Nvidia представила революционный чип с искусственным интеллектом после приобретения Groq за 20 миллиардов долларов.

In a strategic move that echoes with foresight and innovation, Nvidia (NVDA 0.55%) has demonstrated the value of its $20 billion acquisition of AI start-up Groq by unveiling the groundbreaking Groq 3 LPX inference accelerator. Aimed at revolutionizing the AI inference landscape by 2026, this new chip marries the low-latency advantages of Groq's language processing units (LPUs) with the high-throughput capabilities of Nvidia's Rubin GPUs. AI inference, a critical process for deploying trained AI models in real-world applications, often comprises two phases: prefill and decode. This allows models to interpret new data and generate responses, crucial for applications ranging from chatbots like ChatGPT to autonomous vehicles. By integrating Groq's LPU technology, which excels in sequencing natural language with low latency, with its own powerful Rubin GPUs leveraging high-bandwidth memory (HBM), Nvidia promises to significantly enhance AI interactivity and data processing speed. While the Rubin GPU surpasses in memory volume, the Groq 3 LPU offers an impressive 150 TB per second of memory bandwidth, dwarfing the Rubin's 22 TB per second, and dramatically improving throughput. This innovation is more than just a technological leap; it's a market shift. The Groq 3 LPX promises to deliver up to 35 times higher throughput per megawatt for trillion-parameter AI models compared to previous models, a claim that positions Nvidia's offering well above its competitors. This improvement is not only energy-efficient but also ensures rapid and intelligent responses from AI models—an essential factor for retaining user engagement and satisfaction with AI technologies. The introduction of the Groq 3 LPX inference accelerator underscores Nvidia's strategic prowess and technological leadership in the AI chip market, with expectations of bolstering both its sales figures and its stock market allure as the demand for cutting-edge AI technologies continues to rise.