: IOMFB updateFrequencies EDT ERROR: getClockFrequency(0) (400000000) < videoClock 402278400! Giving up on frequencies.
    apple-dcp 386c00000.dcp: RTKit: syslog message: UPPipeDCP_H13P.cpp:1732: IOMFB: videoClock is zero, defaulting to 402280000
    apple-dcp 386c00000.dcp: RTKit: syslog message: UPPipeDCP_H13P.cpp:1775: virtual bool IOMFB::UPPipeDCP_H13P::late_init_ios(bool): Realtime Bandwidth setup failed
    

    Two things I hit on the way

    Not tested yet

    • Suspend and resume.
    • Cold boot from power off.
    • External displays, HDMI, Thunderbolt.
    • Vulkan or GL compute with checked results.
    • Sustained load, power and thermals.
    • The m3-defaults=0 owner-line path from m1n1#32.

    I can run any of these on this unit. Tell me which matter most for qualifying J514M and I will post results here.

  5. webdevtodayjason commented on Nov 10, 2026

    @webdevtodayjason

    J514M follow-up: 10-minute CPU plus GPU load, suspend, GPU memory heap

    Same machine and build as my report above (apple,j514m, 36 GB, 7.2.11-15-1.2-ARCH, Mesa b7b03c5b3b8, m1n1 c6fef490). Still no /etc/m1n1.conf, no module parameters, no firmware override.

    Sustained load

    Ten minutes of combined load on AC power, lid open, starting from idle. Ambient temperature was not measured.

    • CPU: llama.cpp llama-bench, 10 threads, a 17.2 GiB model held in RAM (-dev none -ngl 0).
    • GPU: llama.cpp Vulkan backend, a 0.8B model fully offloaded, alternating 2,048-token prompt and 1,024-token generation passes.
    • Sensors sampled every 10 s from macsmc_hwmon, with a 30 s baseline and 120 s of cooldown.
    Time into load Fan 1 / Fan 2 (rpm) Total System Power Charge Regulator WiFi/BT Module NAND
    idle baseline 0 / 0 12.6 W 31.4°C 32.5°C 27.4°C
    60 s 0 / 0 51.6 W 43.1°C 48.4°C 29.3°C
    90 s 2302 / 2484 52.0 W 48.2°C 55.5°C 30.9°C
    180 s 3296 / 3556 57.4 W 56.4°C 64.3°C 35.1°C
    300 s 3997 / 4278 55.1 W 55.5°C 61.0°C 36.7°C
    420 s 4946 / 5349 52.7 W 54.6°C 59.1°C 37.2°C
    600 s 4950 / 5349 52.8 W 52.8°C 56.4°C 36.5°C
    120 s after load 4898 / 5350 13.4 W 39.8°C 41.0°C 33.4°C

    Summary over the load window:

    • Total System Power: 53.5 W average, 80.6 W peak in one sample. Heatpipe Power: 39.9 W average, 66.7 W peak.
    • Fans stayed at 0 rpm for about 80 s, then stepped up to about 4,950 and 5,350 rpm. The visible temperatures peaked near 180 s and fell as the fans rose.
    • No throughput decay. CPU job: 14 runs between 10.7 and 13.0 tokens/s, the last one the fastest. GPU job: 26 prompt passes between 1,787 and 1,837 tokens/s, 26 generation passes between 47.5 and 55.2 tokens/s.
    • scaling_cur_freq on the performance clusters read 4,056 MHz in nearly every sample.
    • Kernel log for the whole run: no GPU faults, no DART faults, no timeouts, no OOM kills. The only new lines were firewall drops.
    • Fans were still near 5,000 rpm two minutes after the load ended at 13 W. They were back at 0 rpm when I checked later. I did not log the time in between.

    Limits of this data:

    • This board exposes no CPU or GPU die temperature to Linux here. sensors shows only NAND, battery hotspot, charge regulator and WiFi/BT module, so I cannot say how hot the SoC ran.
    • One run, one ambient, AC only. No battery-power run and no lid-closed run.
    • The six tas2764 temperature inputs read FAULT at idle. I did not sample them under load.

    Suspend

    One suspend and resume cycle so far. The desktop came back, and the GPU work above ran after that resume. I did not capture logs for that cycle, so treat it as a single informal pass. Cold boot from power off is still untested.

    GPU memory heap (Mesa, not kernel)

    • The Vulkan device-local heap reports 17.40 GiB on this 36 GB machine. With HK_SYSMEM=28000000000 it reports 27.94 GiB.
    • A 17.2 GiB model fully offloaded fails with ErrorOutOfDeviceMemory at llama.cpp's default batch size under the default heap. It loads only with -b 128 -ub 128.
    • With the larger heap the same model loads at the default batch size, and prompt processing goes from 82 to 201 tokens/s.
    • Generation on that model stays at 4.7 tokens/s either way, against 19.6 tokens/s CPU only. It is a mixture-of-experts model, and a small dense model generates at about 54 tokens/s on the same GPU. I have not profiled where the time goes, so I am not calling this a driver problem.

    Happy to move the heap note to omacom/mesa if that is the better place.

    Can run next

    • The same load on battery, and a longer 30-minute run.
    • A logged suspend and resume loop, and a cold boot.
    • Anything specific you want sampled during load. I have the raw 10 s sensor log and can attach it.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions