Skip to content

Instantly share code, notes, and snippets.

@allenk
allenk / ARCR-Outperform-AI-RT-Tasks.md
Created October 21, 2025 19:45
Android Realtime Content Recognition: Outperforms AI for Real-Time Tasks
AI Inference Render-Based Method
Heavy GPU/NN load Zero ML cost
Frame delay / batching Real-time per-frame
Statistical guess Deterministic signal
Pixel-bound System-intent bound
@allenk
allenk / Frame-Time-Metadata.md
Created October 21, 2025 19:41
The Core Concept: Frame-Time Metadata
Bit Range Meaning
0 Frame Reset
1 Video Layer Detected
2 GUI Overlay Present
3 Vertical Scroll
4 Horizontal Scroll
5 Subtitle/Text Present
6 Frame Skip
7 Verification Flag
@allenk
allenk / Android-Canvas-Semantic-Patterns.md
Created October 21, 2025 19:38
Reading the Canvas: Extracting Semantic Patterns
Command Meaning Possible Scene Element
DrawRect Background or clear region Scene reset / transition
DrawImageRect Texture or video frame Video playback
DrawTextBlob Rendered text glyphs Subtitle or caption
DrawDrawable Composed child nodes GUI overlays, buttons
@allenk
allenk / MyHooker-Detours-Compare.md
Created October 1, 2025 17:30
Comparison Our Game Hooker with Microsoft Detours
Aspect Our Framework Microsoft Detours
Platform ARM64/Android x86/x64/ARM/ARM64 Windows
Language Modern C++ C/C++
Hook Method GOT/PLT + Inline Inline Patching
API Style Declarative/Chainable Imperative
Thread Safety Built-in Manual Synchronization
@allenk
allenk / cuda-vulkan-power.md
Created August 31, 2025 17:31
CUDA VULKAN Power Consumption Review
Mode SM Utilization Power Increase Temp Change Notes
Vulkan Medium-low +0W (baseline) +0°C Auto fallback, relatively low load
CUDA (On) 100% ~+250W ~+8°C Full load, remains thermally stable
@allenk
allenk / blackwell-pytorch-benchmark.md
Last active August 31, 2025 17:29
Blackwell, PyTorch CUDA Support
ID Python Version PyTorch Version Inference Mode GPU Support Status Avg. Time per Image GPU Utilization Notes
3.9.13 2.4.1 Vulkan fallback ❌ Not supported (sm_120) ~2.12 sec Low (occasional spikes) Auto fallback to Vulkan/NCNN
3.13.7 2.9.0 Nightly Vulkan fallback ✅ Supported (sm_120, CUDA off) ~1.94 sec Medium-low Forced with --nocuda
3.13.7 2.9.0 Nightly CUDA ✅ Supported (sm_120, enabled) ~1.04 sec High (stable 100%) Full CUDA inference enabled
@allenk
allenk / cuda-vulkan-power-tc.md
Last active August 31, 2025 17:30
CUDA VULKAN Power Consumption Review
模式 SM 使用率 功耗增加 溫度變化 備註
Vulkan 中等偏低 +0W(基準) +0°C 自動 fallback,負載不高
CUDA(啟用) 100% 約 +250W 約 +8°C 滿負載運行,仍保持穩定
@allenk
allenk / blackwell-pytorch-benchmark-tc.md
Created August 31, 2025 17:17
Blackwell, PyTorch CUDA Support
編號 Python版本 PyTorch版本 推論模式 GPU支援狀態 每張處理時間 GPU使用率 備註
3.9.13 2.4.1 Vulkan fallback ❌ 不支援 sm_120 約 2.12 秒 低(偶爾跳高) 自動 fallback
3.13.7 2.9.0 Nightly Vulkan fallback ✅ 支援 sm_120 未啟用 CUDA 約 1.94 秒 中等偏低 關閉CUDA
3.13.7 2.9.0 Nightly CUDA ✅ 支援 sm_120 並已啟用 約 1.04 秒 高(穩定 100%) 完整啟用 CUDA 模型推論
@allenk
allenk / deepclean_test.cmd
Last active September 15, 2025 15:42
ASUS Software Clean Up Tool Test version (This version is for testing purposes only, please use the released version)
@echo off
:: ------------------------------------------------------------------------------------------------------------
:: Clean Up ASUS All
:: ------------------------------------------------------------------------------------------------------------
:: The tool helps to clean up all ASUS software from system
:: ------------------------------------------------------------------------------------------------------------
:: Before running the tools,
:: 1. Complete backup your system.
:: 2. Disable ASUS Apps from BIOS (MyASUS and Armoury)
:: 3. Run ASUS remove tools (Armoury Crate Uninstall Tool.exe, or geek_uninstall.exe).
@allenk
allenk / cache_mon.sh
Created April 16, 2024 14:04
WSL2 Cache Memory Auto Release
#!/bin/bash
# Memory threshold
THRESHOLD_AVAIL=10 # MemAvailable
THRESHOLD_FREE=10 # Free
# Loop to test memory
while true; do
# fetech memory status
TOTAL_MEM=$(grep MemTotal /proc/meminfo | awk '{print $2 / 1024}')