🚀 SmoothGL V7-Stable — Official Patch Notes / Официальный Патчноут

English version below / Русская версия ниже

🇺🇸 English Changelog In this update, we conducted a total codebase sweep (4 stages of deep code audits), fixed complex fatal crashes, and added custom features from our team!

🛠️ Critical Fixes (Crashes) CAUTION

FATAL ERROR: OpenGL Background Thread Crash A past agent added java.lang.ref.Cleaner to automatically clean up GlBuffer and GlVertexArray memory. However, OpenGL strictly forbids deleting resources outside the render thread! This caused hard JVM crashes. We completely ripped the Cleaner out of low-level classes, fixing the crash entirely.

WARNING

BufferOverflowException on RTX 5050 (PulseVisuals) The mod tried to overwrite the Windows command line to enable NVIDIA driver optimizations, but the buffer overflowed due to the short command line of the PulseVisuals launcher. We rewrote the workaround to gracefully skip this step if the buffer is too small, allowing the game to launch successfully.

🧊 Rendering & Graphics Fixed "Broken" Translucent Glass (Translucent Culling Bug): We discovered that the built-in IntervalTree library (used for sorting transparent blocks) was fundamentally broken.

Deleted glass faces remained in memory as "ghosts". The algorithm returned reversed booleans (Set Contract Violation). We rewrote the library: replaced the slow legacy Stack with a fast ArrayDeque and fixed the logic. Glass, ice, and water layers now render perfectly smoothly without missing faces. Restored Custom Blocks (NeoForge): The NeoForge port aggressively canceled standard block rendering (ModelBlockRendererMixin). This made mechanisms from Create or Xaero's Minimap waypoints completely invisible. We added the missing isVanillaAdapter() check, restoring full support for third-party mods!

⚙️ UI Optimization & Concurrency TIP

Smooth Settings without Micro-Stutters Previously, closing the settings menu caused the game to stutter because the mod synchronously saved the JSON config file on the UI render thread. We wrapped the saving process in CompletableFuture.runAsync(). The menu now closes instantly while the config writes in the background!

Data Race Protection: In SodiumGameOptions, the chunk builder thread settings were read asynchronously without thread protection. We added the volatile modifier to all parameters, completely eliminating the risk of ConcurrentModificationException when changing settings on the fly. 💎 Team_brat Exclusives We embedded a branded section right into the mod's graphics menu!

Появилась новая вкладка "SmoothGL (Team_brat)", содержащая уникальные переключатели:

Pulse Engine Boost: Optimization tailored specifically for the PulseVisuals launcher. Aggressive Memory Unloading: Keeps RAM usage at a minimum by utilizing an aggressive garbage collector upon world unloads. Strict Glass Sorting: A toggle for our newly fixed, highly accurate transparency algorithm. 🇷🇺 Русский Патчноут В этом обновлении мы провели тотальную зачистку архитектуры мода (4 стадии глубокого аудита кода), исправили сложнейшие краши и добавили кастомные фишки от нашей команды!

🛠️ Критические исправления (Краши и Вылеты) CAUTION

FATAL ERROR: OpenGL Background Thread Crash Один из прошлых агентов добавил java.lang.ref.Cleaner для автоматической очистки памяти буферов (GlBuffer, GlVertexArray). Но OpenGL строго запрещает удалять ресурсы вне потока рендеринга! Из-за этого игра намертво вылетала (Fatal JVM Crash). Мы вырезали Cleaner из низкоуровневых классов, полностью устранив этот вылет.

WARNING

BufferOverflowException на RTX 5050 (PulseVisuals) Мод пытался подменить командную строку Windows для включения оптимизаций NVIDIA, но буфер переполнялся из-за особенностей лаунчера PulseVisuals. Мы переписали хак: теперь мод мягко пропускает этот шаг, если места в буфере не хватает, позволяя игре успешно запускаться.

🧊 Рендеринг и Графика Фикс "сломанного" полупрозрачного стекла (Translucent Culling Bug): Мы обнаружили, что встроенная библиотека IntervalTree, отвечающая за сортировку прозрачных блоков, была насквозь дырявой.

Удаленные грани стекла оставались в памяти как "призраки". Алгоритм неправильно отвечал на проверки добавления/удаления (Set Contract Violation). Мы переписали библиотеку: заменили медленный Stack на ArrayDeque и исправили логику. Теперь слои стекла, льда и воды отрисовываются идеально ровно, без исчезающих граней. Возвращение кастомных блоков (NeoForge): Порт под NeoForge слишком агрессивно отменял стандартный рендер блоков (ModelBlockRendererMixin). Из-за этого механизмы из Create или вейпоинты Xaero's Minimap были тупо невидимыми. Мы добавили проверку isVanillaAdapter(), вернув поддержку сторонних модов!

⚙️ Оптимизация интерфейса и Многопоточности TIP

Плавные настройки без микро-фризов Раньше при закрытии меню настроек игра "заикалась", потому что мод синхронно сохранял JSON-файл прямо в потоке отрисовки кадров. Мы обернули сохранение в CompletableFuture.runAsync(). Теперь меню закрывается мгновенно, а конфиг пишется в фоне!

Защита от состояния гонки (Data Race): В SodiumGameOptions настройки количества потоков (Chunk Builder Threads) читались асинхронно без защиты. Мы добавили модификатор volatile ко всем параметрам, навсегда убрав риск ConcurrentModificationException при смене настроек "на лету". 💎 Эксклюзив от Team_brat Мы внедрили фирменный раздел прямо в графическое меню мода!

Появилась новая вкладка "SmoothGL (Team_brat)", содержащая уникальные переключатели:

Pulse Engine Boost: Оптимизация специально под лаунчер PulseVisuals. Aggressive Memory Unloading: Удержание потребления ОЗУ на минимальном уровне за счет агрессивного сборщика мусора при выгрузке локаций. Strict Glass Sorting: Переключатель того самого точного, исправленного алгоритма прозрачности. Build / Сборка: SmoothGL-V7-Stable.jar Version / Версия: Minecraft 1.21.4 (Fabric/NeoForge)

SmoothGL V6 Alpha English | Русский

🇬🇧 English SmoothGL is a custom rendering engine focused on improving game stability and eliminating micro-stutters during gameplay.

Changes in V6 Alpha Frame Budget Broker: Added a new load balancer that dynamically limits the chunk loading rate if it detects FPS drops. This helps maintain smoothness during high-speed traversal, such as flying with Elytra. Upload Smoothing Scheduler: Altered the algorithm for uploading geometry to the GPU. Geometry is now dispatched in more evenly distributed batches to prevent overloading the PCIe bus. Memory Optimizations: Removed hidden memory allocations (BoundingBox objects) in the hot loops of chunk building, which significantly reduces the pressure on the Java Garbage Collector (GC). Culling Fixes: Fixed bugs causing foliage and translucent blocks to disappear at the render distance borders, and disabled an experimental chunk scanner that caused severe stuttering when running. Installation Place smoothgl-v6-alpha.jar into your mods folder. The mod works independently and completely replaces the standard rendering algorithm.

🇷🇺 Русский SmoothGL — это кастомный движок рендеринга, нацеленный на улучшение стабильности и устранение микрофризов во время игры.

Изменения в версии V6 Alpha Управление нагрузкой (Frame Budget Broker): Добавлен новый балансировщик, который динамически ограничивает скорость загрузки новых чанков, если замечает просадки FPS. Это помогает сохранять плавность при полете на элитрах. Сглаживание загрузки (Upload Smoothing Scheduler): Изменен алгоритм отправки геометрии на видеокарту. Геометрия теперь отправляется более равномерными порциями, чтобы не перегружать шину данных. Оптимизация памяти (Memory Optimizations): Убраны скрытые выделения памяти (объекты BoundingBox) в горячих циклах сборки чанков, что значительно снижает нагрузку на сборщик мусора (Java GC). Исправления отсечения (Culling Fixes): Исправлены баги с исчезновением листвы и прозрачных блоков на границах прорисовки, а также отключен экспериментальный сканер чанков, вызывавший жесткие зависания при беге. Установка Поместите smoothgl-v6-alpha.jar в папку mods. Мод работает самостоятельно и полностью заменяет собой стандартный алгоритм рендера.

SmoothGL V5 — Performance & Stability Update (Modrinth Changelog)

🔧 Fixes & Optimizations

1. FogRenderer crash fix

• Updated setupFog mixin injection signature for Minecraft 1.21.4
• Disabled vanilla fog by forcing extremely large fog distances (999999.0f)

2. Thread-safety improvements — PipelineTrace

• Replaced HashMap with ConcurrentHashMap for chunkStates

• Replaced counters with AtomicLong

• Added cleanup of chunkStates entries after RENDER_SUBMITTED

  • Prevents unbounded memory growth
  • Reduces GC spikes and long pauses

3. Hot path optimization — Double-Checked Locking

Implemented volatile + DCL singleton access for:

• GeometryKernel
• PipelineTrace
• ParticleCullingSystem
• SmoothGLDecisionEngine
• StabilityLayer
• FrameBudgetController
• SmoothGLController
• SmoothGLApi
• RenderImpactProfiler

Additional changes:

• Removed unnecessary synchronized methods from hot paths

4. Removed expensive logic from render hot paths

• Removed per-face distance-based leaf LOD checks from AbstractBlockRenderContext.isFaceCulled()

• Eliminated repeated:

  • getInstance() calls
  • BlockPos allocations
  • instanceof checks per face

• Sodium occlusion cache now handles inter-leaf culling more efficiently

5. Cached renderer references — BlockRenderer

Added cached fields:

• cachedGeom
• cachedCuller
• cachedLeafCullDistSq

References are now resolved once during prepare() instead of per-block rendering.

6. Debug overhead disabled

• debugMode = false by default across SmoothGL systems
• Removed System.out.println calls from rendering hot paths
• Disabled unused PipelineTrace.enable() initialization

Result:

• Lower CPU overhead
• Reduced console spam
• Better frame consistency

7. More aggressive performance tuning

GeometryKernel

• leafCullDistance: 24.0f (was 32)
• simplificationDistance: 48.0f (was 64)
• Enabled alphaOverdrawReduction

ParticleCullingSystem

Enabled:

• Distance culling
• Frustum culling

Particle thresholds:

• Smoke: 32
• Rain: 48
• Default: 24

8. Particle culling restored

• Restored missing logic in SingleQuadParticleMixin.renderRotatedQuad()
• Particle culling works correctly again

9. Removed atomic operations from hot paths

• Removed addAlphaOverdrawReduced(1) from leaf culling in BlockRenderer.renderModel()
• Reduces synchronization overhead during chunk rendering

10. Leaf rendering fixes

Fixed issue where leaves would not appear until a neighboring block was updated.

Changes:

• Added explicit leaf-to-leaf checks in AbstractBlockRenderContext.isFaceCulled()
• Leaves now always render faces between each other
• Bypasses problematic occlusion behavior

11. Updated mod metadata

• Updated fabric.mod.json
• Added detailed feature descriptions for Modrinth/launcher display

🚀 Overall Results

• Lower CPU overhead in chunk rendering
• Reduced memory pressure and GC spikes
• Improved thread safety
• Better particle rendering performance
• More stable rendering pipeline on Minecraft 1.21.4
• Faster and more consistent frame times

SmoothGL V4 Beta 4 — Max FPS Release ✨ New Features Priority Calculator v2 with chunk state tracking Controller v2 with adaptive decisions StabilityLayer for anti-oscillation L3 Cache with GPU upload optimization Render Impact Profiler with Debug HUD 🔧 Improvements Simplified architecture Clean hooks interface Stable Mixin integration Integration with Sodium 0.6.13 Reduced runtime overhead Improved scheduling stability 🐛 Fixes Fixed Mixin package paths Fixed Gradle configuration Removed unstable reflection adapters Improved build stability ⚙️ Enabled by Default

All core optimizations are enabled for maximum FPS:

Upload Ring (128MB) Rebuild Debounce (2ms) Frame Arena (32MB) Priority v2 Controller v2 StabilityLayer L3 Cache 📊 Main Goal of V4

SmoothGL V4 focuses on:

higher FPS reduced micro-stutter more stable frametime lower CPU/GPU spikes smoother gameplay

🎨 SmoothGL v0.3.1.0-beta (The GUI Update) This update introduces a fully-featured, intuitive graphical user interface (GUI) for rendering settings, allowing you to manage and fine-tune all optimization modules effortlessly without typing console commands.

🖼️ Graphical User Interface (Render Settings GUI) No more commands! We've built a clean, native-feeling configuration menu:

Interactive Settings Menu: A complete Render Settings GUI equipped with toggle switches and sliders for real-time adjustments.

Module Toggle Buttons: Easily turn individual optimization modules ON/OFF with a single click:

L3 Cache & Hash Cache

Frame Pacer & Input Pipeline

Packet Pump

Dynamic Render Distance & Entity Tick Opt

FSR Upscaling

Precision Fine-Tuning Sliders: Added sliders for precise, real-time control over core parameters:

L3 Cache Size — Adjust allocated cache memory.

Frame Pacer Frames — Tweak internal frames-in-flight depth.

FSR Quality & FSR Scale — Smoothly scale your upscaling settings.

Integrated Performance Metrics: Real-time FPS indicator built directly into the settings menu for instant benchmarking.

🖱️ Controls & Navigation Smooth Mouse Scroll: Implemented mouse-wheel scrolling support for seamless navigation through long lists of settings.

New Keybind (F8): The Render Settings menu now opens via the F8 key (migrated from Right Shift for better accessibility).

HUD Toggle (F7): Use the F7 key to quickly toggle the performance breakdown HUD overlay.

💾 Configuration & Auto-Saving Save & Close: Clicking the Save & Close button instantly commits your changes to the config/smoothgl.properties file.

Cancel Changes: The Close button exits the menu without saving, reverting any accidental tweaks.

🚀 SmoothGL v0.3.0-beta30 This major update introduces full support for AMD FidelityFX Super Resolution (FSR) paired with CAS Shaders, alongside a revolutionary overhaul of the L3 Vertex Buffer Cache (v2-v14) for massive rendering pipeline optimizations.

🔧 Graphics & Rendering Engine (FSR & CAS) Added FSR Upscaling: Implemented full integration of AMD FidelityFX Super Resolution with 4 quality presets:

Performance

Balanced

Quality

Ultra Quality

CAS Shaders Integration: Embedded Contrast Adaptive Sharpening shaders to eliminate blurriness and drastically improve post-upscaling image clarity.

Full-Screen Quad Rendering: Implemented a highly optimized VAO/VBO/EBO pipeline tailored for full-screen upscaled blitting.

⚡ Revolutionary L3 Vertex Buffer Cache (12 Improvements, v2-v14) The vertex buffer caching subsystem has been completely re-architected to eliminate micro-stutters and minimize CPU-to-GPU overhead:

Enhanced Hashing: Replaced legacy FNV-1a with the ultra-fast xxHash64 algorithm to ensure more uniform distribution and prevent collision bottlenecks.

Predictive Caching: Integrated ML-driven prediction to forecast upcoming buffer allocations based on historical access patterns.

GPU-Side Buffers: Implemented persistent mapped buffers to enable fast, zero-copy data uploads directly to the GPU.

Delta Encoding: Introduced inter-frame data compression (storing only changes) to drastically reduce RAM/VRAM footprint.

Hierarchical Cache: Deployed a multi-tier (L1/L2/L3) cache structure that dynamically ranks data by access frequency.

Spatial Hashing & Compute Shaders:

Added spatial hashing based on buffer contents and world position.

Offloaded heavy hashing workloads to the GPU via Compute Shaders.

Parallelization & Filtering:

Implemented multi-threaded hashing to effortlessly handle massive data buffers.

Integrated a Bloom Filter for instantaneous tracking of missing entries (negative lookups).

Smart Memory Management:

Adaptive Sizing: Dynamic cache resizing on the fly based on runtime memory pressure.

Compression-Aware Eviction: Evicted L3 records are compressed before discarding to optimize storage overhead.

Prefetching: Proactively pre-loads vertex buffers by analyzing camera movement vectors and velocity.

🧹 Code Cleanup & Refactoring Removed GPU Detection: Stripped out redundant GPU detection routines since FSR is hardware-agnostic and runs out-of-the-box on all modern GPUs (NVIDIA, AMD, Intel).

Dropped Vulkan Backend: Removed experimental Vulkan support to fully double down on a pure, highly optimized OpenGL solution leveraging FSR.

📊 Configuration & Commands Added new console commands for dynamic, on-the-fly upscaling management:

/smoothgl set enableFSR <true/false> — Toggle FSR scaling.

/smoothgl set fsrQualityMode — Switch between quality modes (Performance, Balanced, Quality, Ultra Quality).

/smoothgl set fsrRenderScale — Manually fine-tune the internal rendering scale.

📝 Changelog (Version 0.2.0-Beta)

This is the first major step towards stable graphics optimization. In this version, we focused on eliminating micro-stutters during world generation and stabilizing peak FPS.

🔧 Graphics & Engine (Rendering & JIT)

• Added JIT Warmup: Implemented pre-warming for critical rendering and math classes during game startup. This completely eliminates stutters in the first few seconds after joining a world.
• Integrated Smart Cache: Added a custom cache manager to optimize native memory buffers, significantly reducing Garbage Collector (GC) overhead.

📊 Performance Monitoring & Networking

• Introduced Frame Profiler: Implemented an internal frame profiler that tracks performance drops in real-time to help smooth out the FPS frame time graph.
• Added Network Tuner: Optimized network packet processing, decreasing CPU load when playing on heavy multiplayer servers.

🧹 Code Cleanup & Stability

• Pure Java Optimization: Refactored the core architecture into pure Java, stripping away bloated runtime dependencies for maximum lightweight performance.
• Improved Mixin Stability: Reworked injection points within the Fabric loader to prevent crashes and ensure better compatibility with other mods.

0.1.0 (Initial Release)

Features

• Frame profiler with stutter detection and breakdown HUD (F7)
• Adaptive distance culling for entities, block entities, particles, and armor layers
• Occlusion culling via raycast with TTL cache for entities and particles
• GPU frame pacer with configurable frames-in-flight depth
• JIT warmup on startup to reduce early-game micro-stutters
• Direct-buffer pre-touch on startup (configurable size)
• GC watcher to monitor pause times
• Network tuner (TCP_NODELAY, larger buffers, IP_TOS_LOWDELAY)
• Packet pump for reduced input lag on high-FPS clients
• Ping monitor with HUD display
• In-world debug renderer (F9 toggle) for cull boundary and occlusion rays
• Custom telemetry graphs (frame-time bars + ping line) in HUD
• Startup phase timer with /smoothgl startup command

Commands

• /smoothgl - status overview
• /smoothgl breakdown - frame breakdown
• /smoothgl startup - startup timing report
• /smoothgl dump - stutter dump
• /smoothgl reset - reset profiler
• /smoothgl set <key> <value> - runtime config

Keybinds

• F7 - Toggle frame breakdown HUD

Configuration

• All settings in config/smoothgl.properties
• Runtime-configurable via commands