Вони купу оптимізацій зробили, щоб не провокувати статери.
Recent updates to
OptiScaler focus heavily on lower overhead, enhanced frame generation efficiency, and broader architecture support.
- DirectX 12 Upscaler Stability: Improved DirectX 12 hook handling significantly reduces frame pacing jitter and eliminates random ResizeBuffers1 crashes.
- DLSS Input Overhead Fixes: Fixed subtle thread-blocking issues where selecting DLSS inputs on non-NVIDIA GPUs caused micro-stutters.
- спойлер
-
Recent updates to OptiScaler focus heavily on lower overhead, enhanced frame generation efficiency, and broader architecture support.
Engine & Back-End Optimizations
* DirectX 12 Upscaler Stability: Improved DirectX 12 hook handling significantly reduces frame pacing jitter and eliminates random `ResizeBuffers1` crashes.
* DLSS Input Overhead Fixes: Fixed subtle thread-blocking issues where selecting DLSS inputs on non-NVIDIA GPUs caused micro-stutters.
* Resource Loading & Cache Cleanup: Streamlined memory allocation during initial upscaler hook setup, speeding up game launch times and reducing initial frame-drop spikes.
Frame Generation (FG) Improvements
* NukemFG & OptiFG Overhaul: Integrated native swapchains rather than relying entirely on Streamline hooks. This lowers frame delivery latency and delivers noticeably smoother frame-time pacing.
* FlipMetering Control: Added automated toggles for `FlipMetering` via FakeNVAPI. Disabling this prevents erratic frametime spikes when running DLSS frame generation inputs alongside non-NVIDIA GPUs or custom frame generators.
* Latency Reduction Integration: Deeper bundling with FakeNVAPI allows native injection of AMD Anti-Lag 2 (Vulkan/DX12) and Intel XeLL to offset frame generation latency penalties.
Next-Gen Upscaling & GPU Architecture Enhancements
* FSR 4 INT8 Optimization (RDNA 2 & Older Architectures): Includes dedicated optimizations for running INT8 FSR 4 models on RDNA 2 (RX 6000 series) and older GPUs. This reduces the typical 10–20% performance regression seen when running heavy upscalers on non-matrix hardware and eliminates motion ghosting.
* Smart Architecture Auto-Selection: Automatically detects GPU capabilities to pair the most performant model (e.g., auto-enabling `Fsr4Update` on RDNA 3 dGPUs while applying optimized INT8 fallbacks for unsupported hardware).
Sharpening & Post-Processing Speed
* Depth-Aware RCAS: Sharpening logic now features edge-aware masking options, preventing shimmering artifacts without adding notable GPU pass overhead.
* Precompiled Shaders: Native precompiled shader passes for RCAS and Output Scaling reduce shader compilation stutters when upscaling ratios change dynamically.