來源IT之家•較早收集於 4h
Nvidia Vera CPU 將採用 SK Hynix 記憶體產品

💡Nvidia 專為 AI 智能體打造的 Vera CPU 將採用 SK Hynix 記憶體,了解此合作對 AI 硬體效能的影響。
⚡ 30 秒速覽
有什麼變化
Vera CPU 是一款專為 AI 智能體打造的革命性處理器。
為什麼重要
將高性能記憶體整合至 Vera CPU,凸顯了記憶體頻寬對於擴展 AI 智能體能力及降低複雜工作負載延遲的關鍵作用。
下一步行動
評估你的 AI 智能體架構,確認是否能利用基於 Vera 的系統所提供的專用計算與記憶體吞吐能力。
誰應關注:Developers & AI Engineers
關鍵要點
- •Vera CPU 是一款專為 AI 智能體打造的革命性處理器。
- •Nvidia 與 SK Hynix 加深合作,確保未來一年的記憶體供應。
- •Vera CPU 將驅動各行各業的多樣化 AI 工作負載。
- •Nvidia 與 SK 集團達成戰略合作,支持高性能 AI 硬體發展。
🧠 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 Vera CPU features 88 custom NVIDIA "Olympus" cores, designed for control-heavy, latency-sensitive agentic AI and reinforcement learning workloads, and utilizes Spatial Multithreading to create 176 threads with partitioned core resources.
- •Nvidia's Vera CPU delivers up to 1.2 terabytes per second (TB/s) of LPDDR5X memory bandwidth and supports up to 1.5 TB of memory capacity, offering twice the bandwidth at half the power of traditional CPU memory.
- •The Vera CPU is engineered to accelerate agentic sandbox performance by up to 1.8 times compared to x86 CPUs, specifically optimizing tasks such as code execution, tool calls, and evaluation loops within AI agent reasoning loops.
- •Vera represents Nvidia's first standalone data center microprocessor, directly challenging established server CPU providers like Intel's Xeon, AMD's EPYC, and custom chips from major cloud operators such as Amazon's Graviton.
- •The partnership with SK Hynix for Vera's DRAM is expected to lead to substantial growth in business between the two companies through the second half of 2026 and into 2027, reinforcing SK Hynix's critical role in the AI hardware supply chain.
📊 競品分析▸ Show
A Markdown table comparing this with competitors (Feature/Pricing/Benchmarks). Return null if not applicable (e.g. op-ed, interview, single-product announcement with no clear competitors).
🛠️ 技術深入
- Cores: 88 custom NVIDIA Olympus cores.
- Architecture: Based on a custom Arm CPU core, compatible with Arm v9.2 ISA.
- Fabric: Second-generation NVIDIA Scalable Coherency Fabric (SCF) connects all cores, cache, memory, I/O, and NVLink-C2C across a single compute die, providing 3.4 TB/s of bisectional bandwidth.
- Memory: LPDDR5X memory subsystem, offering up to 1.2 TB/s bandwidth and up to 1.5 TB capacity.
- Multithreading: Spatial Multithreading enables 176 threads with partitioned core resources for predictable throughput.
- Interconnect: NVLink-C2C delivers up to 1.8 TB/s of coherent bandwidth between Vera CPUs and NVIDIA GPUs (e.g., Rubin GPUs).
- Manufacturing Process: Built on a 3nm process.
- Power: Configurable Thermal Design Power (TDP) range of 250 W to 450 W.
- Packaging: Monolithic compute die with memory and I/O disaggregated into adjacent chiplets via CoWoS-R packaging.
🔮 前景展望基於引用來源的 AI 分析
Nvidia will significantly expand its market share in the data center CPU segment.
The Vera CPU is positioned as Nvidia's first standalone data center CPU, directly challenging established x86 players and custom cloud chips, indicating a strategic push beyond GPUs into a broader compute market.
The demand for specialized CPUs optimized for AI agent workloads will accelerate industry-wide innovation in processor design.
Vera's focus on agentic AI and reinforcement learning highlights a growing need for CPUs that can efficiently handle latency-sensitive, control-heavy tasks, pushing other manufacturers to adapt their designs.
SK Hynix's role as a critical memory supplier for high-performance AI hardware will be further solidified.
The deepened partnership and anticipated substantial growth in business with Nvidia for Vera DRAM indicate SK Hynix's strategic importance in the evolving AI supply chain, particularly for high-bandwidth, power-efficient memory.
⏳ 時間線
1993-04
Nvidia founded by Jensen Huang, Chris Malachowsky, and Curtis Priem.
2021-04
Nvidia announces Grace CPU, its first data center CPU, designed for HPC and AI workloads.
2022-03
Nvidia announces Grace CPU Superchip, featuring 144 cores and 1TB/s memory bandwidth.
2023
Nvidia Grace CPU becomes generally available.
2026-03-16
Nvidia officially launches Vera CPU, purpose-built for agentic AI, and announces it is in full production.
2026-06-07
Nvidia CEO Jensen Huang confirms Vera CPU will utilize SK Hynix DRAM, deepening the partnership.
📎 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: IT之家 ↗
每週電子報
每週一封,可隨時退訂。