aesthetic.computer
Monorepo for Aesthetic.Computer
aesthetic.computer
docs: update NixOS analysis — deprecate for production, use bare-metal
After ~20 build iterations, NixOS DRM-direct runs at 27fps vs 60fps
bare-metal on ThinkPad 11e. The generic kernel's i915 page flip takes
27ms vs 16ms. Falling back to ac-os build for @oskie's USB.
NixOS path preserved in git for future investigation.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
+100
-3
+14
-3
reports/2026-04-06-ac-native-nix-vs-baremetal-pipeline-analysis.md
+14
-3
reports/2026-04-06-ac-native-nix-vs-baremetal-pipeline-analysis.md
···
90
90
**Pros:** Simplest path. Proven. Fast.
91
91
**Cons:** Loses reproducibility narrative. Build depends on host state.
92
92
93
-
## Verdict
93
+
## Verdict (Updated 2026-04-07)
94
+
95
+
**Option D confirmed after extensive testing.** We implemented Option B (DRM-direct, no cage) across ~20 build iterations. Results:
96
+
97
+
- Cage/Wayland bugs were fixed (cursor, double input, shutdown)
98
+
- DRM-direct mode works but runs at **27fps** (vs 60fps bare-metal)
99
+
- `present_us` consistently 24-28ms despite: removing frame_sync, disabling mitigations, disabling WiFi, disabling logind
100
+
- The NixOS generic kernel's DRM/i915 stack simply performs worse than the custom kernel
101
+
- Input required MAX_INPUT_DEVICES bump from 8→24 (ThinkPad has 18 evdev nodes)
94
102
95
-
**Option B is likely the sweet spot.** The Wayland compositor is the single biggest source of bugs and performance loss. Removing cage while keeping NixOS for the system layer would eliminate 80% of the issues encountered in this session while retaining reproducibility.
103
+
**For @oskie and production: use `ac-os build` (bare-metal).** NixOS path is preserved in git for future work but not suitable for shipping today.
96
104
97
-
However, **for @oskie's USB today, Option A is correct** — ship the bare-metal build that works, and iterate on NixOS in parallel.
105
+
### What would fix NixOS performance
106
+
1. Custom kernel config matching the bare-metal build's `.config`
107
+
2. Or identify why the NixOS i915 DRM page flip takes 27ms vs 16ms (kernel config diff)
108
+
3. Or render directly into DRM buffer instead of fb_copy_scaled
98
109
99
110
## Open Questions
100
111
+86
reports/2026-04-07-thinkpad-11e-hardware-profile.md
+86
reports/2026-04-07-thinkpad-11e-hardware-profile.md
···
1
+
# ThinkPad 11e Yoga Hardware Profile (for initrd optimization)
2
+
3
+
**Date:** 2026-04-07
4
+
**Machine:** Lenovo ThinkPad 11e Yoga (5th Gen)
5
+
**Build:** NixOS DRM-direct (no cage/Wayland)
6
+
7
+
## Hardware Summary
8
+
9
+
| Component | Details | Kernel Module |
10
+
|-----------|---------|---------------|
11
+
| **CPU** | Intel Core m3-8100Y @ 1.10GHz (4 cores) | - |
12
+
| **GPU** | Intel UHD Graphics 615 [8086:591c] | i915 |
13
+
| **RAM** | 8 GB | - |
14
+
| **WiFi** | Intel Wireless-AC 9260 [8086:2526] | iwlwifi, iwlmvm |
15
+
| **NVMe** | KIOXIA BG4 256GB [1e0f:0001] | nvme |
16
+
| **Audio** | Sunrise Point-LP HD Audio [8086:9d71] | snd_hda_intel |
17
+
| **USB** | Sunrise Point-LP USB 3.0 xHCI [8086:9d2f] | xhci_pci |
18
+
| **Touchpad** | Elan i2c [04f3:2a40] | elan_i2c, hid_multitouch |
19
+
| **Touchscreen** | ELAN901C i2c [04f3:2a40] | i2c_hid_acpi |
20
+
| **Bluetooth** | Intel (on WiFi card) | btusb, btintel |
21
+
| **Card Reader** | Realtek RTS522A [10ec:522a] | rtsx_pci |
22
+
| **DRM device** | /dev/dri/card1 (not card0!) | i915 |
23
+
24
+
## Input Devices (18 evdev nodes)
25
+
26
+
- event0: Lid Switch
27
+
- event1: Power Button
28
+
- event2: Sleep Button
29
+
- event3: Power Button
30
+
- event4: **AT Translated Set 2 keyboard** (main keyboard)
31
+
- event5-9: HDA audio jacks
32
+
- event10-12: ThinkPad Extra Buttons, sensor hub
33
+
- event13-14: ELAN touchpad (SMBus + i2c)
34
+
- event15-17: ELAN touchscreen (multitouch + stylus)
35
+
36
+
**Critical:** MAX_INPUT_DEVICES was 8, missing event4 (keyboard). Fixed to 24.
37
+
38
+
## Performance Profile (NixOS DRM-direct)
39
+
40
+
| Metric | Value | Target |
41
+
|--------|-------|--------|
42
+
| **Avg frame time** | 35-52ms (19-28 fps) | 16.7ms (60 fps) |
43
+
| **Max frame time** | 65ms | <20ms |
44
+
| **Paint time** | 9-16ms | OK |
45
+
| **Present (page flip)** | 24-28ms | 16ms (1 vblank) |
46
+
| **Slow frames** | 300/300 (100%) | <5% |
47
+
| **Boot time** | ~90s | <10s |
48
+
49
+
**Root cause of low FPS:** DRM page flip (`present_us`) takes 24-28ms instead of 16ms. Suggests vsync contention or double-sync with `frame_sync_60fps`. The old bare-metal build hit 60fps on the same hardware.
50
+
51
+
## Modules Needed for Slim Initrd
52
+
53
+
Based on this hardware profile, a targeted initrd would need:
54
+
55
+
```nix
56
+
boot.initrd.includeDefaultModules = false;
57
+
boot.initrd.availableKernelModules = [
58
+
# Storage (USB boot)
59
+
"xhci_pci" "usb_storage" "uas" "sd_mod"
60
+
# NVMe (internal disk)
61
+
"nvme"
62
+
# GPU (i915 KMS)
63
+
"i915"
64
+
# Filesystems
65
+
"ext4" "vfat" "nls_cp437" "nls_iso8859_1"
66
+
];
67
+
68
+
boot.kernelModules = [
69
+
# WiFi
70
+
"iwlwifi" "iwlmvm"
71
+
# Audio
72
+
"snd_hda_intel" "snd_hda_codec_realtek"
73
+
# Input
74
+
"elan_i2c" "i2c_hid_acpi" "hid_multitouch"
75
+
# Bluetooth
76
+
"btusb" "btintel"
77
+
];
78
+
```
79
+
80
+
This would reduce initrd from ~1.4GB to ~50MB, cutting boot from 55s to ~5s.
81
+
82
+
## Open Issues
83
+
84
+
1. **FPS**: present_us too high — investigate double vsync or page flip contention
85
+
2. **WiFi scan**: still running despite creds check — `/mnt` mount may differ
86
+
3. **card1 not card0**: ThinkPad 11e uses card1 for i915, needs wildcard in DRM wait