來源Computerworld•較早收集於 49m
Intel 憑藉 Core Ultra Series 3 重返機器人市場
💡Intel 憑藉邊緣優化晶片重返機器人領域，可能重新定義我們在實體機器人中部署本地 AI 代理的方式。
⚡ 30 秒速覽
有什麼變化
Intel Core Ultra Series 3 處理器目前已整合至 130 種邊緣 AI 與機器人設計中。
為什麼重要
此舉標誌著邊緣運算的重大轉變，透過消除對雲端的依賴，有望降低機器人系統的延遲。這使 Intel 成為下一代自主人形機器人的關鍵硬體供應商。
下一步行動
如果您正在開發需要本地低延遲 AI 推論的邊緣機器人應用，請評估 Intel Core Ultra Series 3 開發套件。
誰應關注：Developers & AI Engineers
關鍵要點
- •Intel Core Ultra Series 3 處理器目前已整合至 130 種邊緣 AI 與機器人設計中。
- •新型晶片架構可將圖形、運動與控制功能整合至單一矽晶片中。
- •策略重心在於「邊緣 AI」，使設備能在本地運行複雜 AI 模型，無需依賴雲端處理。
- •Intel 正瞄準預計於 2050 年達到 5 兆美元規模的人形與工業機器人市場。
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 13 個來源。
🔑 增強重點摘要
- •Intel's Core Ultra Series 3 processors integrate a CPU, GPU, and NPU into a single system-on-chip (SoC), enabling real-time, on-device processing for complex AI workloads like language, vision, reasoning, and motion control, thereby reducing reliance on discrete GPUs and cloud infrastructure.
- •To facilitate robotics development, Intel has introduced OpenVINO Physical AI, an open-source framework designed to simplify the deployment and scaling of physical AI models by providing a unified, open, and scalable path from AI experimentation to production-grade robots.
- •The new architecture supports multi-agent Physical AI systems, as demonstrated by Sensory AI's Ella robot barista, which runs three specialized AI agents (Avatar, Guardian, and Ella Agent) concurrently on a single Core Ultra Series 3 SoC for customer interaction, system operations, and business intelligence.
- •The Core Ultra Series 3 processors are built on Intel's 18A manufacturing process, representing the company's most advanced node to date and a key step in its strategy to bring leading-edge chip production back in-house.
- •Intel is positioning its Core Ultra Series 3 as a cost-effective alternative to discrete GPUs, claiming competitive performance against NVIDIA Jetson AGX Orin and Jetson Thor T5000 modules, with some benchmarks showing roughly half the system cost for certain vision-language-action (VLA) model workloads.
📊 競品分析▸ Show
| Feature/Metric | Intel Core Ultra Series 3 (e.g., X7 358H) | NVIDIA Jetson AGX Orin 64GB | NVIDIA Jetson Thor T5000 | NVIDIA RTX 4060/5080 (discrete GPUs) |
|---|---|---|---|---|
| Total Platform TOPS | Up to 180 TOPS (select H-SKUs) | 275 TOPS | N/A (implied higher than Orin) | N/A (focus on discrete GPU performance) |
| LLM Performance | Up to 1.9x higher than Jetson AGX Orin | Baseline | N/A | N/A |
| Image Classification | Up to 1.7x higher than Jetson AGX Orin | Baseline | N/A | N/A |
| Performance per Watt per $ (End-to-End Video Analytics) | Up to 2.3x better than Jetson AGX Orin | Baseline | N/A | N/A |
| Vision Language Action (VLA) Throughput | Up to 4.5x higher than Jetson AGX Orin | Baseline | Competitive with Thor on medium-sized VLA models | N/A |
| TTS (Text-to-Speech) Performance | 3.7x faster than Jetson AGX Orin | Baseline | N/A | N/A |
| Multi-task Vision Performance | 5.4x faster than Jetson AGX Orin (at 25W) | Baseline | N/A | N/A |
| Policy Throughput (LeRobot pipeline) | 4.9x to 8.6x faster than Jetson AGX Orin | Baseline | N/A | N/A |
| Throughput (vs. Jetson Nano) | Nearly 9x faster | N/A | N/A | N/A |
| Latency (3-camera Pi0.5 Droid workload) | Lower than Jetson AGX Orin | Baseline | N/A | N/A |
| Power Efficiency (vs. RTX 4060 GPU) | 2.7x higher | N/A | N/A | Baseline |
| Time-to-First-Token (vs. RTX 5080 GPU) | Nearly 2x faster (at 22W vs 90W) | N/A | N/A | Baseline (at 90W) |
| System Cost | Roughly half the system cost of Jetson Thor T5000 (for VLA models) | N/A | Higher | Higher (for discrete GPUs) |
| Architecture | Integrated CPU, GPU, NPU (SoC) | Dedicated AI accelerators (SoC) | Dedicated AI accelerators (SoC) | Discrete GPU |
🛠️ 技術深入
- The Core Ultra Series 3 processors integrate a CPU, GPU (Intel Arc graphics), and NPU into a single system-on-chip (SoC) for heterogeneous computing.
- Select H-SKUs of Core Ultra Series 3 can achieve up to 180 total platform TOPS (Tera Operations Per Second).
- The integrated GPU features up to 12 Xe cores, offering up to 50% more GPU cores than the previous generation for AI-intensive edge workloads.
- A dedicated NPU within the Core Ultra Series 3 is rated at 50 TOPS.
- The architecture supports up to 64GB of LPDDR5 8533 or up to 128GB of DDR5 7200 memory, providing flexibility for large AI workloads compared to dGPUs with fixed VRAM.
- Power consumption for these processors can be as low as 15W for fanless designs and up to 65W.
- Core Ultra Series 3 processors are built on Intel's 18A manufacturing process, which is the company's most advanced node to date.
- They include industrial-grade features such as support for Functional Safety, Intel® TCC (Time Coordinated Computing) for real-time performance, and extended temperature support ranging from -40°C to +100°C.
- Select SKUs are designed for long-term durability, ensuring 100% operation over 10 years in industrial conditions.
- Connectivity options include Thunderbolt 4 and integrated Wi-Fi 7 R2.
- Intel Application Energy Telemetry (AET) is supported for real-time workload-level energy monitoring.
🔮 前景展望基於引用來源的 AI 分析
Intel's unified hardware-software stack, including Core Ultra Series 3 and OpenVINO Physical AI, will significantly accelerate the commercial deployment of diverse robotic fleets.
By simplifying development, reducing software complexity, and lowering total cost of ownership, Intel addresses key barriers to scaling robotics solutions across various industries.
The shift to integrated CPU/GPU/NPU architecture will lead to a broader adoption of 'agentic AI' in physical robots, moving away from a reliance on bulky and expensive discrete GPUs.
The Core Ultra Series 3's ability to run multiple specialized AI agents concurrently on a single SoC, with lower power consumption and reduced cost, makes advanced on-device AI more feasible for commercial applications.
⏳ 時間線
2023-07
Intel Capital invests $9 million in humanoid robot company Figure.
2025-07
Intel spins off its AI robotics unit RealSense into a standalone business with $50 million in Series A funding.
2025-11
Intel discusses its Robotics AI Suite and strategy for converging real-time control and AI perception on singular computing platforms.
2026-01
Intel introduces Core Ultra Series 3 processors at CES 2026, built on the 18A process, extending the platform beyond PCs to embedded and industrial use cases like robotics.
2026-05
Intel highlights Core Ultra Series 3 as the new standard for edge AI robotics compute, with examples like the Ella barista robot fully transitioning to Intel architecture.
2026-06
Intel announces over 130 design wins for its Series 3 processors in edge AI and robotics and launches the OpenVINO Physical AI framework at Computex 2026.
📎 來源 (13)
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： Computerworld ↗
每週電子報
每週一封，可隨時退訂。