AI Booster - NVIDIA GTC 2023 (1)
Acceleration libraries solve new challenges and open new markets.
這系列文章是透過蒐集、分享我覺得有意思AI服務、相關文章、影片，期許自己能更了解這世界上正在發生什麼。
NVIDIA GTC 2023 於前幾天登場，由於近期ChatGPT的熱潮，因此影片不管在NVIDIA的要構建的生態系包括新的硬體產品、軟體服務，以及前陣子很紅的元宇宙，大多圍繞著人工智慧，這也是老黃在影片講了三次AI的iPhone時刻正在發生。在一個多小時官方釋出的影片裡，老黃提到了加速運算函式庫(Acceleration Libraries)、DGX超級電腦、DGX 生成式 AI的雲端服務、4 款推論平台(Inference Platform)以及Omniverse等。
要支持AI大量的運算就需要大量運算晶片，將近半個世紀，整個電腦產業的發展踩著摩爾定律步伐前進，儘管近期的發展因為運算量已逐漸脫離摩爾定律的趨勢，能耗也會跟著不斷飆升，這也與政府和企業的淨零碳排相衝突，老黃認為加速運算和人工智慧可以解決眼前困境。
NVIDIA與台積電 (TSMC)、艾斯摩爾(ASML)和新思(Synopsys)合作推出cuLitho加速運算函式庫，聲稱運算式微影速度加快了超過 40 倍，原先需要兩週的時間來處理的單一光罩可縮短為8個小時，所需功率也能大大減少( 35MW ->5MW)。
NVIDIA cuLitho is a library with optimized tools and algorithms for GPU-accelerating computational lithography and the manufacturing process of semiconductors by orders of magnitude over current CPU-based methods. Manufacturing computer chips requires a critical step called computational lithography — a complex computation — involving electromagnetic physics, photochemistry, computational geometry, iterative optimization, and distributed computing. This computational lithography step is already one of the largest compute workloads in semiconductor production, necessitating massive data centers, and the silicon miniaturization evolution process exponentially amplifies the computation requirements over time.
除了晶片製造外，加速運算可以加速解決最佳化問題，這一類問題通常都是NP-Hard的問題，比如供應鏈物流的收、送貨問題(Pick and delivery problem)，問題不僅量大(比如每年全球有將近4000億個包裹)且複雜(比如每年全球有將近3770億個配送點)。客戶像是AT&T可以使用NVIDIA加速運算函式庫cuOpt及時處理派工問題，這類型的問題過去需要花費一整晚的時間才能完成，而現在可以以比過去迅速100倍速度解決。
PDP is a generalization of the Traveling Salesperson Problem and is NP-hard meaning there is no efficient algorithm to find an exact solution. The solution time grows factorially as the problem size increases. Using an evolution algorithm and accelerated computing to analyze 30 billion moves per second. NVIDIA cuOpt has broken the world record and discovered the best solution for Li&Lim’s (PDP) challenge.
老黃表示NVIDIA與全球量子運算社群合作開發cuQuantum加速運算函式庫，此函式庫可供研究人員發展量子程式設計模型來模擬複雜量子電路，目前IBM Qiskit、Google Cirq、百度量易伏 (Baidu Quantum Leaf)、QMWare、QuEra、Xanadu Pennylane、Agnostiq 和 AWS Bracket都已將 cuQuantum 整合到他們的模擬框架中
cuQuantum Appliance helps developers get started by making simulation software available in a container optimized to run on the latest NVIDIA DGX™ systems, and HGX systems.
用於電腦視覺(Computer vision)的CV-CUDA和影片處理(Video processing)的VPF是新一代的雲端加速運算函式庫。老黃提到到透過此函式庫，騰訊目前每天可以處理高達300,000 支影片、微軟則用於其Bing的圖片搜尋、Runway用來處理、編輯他們的雲端生成式AI影片服務。
CV-CUDA includes 30 computer vision operators for detection, segmentation, and classification. VPF is a python video encode and decode acceleration library.
NVIDIA Clara Parabricks 是用於基因定序(Genome Sequencing)的加速運算函式庫，透過此加速器，醫師可以當場對患者的血液進行 DNA 定序，並大幅降低基因定序的成本。
NVIDIA® Parabricks® is the only GPU-accelerated computational genomics toolkit that delivers fast and accurate analysis for sequencing centers, clinical teams, genomics researchers, and next-generation sequencing instrument developers.
