zircon-1.0.4.jar
Curse Maven Snippet
What's new
⚡ Highlights
- Shader Mode. Zircon now recognises an active shader pack, adds no GPU work of its own, and keeps its own work out of the way of the shader pipeline.
- GPU Governor (optional, off by default). Keeps heavy shader packs smooth by nudging a handful of render-only settings only when your GPU is actually running late, then putting them back. Fully reversible, with one-click restore.
- Steady Mode. Frames are delivered at even intervals instead of bouncing between fast and slow, with a live stability score.
- Still Life. Entities that haven't changed since last frame, like item frames, armor stands and resting mobs, are no longer re-processed every frame.
- Parallel extraction. The heaviest per-entity work in big bases can now run across your CPU cores.
- "Why did that stutter?" Every hitch is classified with the evidence behind it: garbage collection, upload burst, shader compile, disk, lock contention and more.
- Smoother joins. Zircon stays out of the way while the world loads, so joining a world or changing dimension spikes less.
- Iris compatibility fixed. Zircon no longer silently turns its Iris support off when Iris updates.
🎮 Rendering & FPS
- Still Life (retained entity states): unchanged entities reuse their previous render state instead of being rebuilt each frame.
- Only renderers that passed an automated input-by-input check take part. Unknown and modded renderers are left alone.
- Verified pixel-identical against vanilla across every vanilla entity type, standing still and moving.
- Parallel render-state extraction: for renderers proven safe, Zircon prepares entity and block-entity render states on background threads.
- The game never waits on a late worker. Anything not ready in time is simply done the normal way.
- If anything goes wrong on a worker, Zircon switches the feature off and the game carries on as vanilla.
- Block-update coalescing: redstone clocks, piston farms and flowing fluids no longer trigger repeated rebuilds of the same chunk section within one tick.
- Text and nametag caching: laid-out sign text and entity nametags are reused until they change. Big sign walls and name-tagged herds cost far less.
- Instanced drawing (where supported): repeated identical objects such as chests, beds and banners can be drawn together on graphics backends that support it. Never used while a shader pack is active.
- Persistent upload staging (where supported): Zircon's own small uploads reuse persistently mapped buffers on backends that offer it. Zircon says "unavailable" where it isn't.
- Depth-readback occlusion (experimental, off by default): uses the previous frame's depth to hide objects that are fully buried behind solid geometry. It is deliberately conservative, one-way only, and never runs with shaders.
- Menu-cover throttle (optional, off by default): while an opaque full-screen menu hides the world, the world behind it can render less often.
- Particle level of detail (optional, off by default): distant particles can be thinned out. It is a visual trade-off, so it lives on the Visual workload page and no preset turns it on.
🧠 CPU & multi-core
- Work-stealing scheduler: Zircon's worker pool now uses per-thread queues with work stealing, so threads spend less time waiting on each other. Priorities and cancellation behave exactly as before.
- Smarter frame guard: Zircon predicts how long the next frame will take and moves its own work to frames with spare time.
- Join and transition smoothing: for a short window after joining a world or changing dimension, Zircon postpones its own maintenance and lowers its budgets, so it adds nothing to the chunk flood.
- Autopilot 3 (modpack fingerprint): Zircon remembers what it learned per modpack. When your mod list changes, it keeps what still applies and drops what doesn't.
🖥️ Shaders & GPU
- Shader Mode: Zircon detects an active shader pack and switches to a GPU-protect policy. It adds no discretionary GPU work, keeps its overlay cheap and defers its own maintenance to frames with slack.
- GPU Governor (new, optional):
- Off by default, with a clear toggle on the new GPU page.
- Only ever moves render-only options: entity distance, cloud range and type, weather radius, entity shadows, particle level, ambient occlusion and improved transparency. Options that could reload chunks or resources are excluded from automatic changes.
- Never touches render distance, simulation distance, FOV, resolution or your shader pack.
- Steps down quickly when the GPU runs late, recovers slowly, never oscillates, and respects floors you set.
- Every change is logged in the Monitor with its measured reason. Restore my settings is always one click away, and your original values are saved to disk before the first change, so they come back after a crash too.
- Needs real GPU timing to act. If timing isn't available, it does nothing.
- Multi-timestamp GPU timing: Zircon now uses multiple GPU timestamps per frame. In the vanilla pipeline it shows GPU time for major phases. With a shader pack active, Iris owns the pipeline, so Zircon reports whole-frame GPU time only and says so. It never invents per-pass numbers.
- Shader Advisor: measures how much GPU time each render-only setting costs on your machine and shows a table you can apply from. Settings inside the shader pack itself can't be changed by Zircon, so a button opens Iris's own screen instead.
- Shader-aware frame pacing: with a pack on and the GPU as the limit, Zircon picks its pacing from your measured GPU frame times, so frames come out evenly instead of jittering.
- Shader-safe culling, verified: culling and retention were checked with shader packs on, with paired screenshots and zero differing pixels, including shadows cast from objects just off-screen.
- Vulkan completed: the full Vulkan backend run, including GPU timing, fences and every new feature's detection path.
📈 Smooth frame delivery
- Steady Mode: picks a frame cap just below what your machine holds at the 95th percentile and delivers frames at even intervals. It never raises your cap and never overrides a lower one.
- Stability score: shows the share of frames within ±10% of the median and your worst-frame ratio. This is Zircon's own measurement of frame times, not input latency.
- Fence-bounded render-ahead: measures how many frames the CPU is running ahead of the GPU, and in Steady Mode can bound that queue. Reports the queue depth in frames and never a made-up latency figure.
- Garbage collection at safe moments: Zircon only asks for a collection during loading screens, the pause menu, dimension changes and after saves, and only when memory pressure is building. Never during active play. Turned off automatically for concurrent collectors where it doesn't help.
- Hitch library: every stutter is classified with the evidence used, and "unknown" is a valid answer. The Monitor lists your latest hitches with cause, duration and a suggested action.
💾 Memory, GC & Java
- Where is my RAM? On request, Zircon samples a class histogram and attributes memory to the mods that own it. It shows the cost of taking it, since it can briefly pause Java.
- Native and direct memory watch: tracks direct and mapped buffers and, where available, GPU allocation totals. Warns on steady growth after leaving a world.
- Texture memory advisor: estimates texture memory per mod and resource pack, clearly labelled as an estimate, and highlights the biggest offenders.
- GC Lab: compares your own sessions locally, so it can show whether a collector or flag set gave you fewer hitches on this machine. Nothing leaves your computer.
- Heap headroom prediction: warns early when memory is trending toward GC thrash, with a plain-language cause.
- AOT cache helper: generates the exact flags for Java's startup cache and tells you when the cache is stale because your mod list changed. Zircon never edits your launcher.
- More leak protection: dimension changes, resource reloads, screen closes and profile switches are now checked for leaks the same way world exits are.
🔍 Intelligence & stability
- "Why is my FPS low?": one Monitor screen ranks what to look at, combining bottleneck detection, hitch causes, memory state, shader state and the top mods by cost. Each item is tagged Measured or Inferred.
- Regression detector: notices when this session is clearly worse than your own recent baseline and points to the most likely correlated change: a new mod, driver, setting or Zircon module. Reported as a correlation, not proof.
- Auto-bisect (optional): a guided test that toggles Zircon's own modules one at a time and offers to disable any that hurt your machine. It never touches other mods or your graphics settings.
- Mixin conflict detector: scans other mods' mixin targets (read-only) and shows overlaps with Zircon on the Compatibility page, along with what Zircon did about each.
- Shareable report: one button writes a support report with hardware class, versions, mod list, settings, frame-time statistics and Zircon's module states. No personal data, server addresses or world names, and you see exactly what it contains before saving.
🐛 Bug fixes
- Fixed: Zircon turned its Iris support off whenever your Iris version differed from the one it was tested with, so every Iris update silently disabled shader awareness. Zircon now checks what Iris actually supports instead of its version number.
- Fixed: Iris state was queried through a slower, generic call every shader pass.
- Fixed: shader-pass culling safety had only been verified in tests. It is now verified with real screenshots under shader packs.
- Fixed: Vulkan backend behaviour had not been verified end to end.
- Fixed: repeated chunk-section invalidation from redstone, pistons and fluids.
- Fixed: possible memory held after dimension changes, resource reloads, screen closes and profile switches.
- Fixed: hostile, truncated or corrupt config and profile files are now handled safely, and old configs from every previous schema migrate cleanly.
- Hardened: an 8-hour soak test, repeated world joins and leaves, window resizes, shader pack switching while in a world, and dedicated-server joins. Zircon stays fully inert on dedicated servers.
🤝 Compatibility
Zircon still steps aside before loading any feature another mod already handles, so nothing conflicts or runs twice. New in 1.0.4:
- Retained extraction, parallel extraction, instanced drawing, depth-readback occlusion, text caching and the menu throttle each detect mods that own those systems and stand down for them.
- Iris and Sodium are never reached into. Zircon only uses their published APIs, and terrain-side features stay out of the way of Sodium.
- Instanced drawing and depth-readback occlusion never run while a shader pack is active.
- Tested alongside Sodium, Lithium, FerriteCore, ImmediatelyFast, Entity Culling, More Culling, C2ME, Iris, Distant Horizons, Dynamic FPS and spark.
- Shader shadow, reflection and portal passes are never culled using the main camera's view.
⚙️ Settings & interface
- New pages: GPU and Memory & Java additions, plus new Monitor views for hitches, stability and diagnosis.
- New Monitor tools: hitch library, stability score, "Why is my FPS low?", Shader Advisor, class histogram and shareable report.
- Config schema 4: your existing settings carry over automatically. New options start at their defaults, and anything that changes your visuals starts off.
- Interface: all new screens were checked at GUI scales 1–4, with full narration support.
Zircon only claims what it measures. If something looks wrong, please use Record performance problem in Zircon's diagnostics and attach the report to your issue.
All Relations
- All Relations
- Embedded Library
- Optional Dependency
- Required Dependency
- Tool
- Incompatible
- Include