COMPUTEX 2024 臺北國際電腦展將於6月4日登場，昨日(6月2日晚間) 輝達（NVIDIA）執行長黃仁勳於在臺大體育館開講，先行打響AI科技盛會第一槍。
黃仁勳指出，「臺灣奠基了全世界人工智慧的發展，而輝達正處於電腦圖學、虛擬建構及人工智慧發展的關鍵交叉口」，提出輝達創造了虛擬現實「Ominiverse」，結合「加速運算 Accelerated Computing 」、「人工智慧AI」技術，可以將人工智慧及大數據視覺化，將會重塑迄今已發展六十年的IT資訊業。
黃仁勳強調，十八世紀時，英國工業革命讓英國偉大；現今，全世界正面臨AI產業革命，而臺灣就位於世界的中心，二十一世紀的今日，希望臺灣人相信自己，也能站在下個時代的浪尖上。
The Taipei International Computer Exhitbition, COMPUTEX 2024, kicks off on June 4th. Last night (June 2nd), NVIDIA CEO Jensen Huang gave a keynote speech at the NTU Sports Center, marking the start of this major AI technology event.
Huang emphasized that "Taiwan has created the world’s AI infrastructure", and NVIDIA stands at the pivotal intersection of computer graphics, simulations, and AI development." He introduced NVIDIA's creation of the virtual reality platform "Omniverse," which integrates Accelerated Computing and AI technologies. This platform can visualize AI and big data, promising to reshape the IT industry, which has been evolving for sixty years.
Huang highlighted that in the 18th century, the British Industrial Revolution made Britain great; today, the world faces an AI industrial revolution, and Taiwan is at its center. In the 21st century, he hopes the people of Taiwan will believe in themselves and lead the next era.
(NVIDIA CEO Jensen Huang)
演講筆記 Keynote
CUDA運算使輝達生態系成形 CUDA Computing Shapes the NVIDIA EcosystemCUDA（Compute Unified Device Architecture）運算已經成為應用開發的標準。CUDA是由NVIDIA開發的平行計算平台和應用程序接口（API），能夠利用圖形處理器（GPU）進行通用計算，客戶可以直接使用NVIDIA建立的大量資源，包括軟硬體工具和開發社群等。隨著技術進步，運算能力的單位成本將不斷降低，將推動更多元廣泛的應用。
黃仁勳宣佈，NVIDIA推出了AI模型的推論「微服務 NIM」，可以即時部署生成式AI，讓開發者能更簡單、更快速部署AI應用 (輔助夥伴copilot、聊天機器人、生成文字、圖片、影片、語音等)，使用上只需像運用ChatGPT一樣輸入指令即可。
NVIDIA's CUDA (Compute Unified Device Architecture) computing has become a standard for application development. CUDA is a parallel computing platform and API developed by NVIDIA that uses GPUs for general computing. Customers can leverage a vast array of resources provided by NVIDIA, including software, hardware tools, and development communities. As technology advances, the cost per unit of computing power will continue to decrease, driving broader and more diverse applications.
Jensen Huang announced that NVIDIA has introduced "NIM" (NVIDIA Inference Microservices) for AI model inference, enabling real-time deployment of generative AI. This allows developers to deploy AI applications (such as copilot assistants, chatbots, text generation, image creation, video production, and voice synthesis) more easily and quickly. Users can simply input commands, just like using ChatGPT.
晶片計算力串聯 Interconnecting Chip Computing Power在高效能計算領域，客戶預算直接影響可用的計算能力。然而，將多個高效能晶片串聯起來的過程需要考慮多方因素，包括電源供應、機殼設計、液冷散熱系統、矽光子高速傳輸技術，以及高頻寬記憶體（HBM，High Bandwidth Memory），這些元素可以共同確保系統的穩定性和高效運作。
In high-performance computing, customer budgets directly affect available computing power. However, interconnecting multiple high-performance chips involves various considerations, including power supply, case design, liquid cooling systems, silicon photonics high-speed transmission technology, and High Bandwidth Memory (HBM). These elements collectively ensure system stability and high efficiency.
輝達聯手台積電，建立客戶忠誠度 Building Customer Loyalty with TSMC Partnership一旦客戶體驗過NVIDIA產品的強大性能，往往會對其產生依賴，如同許多科技公司必須選擇台積電（TSMC）進行晶片代工。即使客戶有自訂晶片的需求，依然難以忽視NVIDIA產品帶來的優勢和便利。
Once customers experience the powerful performance of NVIDIA products, they often become reliant on them, much like many tech companies must choose TSMC (Taiwan Semiconductor Manufacturing Company) for chip production. Even when customers need custom chips, they find it hard to overlook the advantages and convenience of NVIDIA products.
一年一案新計算平台開發計畫 Annual New Computing Platform DevelopmentNVIDIA每年都會推出新的GPU計算平台，以維持技術領先地位，從2023年Hopper，到2024年Blackwell Ultra，再到2025、2026年的Rubin平台，每一個新平台都能與前一代兼容互聯，確保客戶無縫升級和擴展系統。
NVIDIA launches a new GPU computing platform every year to stay at the forefront of technology. From 2023’s Hopper to 2024’s Blackwell Ultra, and the upcoming Rubin platforms in 2025 and 2026, each new platform is backward compatible, ensuring seamless upgrades and system expansion for customers.
AI應用層面更加廣泛全面 Broader and More Comprehensive AI ApplicationsNVIDIA的技術應用深入各個領域，從AI邊緣運算、日常機器人、生成式應用，到計算中心、超級工廠(智慧工廠)、科學研究，這些應用不再只是概念，而是實際存在運行的系統，顯示出NVIDIA技術對現今各產業帶來的廣泛影響力。
NVIDIA's technology applications span various fields, from AI edge computing, everyday robotics, and generative applications to computing centers, smart factories, and scientific research. These applications are no longer just concepts but real, operational systems, showcasing NVIDIA's extensive impact across different industries.
結語 Conclusion
黃仁勳對臺灣作出感謝自白及未來期許。他表示，是臺灣一路協助輝達實現願景，從加速運算、電腦繪圖、科學研究到人工智慧發展，每一個晶片、每一台電腦，背後都有無數的夥伴支持，「臺灣是無名的英雄，卻是世界的支柱」，一同推動產業復興，讓世界踏入嶄新運算的領域，相信未來輝達製作的電腦，將會轉化成新型的智慧工廠，在人工智慧發展上更進一步，創造出更加智慧便利的生活，解決全球疾病、自然災害等問題。
Jensen Huang expressed his gratitude to Taiwan and shared his hopes for the future. He stated that Taiwan has been instrumental in helping NVIDIA achieve its vision, from accelerated computing and computer graphics to scientific research and AI development. Behind every chip and computer are countless partners' support.
"Taiwan is an unsung hero, yet a pillar of the world," collectively driving industry revival and ushering in a new era of computing. He believes that future NVIDIA computers will transform into new smart factories, advancing AI development, creating smarter and more convenient lives, and addressing global issues such as diseases and natural disasters.
(演講畫面：列出輝達在臺灣提供支持的合作夥伴單位)
Σχόλια