0.7.5-alpha Highlights

• Added the terrain engine to the Vulkan renderer — batched draw submission now runs in Vulkan-native indirect form, with draw commands living on the GPU and re-issued in a single call when nothing changed
  • With it, Onyx now beats vanilla-Vulkan in ALL six benchmark scenes — up to +752 FPS in varied builds, +69% in mob crowds, and the last vanilla-led scene (caves) flipped to an Onyx win
• Added compact 16-byte terrain vertices on Vulkan — median-verified gains of +9% in open terrain, +12% in varied builds, and +15% at max render vistas, on top of Vulkan already being ~2x OpenGL
• Updated the terrain engine to pick the right draw-submission form for the graphics device automatically
• Added experimental chunk-update pacing ("uploadPacing" in the config, off by default) — spreads big chunk-loading bursts across a few frames instead of letting them land in one
  • When a batch of chunks arrives, up to 200 mesh rebuilds could hit a single frame — that one long frame is the stutter you feel even at high FPS; pacing caps acceptance at 32 per frame and the rest simply wait a frame or two
  • Measured 25% faster worst-frames in the chunk-loading flight test, with block-breaking always exempt so your own edits never wait
• Fixed the terrain engine wasting render time in small scenes — its per-frame lookup table was being wiped at full size every frame, so a cave paid the cost of a mountain vista
  • Found by profiling the one scene vanilla still led; the fix turned that scene into a win and made every other scene faster too
• Updated the F3 panel's GPU and CPU meters to show steady rolling averages instead of fast-flickering snapshots — the number now means "what you're actually using"
• Added a frame time line to the F3 panel — your current and 10-minute-average frame time in milliseconds, computed from the exact same measurements as the FPS lines so the two can never disagree
• Updated the F3 TPS line to label its milliseconds number "tick" — it is the server's time to simulate one game tick (50ms is the budget), not a frame time
• Updated the benchmark methodology (1.2) — random ticks frozen and stray items purged in test scenes, so every published number is now a median of deterministic, repeatable runs