苹果公司计划在全球开发者大会上展示其大规模人工智能技术，但据报道，该公司将在5月7日提前发布OLED iPad Pro。新款iPad Pro将搭载M4芯片和神经引擎，成为苹果的首款真正的人工智能设备。此外，新款iPad Pro还将获得OLED升级和一些新配件，如具有触觉反馈的Apple Pencil。这次发布被认为是iPad Pro自2018年以来最重要的一次改版。

Nvidia introduces Blackwell B200 GPU and GB200 superchip, offering improved performance and energy efficiency. B200 GPU provides 20 petaflops of FP4 horsepower, while GB200 combines two GPUs with a single Grace CPU for 30 times the performance. Nvidia targets companies with larger designs like GB200 NVL72, delivering 720 petaflops of AI training performance. Amazon, Google, Microsoft, and Oracle plan to offer NVL72 racks in their cloud services. Nvidia's systems can scale to tens of thousands of GB200 superchips.

由于需求超过供应，Nvidia的H100 GPU变得抢手和昂贵，成为首家市值万亿美元的公司。微软、Meta、OpenAI、亚马逊和谷歌等开始研发自己的AI处理器。Nvidia和AMD、英特尔等芯片制造商正在进行新一轮的AI芯片竞争。

研究人员使用内部数据训练大型语言模型，创建了一个名为ChipNeMo的自定义LLM，用于生成和优化软件并协助人类设计师。研究团队选择了三个用例：聊天机器人、代码生成器和分析工具。研究人员发现，定制化LLM的价值非常高。在芯片设计任务中，具有仅130亿参数的自定义ChipNeMo模型的性能甚至超过了具有70亿参数的通用LLM。

研究表明，生成性AI能帮助设计半导体，提高生产力。NVIDIA研究人员创建了自定义大型语言模型ChipNeMo，用于生成代码和辅助设计。该模型经过特定数据训练，表现优于通用模型。研究还强调了定制LLM的重要性和使用NVIDIA NeMo框架的便利性。

谷歌的新Tensor G3芯片升级了每个主要子系统，为设备上的生成式人工智能铺平了道路。最新的Pixel手机在设备上运行了两倍以上的机器学习模型，增强了用户体验。Tensor G3的高效架构与Google Research共同设计，可以处理设备上生成式人工智能的复杂性。

本文介绍了先进封装技术对芯片制造商的影响，逻辑能力和3D堆叠技术是关键。制造商需要掌握先进封装技术，并与客户合作开发解决方案，以在市场上获得竞争优势。先进封装市场提供了颠覆性机会，但也带来了挑战。

全球各行业面临半导体供应挑战，首席运营官采用人工智能解决方案增强供应链弹性。半导体瓶颈原因包括产能有限、需求高涨和过度订购，问题可能持续到2023年。制造能力增长不稳定，小节点能力增长最快。最近的瓶颈已减轻，包括亚洲COVID-19影响和需求激增。

Hey, advertisers. Register for our upcoming Webinar: Digital Advertising Benchmarks for 2022, where you’ll learn emerging trends and best-in-class examples in advertising gleaned from the digital performance of over 1,200 B2C and B2B brands across six sectors. Rising inflation, war in Ukraine, enduring supply chain disruptions, labor shortages (and in some cases, layoffs) — it seems there’s no shortage of macro environmental disruptions impacting advertisers today. And let’s not forget about ongoing privacy developments from Apple and Google (will Google finally deprecate third-party cookies on Chrome? Check out Google’s Latest Cookie Delay Puts Privacy Sandbox in the Trough of Disillusionment).  For many organizations, advertising is one of the first budgets to scrutinize (read: cut) when business plans turn sour and future outlooks appear murky. Advertising’s generally higher share of the marketing budget coupled with the intensifying pressure among CMOs to prove marketing ROI and demonstrate performance make web-based advertising an easy target. Take a look at the figure below. This chart maps total desktop display, desktop video and mobile display advertising impressions across 1,260 brands between the second half of 2021 and the first half of this year. As you’ll see, web-based advertising impressions across brands have fallen dramatically from its 2021 peak. While desktop and mobile display and video impressions alone don’t tell the whole story (paid social advertising and other ad channels fared better), it does indicate shaken confidence and a certain reluctance among brands to march forward.  However, rather than drastically cut back on ad spend – or worse, continue with “business as usual” – marketers must use this disruption as an opportunity to rethink their media and channel mix, adapt ad messaging to appeal to ad-avoidant and price-sensitive consumers, and take calculated risks that position them ahead of the competition.  Join myself and Gartner analyst, Mike Froggatt on September 13 as we explore emerging trends and best-in-class examples in advertising during our Webinar: Digital Advertising Benchmarks for 2022. We’ll share benchmarking data on desktop display, desktop video and mobile display impressions to provide a high-level view of how 1,260 B2C and B2B brands across sectors are evolving advertising efforts amidst all this turbulence. See you there!   Comments or opinions expressed on this blog are those of the individual contributors only, and do not necessarily represent the views of Gartner, Inc. or its management.

这篇文章介绍了如何在MacOS 12.4上使用M1max芯片安装Leo。只有M1芯片的MacBook Pro才能安装Leo。安装过程包括miniconda、PyQt的安装和创建Leo特殊环境。然后，将PyQt的包从brew复制到miniconda，并从GitHub下载最新的Leo源代码。最后，在conda环境的Python中调用launchLeo.py来启动Leo。建议在.bash_profile中定义别名以方便使用。