NVIDIA RUBIN CPX：用數百萬個上下文窗口革新AI

NVIDIA的Rubin CPX GPU設置為重新定義AI推斷，其能夠處理數百萬個上下文窗口，從而改變軟件開發和視頻生成。

NVIDIA is pushing the boundaries of AI with its Rubin CPX GPU, designed for massive-context processing. This innovation promises to significantly enhance AI's ability to understand and generate complex content, particularly in areas like software development and video creation.

NVIDIA用其Rubin CPX GPU推動了AI的邊界，該邊界設計用於大規模的秘密處理。這項創新有望顯著增強AI理解和生成複雜內容的能力，尤其是在軟件開發和視頻創建等領域。

The Dawn of Massive-Context AI

大規模的黎明AI

The current AI boom is fueled by large language models (LLMs) that manipulate tokens. The more tokens an LLM can process, the better its understanding and output. However, computational limitations restrict most models to smaller context windows. Enter NVIDIA's Rubin CPX, designed to handle context windows of up to a million tokens.

當前的AI繁榮由操縱令牌的大型語言模型（LLM）助長。 LLM的令牌越多，其理解和輸出就越好。但是，計算限制將大多數模型限制在較小的上下文窗口中。輸入NVIDIA的Rubin CPX，旨在處理最多一百萬個令牌的上下文窗口。

Rubin CPX: A Game Changer

魯賓CPX：改變遊戲規則的人

According to NVIDIA, the Rubin CPX delivers up to 30 petaFLOPS of NVFP4 precision compute and includes 128GB of GDDR7 memory. This power allows AI systems to comprehend and optimize large-scale software projects and generative video applications with groundbreaking speed and accuracy.

根據NVIDIA的說法，Rubin CPX最多可提供30個NVFP4 Precision Compute的PETAFLOP，並包括128GB的GDDR7內存。這種功能使AI系統能夠以開創性的速度和準確性理解和優化大型軟件項目和生成視頻應用程序。

Disaggregated Inference: The Key to Efficiency

分解推理：效率的關鍵

The Rubin CPX utilizes a disaggregated infrastructure approach, separating the context and generation phases of inference. This allows for targeted optimization of resources, enhancing throughput, reducing latency, and improving overall resource utilization. The context phase, which is compute-bound, benefits from high-throughput processing, while the generation phase relies on fast memory transfers.

Rubin CPX利用了分解的基礎架構方法，將推理的上下文和生成階段分開。這允許對資源進行有針對性的優化，增強吞吐量，減少延遲並改善整體資源利用率。上下文階段是計算結合的，受益於高通量處理，而生成階段則依賴於快速內存傳輸。

The Vera Rubin NVL144 CPX Rack

Vera Rubin NVL144 CPX機架

NVIDIA envisions the Rubin CPX being combined with other components to create powerful AI systems. The Vera Rubin NVL144 CPX rack, for example, combines 144 Rubin CPX GPUs, 144 Rubin GPUs, and 36 Vera CPUs, delivering 8 exaFLOPS of NVFP4 compute. This robust system promises a substantial return on investment, potentially generating billions in revenue from a $100 million investment.

NVIDIA設想魯賓CPX與其他組件結合使用，以創建強大的AI系統。例如，Vera Rubin NVL144 CPX機架結合了144個Rubin CPX GPU，144 Rubin GPU和36 Vera CPU，提供8個Exaflops NVFP4計算。這種強大的系統承諾將獲得可觀的投資回報，從1億美元的投資中產生了數十億美元的收入。

A New Era for AI Infrastructure

AI基礎設施的新時代

NVIDIA's Rubin CPX is setting new standards in AI infrastructure economics. By integrating disaggregated infrastructure with advanced orchestration through the NVIDIA Dynamo platform, the Rubin CPX paves the way for more sophisticated AI systems capable of handling the most demanding inference tasks. This innovation not only enhances AI capabilities but also sets a new benchmark for future developments in generative AI applications.

NVIDIA的Rubin CPX正在製定AI基礎設施經濟學的新標準。通過通過NVIDIA Dynamo平台將分解的基礎架構與高級編排集成在一起，Rubin CPX為能夠處理最苛刻的推理任務的更複雜的AI系統鋪平了道路。這項創新不僅增強了AI功能，而且為生成AI應用程序的未來發展樹立了新的基準。

Looking Ahead

展望未來

With the Rubin CPX, NVIDIA isn't just launching a product; they're ushering in a new era of AI capabilities. Imagine AI assistants that truly understand the nuances of your code or video generation tools that produce stunningly realistic content. The possibilities are endless, and it's going to be a wild ride to see where this technology takes us! Expect the Rubin CPX to be available by the end of next year, and prepare to be amazed.

借助Rubin CPX，Nvidia不僅僅是推出產品。他們在AI功能的新時代迎來了。想像一下，真正了解您的代碼或視頻生成工具的細微差別的AI助手會產生驚人的現實內容。可能性是無窮無盡的，看看這項技術將我們帶到哪裡，這將是一次瘋狂的旅程！期望魯賓CPX在明年年底之前提供，並準備驚訝。