來源The Verge•較早收集於 15m
PlayStation 導入 AI 畫面生成技術

#frame-generation#ai-graphics#gaming-ai#ml-inferencesony-playstationsonyplaystation-5ps5-promark-cernypssr
💡Sony AI 畫面生成登陸主機—圖形 ML 開發者關注遊戲應用。(28字)
⚡ 30 秒速覽
有什麼變化
未來 PlayStation 平台將支援機器學習畫面生成
為什麼重要
推動 AI 融入遊戲主機,提升圖形效能並與 PC/Nvidia 競爭。顯示遊戲硬體廣泛採用機器學習。
下一步行動
在 Unreal Engine 中測試 ML 畫面生成與類 PSSR 上採樣,用於主機原型。
誰應關注:Developers & AI Engineers
關鍵要點
- •未來 PlayStation 平台將支援機器學習畫面生成
- •AI 在渲染畫面間生成「想像」畫面,提升流暢度
- •PS5 Pro 首席架構師 Mark Cerny 向 Digital Foundry 宣布
- •基於 PS5 Pro 升級 PSSR AI 上採樣;類 PC 技術可能增延遲
🧠 深度解析
背景與延伸:來自公開資料,非原文內容。引用 12 個來源。
🔑 增強重點摘要
- •The technology is the primary output of 'Project Amethyst,' a multi-year co-engineering initiative between Sony Interactive Entertainment and AMD to unify PlayStation's PSSR with AMD's 'FSR Redstone' (formerly FSR 4) architecture.
- •Mark Cerny confirmed the PS5 Pro's custom machine learning silicon delivers 300 8-bit TOPS (Tera Operations Per Second) of performance, specifically designed to handle the 'full-fat' version of AI frame generation without the sparsity-based compromises seen in some PC implementations.
- •To combat the inherent latency of 'fake frames,' Sony filed a 2025 patent for a 'Timed Input/Action Release' system that uses a predictive AI model to anticipate player button presses before they are completed.
- •The frame generation suite is part of a broader 'AI Trinity' that includes upscaling and 'Ray Regeneration,' the latter of which uses ML to denoise ray-traced reflections and shadows more efficiently than traditional hand-tuned denoisers.
- •Sony is implementing a system-level 'Enhance PSSR Image Quality' toggle in the PlayStation 5 Pro settings, allowing the AI model to be updated via firmware to improve all supported games simultaneously without requiring individual developer patches.
📊 競品分析▸ Show
| Feature | PlayStation PSSR (Project Amethyst) | NVIDIA DLSS 4/5 | AMD FSR Redstone |
|---|---|---|---|
| Core Architecture | Custom Sony ML Silicon (300 TOPS) | Tensor Cores (RTX 40/50 Series) | AI-based (RDNA 4/5 Exclusive) |
| Frame Generation | ML-based (Coming 2026+) | Optical Multi-Frame Generation | Fluid Motion Frames (AI-assisted) |
| Latency Tech | AI Input Prediction (Patented) | NVIDIA Reflex | Radeon Anti-Lag 2/3 |
| Implementation | System-level & Game-specific | Game-specific (SDK) | Driver-level & Game-specific |
| Platform Cost | Included in $750 PS5 Pro | $599 - $1,599+ (GPU only) | $499 - $999+ (GPU only) |
🛠️ 技術深入
The implementation relies on several key architectural pillars identified in recent technical disclosures:
- Compute Density: Utilizes the PS5 Pro's 300 TOPS 8-bit ML capability to run inference passes in under 2ms per frame.
- Algorithm Parity: Shares a neural network training recipe with AMD's FSR Redstone, moving away from the spatial-temporal heuristics of FSR 3 toward a pure Convolutional Neural Network (CNN) approach.
- Input Prediction: Integrates a 'Frame Warp' technique that modifies the current frame based on predicted player movement to ensure visual responsiveness matches the increased frame rate.
- Memory Management: Optimized for the PS5 Pro's 576 GB/s memory bandwidth to handle the high-resolution buffers required for 4K AI reconstruction and frame interpolation simultaneously.
🔮 前景展望基於引用來源的 AI 分析
120fps will become the standard 'Fidelity Mode' target
AI frame generation allows developers to render at a 60fps base with high-fidelity assets while outputting a smooth 120Hz signal, effectively doubling perceived performance.
PlayStation 6 will likely feature a dedicated NPU
The shift toward simultaneous upscaling, frame generation, and ray regeneration will require more specialized AI silicon than the current hybrid GPU/ML approach used in the PS5 Pro.
⏳ 時間線
2020-11
PlayStation 5 launches with standard AMD FSR support
2023-06
Sony and AMD initiate 'Project Amethyst' collaboration
2024-11
PS5 Pro launches introducing PSSR AI upscaling
2025-01
Sony files patent for AI-driven input prediction to reduce lag
2026-02
Upgraded PSSR 2.0 launches with Resident Evil Requiem
2026-03
Mark Cerny confirms ML Frame Generation for future platforms
📎 來源 (12)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源: The Verge ↗
每週電子報
每週一封,可隨時退訂。

