NVIDIA的Rubin CPX：大規模推斷的巨大飛躍？

NVIDIA的Rubin CPX處理器旨在通過大規模的文本處理，有希望的百萬句量表來徹底改變AI。這是大規模推斷的未來嗎？

NVIDIA's latest innovation, the Rubin CPX, is making waves in the AI world, promising to redefine the limits of large-scale inference. The buzz is all about its potential to handle massive-context processing, scaling up to a million tokens. Let's dive into what this means and why it matters.

NVIDIA的最新創新魯賓CPX正在AI世界中揮舞著波浪，承諾重新定義大規模推斷的極限。嗡嗡聲涉及其處理大規模處理處理的潛力，擴展到超過一百萬個令牌。讓我們深入了解這意味著什麼以及為什麼它至關重要。

The Rubin CPX: What's the Hype?

魯賓CPX：炒作是什麼？

Jensen Huang, NVIDIA's CEO, describes the Vera Rubin platform as a leap in AI computing. The Rubin CPX is designed for massive-context AI, enabling models to reason across millions of tokens of knowledge. This is a big deal because current large language models (LLMs) are often limited by their context windows. The Rubin CPX aims to change that.

NVIDIA首席執行官Jensen Huang將Vera Rubin平台描述為AI計算的飛躍。 Rubin CPX設計用於大規模的文本AI，使模型能夠跨越數百萬個知識。這很重要，因為當前的大型語言模型（LLM）通常受其上下文窗口的限制。魯賓CPX旨在改變這一點。

Understanding the Token Game

了解令牌遊戲

LLMs work by manipulating tokens, which are essentially the building blocks of language and data. The more tokens a model can process, the better it can understand and respond to complex queries. NVIDIA claims Rubin can scale LLMs and other generative AI models to context windows of up to a million tokens. This includes image and video generation models too.

LLM通過操縱令牌來工作，該令牌本質上是語言和數據的基礎。模型可以處理的代幣越多，它就可以理解並響應複雜的查詢越好。 NVIDIA聲稱Rubin可以將LLMS和其他生成A​​​​I模型擴展到上下文的窗口，最多可達一百萬個令牌。這也包括圖像和視頻生成模型。

Tech Specs and Performance

技術規格和性能

The Rubin CPX boasts up to 30 petaFLOPS of NVFP4 precision compute and includes 128GB of GDDR7 memory. NVIDIA envisions these boards being combined with Rubin GPUs and Vera CPUs in a fully-stocked rack implementation called the Vera Rubin NVL144 CPX. This setup includes 144 Rubin CPX GPUs, 144 plain Rubin CPUs, and 36 Vera CPUs, delivering a total of eight exaFLOPS of NVFP4 compute.

Rubin CPX擁有多達30個NVFP4 Precision Compute的PETAFlops，其中包括128GB的GDDR7內存。 NVIDIA設想這些板與魯賓GPU和Vera CPU相結合，在稱為Vera Rubin NVL144 CPX的全儲備機架實現中。該設置包括144個Rubin CPX GPU，144個普通Rubin CPU和36個Vera CPU，總共提供了八個Exaflops NVFP4 Compute。

The Million-Dollar Question: Profitability

百萬美元的問題：盈利能力

NVIDIA is making a bold claim: $100 million spent on Rubin-based hardware could potentially generate as much as $5 billion in revenue. That's a serious return on investment, if it pans out. It remains to be seen whether these claims are realistic.

Nvidia提出了大膽的主張：在魯賓的硬件上花費了1億美元，可能會產生多達50億美元的收入。如果它消失了，那將是一個嚴重的投資回報。這些主張是否現實還有待觀察。

Looking Ahead

展望未來

The Rubin CPX is expected to be available by the end of next year. While the price point is still under wraps, the potential impact on AI is undeniable. NVIDIA continues to be a bellwether for the AI and crypto markets, making its moves closely watched by investors and enthusiasts alike. The company's GPUs are increasingly vital as the AI boom continues, in contrast to Intel which, despite a recent surge following government investment, still lags behind NVIDIA's market cap.

魯賓CPX預計將在明年年底之前獲得。儘管價格仍在包裹中，但對AI的潛在影響是不可否認的。 Nvidia仍然是AI和加密貨幣市場的領先地位，使投資者和發燒友都密切關注其舉動。隨著AI繁榮的繼續，該公司的GPU越來越重要，與英特爾相比，儘管政府投資最近發生了激增，但仍落後於Nvidia的市值。

Final Thoughts

最後的想法

Is the Rubin CPX the future of large-scale inference? Only time will tell. But one thing is clear: NVIDIA is pushing the boundaries of what's possible in AI. Get ready for a wild ride!

魯賓CPX是大規模推斷的未來嗎？只有時間會證明。但是有一件事很清楚：Nvidia正在推動AI中可能的界限。準備狂野！