变形金刚1

Rajkummar Rao Calls for a Return to Conviction-driven Storytelling: "We’ve Become Too Data-based"_我的网站

刑事侦缉档案4

一 | Rajkummar Rao is known for his instinctive choices as an actor. He recently opened up about the changing landscape of cinema and the growing influence of algorithms on creative decisions. Speaking at a recent event, the actor reflected on how streaming platforms have reshaped audience behaviour, often in ways that defy logic and how this shift is affecting filmmakers and performers alike. Rajkummar shared, "Honestly, I'm still learning. I'm still trying to figure out. It's actually like theatres. Sometimes a below average film works very well on streaming platforms. And you wonder, really? This one did so well, and a really good film - nobody talks about it. And you wonder, what's the update? It's the same like theatres. There is a lot of algorithm, lot of data, analysis which, I think, people don't understand. I'm pretty sure, whether it's Francis Ford Coppola who made Godfather, or when Manmohan Desai made Amar Akbar Anthony - they didn't think about the target audiences. They just wanted to tell those stories. So that is something that I am missing. We have become too analytical and too data-based."He further added, "We are not telling stories which are from the heart - which I think Rishabh Shetty has done wonderfully in Kantara. He didn't think about - “I’m going to make this for a certain audience.” He just told that story with full conviction and made a beautiful film out of it. And hit it out of the park. So, that is something, as a creative person, as an actor, I miss. I really miss. I miss seeing good actors on screen - be it streaming or theatres." On the personal front, Rajkummar welcomed his first baby with Patralekhaa on November 15, 2025. They met in 2010 and tied the knot in November 15, 2021. Also Read: Rajkummar Rao and Patralekhaa Welcome Baby Girl on Wedding Anniversary。 AMD对英伟达数据中心GPU垄断地位的挑战，正从低价替代策略转向定价权叙事。

二 | 华尔街见闻提及，7月20日，微软将在其Azure云平台上部署AMD的Helios机架级解决方案，用于驱动前沿模型AI推理工作负载。

三 | Futurum Group估计，在与微软达成的迄今最全面云合作协议中，Helios机架级AI系统以高于英伟达Rubin约40%的售价获得微软全栈采购，标志着AMD在AI基础设施市场的竞争逻辑出现了根本性转变。微软将在Azure上部署Helios用于前沿AI推理，同时引入第六代EPYC Venice CPU虚拟机系列及Pensando DPU。这是AMD首次在单一云客户中实现"GPU+CPU+网络"的全栈落地。周一芯片板块情绪回暖，叠加AMD本周晚些时候将举行大型产品发布活动，AMD股价一度大涨超5%，尾盘涨幅回落，收至1.58%。比Rubin贵40%却照样卖此次交易中一个易被忽视的数据是价格。据全球性技术研究和顾问公司Futurum Group估计Helios售价在500万至550万美元之间，高于其对英伟达Vera Rubin 350万至400万美元的估价。 Helios每个计算托盘配备四块Instinct GPU，由单颗EPYC CPU驱动，并搭载多达12块Pensando网络芯片。 AMD数据中心负责人Forrest Norrod称其核心优势在于"最佳总拥有成本和最低每词元成本"。

四 | 交易并非仅止于GPU。亚马逊Azure将新增两个基于第六代EPYC Venice处理器的虚拟机系列，并在AI后端网络和部分服务中部署Pensando DPU。这意味着AMD在微软云中的角色，正从"GPU供应商"升级为"综合AI基础设施提供商"。巴克莱近期将AMD目标价从500美元上调至665美元，认为其CPU上行空间在三家主要芯片股中"最未被充分定价"。该行预测AMD 2027年CPU收入将达约290亿美元，仅CPU一项到2030年可贡献约19美元每股盈利。

五 | 产品周前的催化剂矩阵微软的加入使AMD的Helios客户矩阵进一步加密。此前Meta已承诺6吉瓦GPU部署，OpenAI、甲骨文和印度TCS亦作出重大承诺。AMD称全球前十大AI公司中已有八家在其Instinct GPU上运行工作负载。从财务数据看，2026年Q1数据中心营收同比增长57%至58亿美元，公司计划2027年起实现数百亿美元数据中心AI营收，其中大部分来自Helios。 Futurum Group分析师Daniel Newman认为AMD有望将数据中心GPU市场份额从当前的4.5%提升至20%-25%，"这意味着数千亿美元的营收规模"。芯片板块周一整体反弹提供了有利的情绪背景。而本周产品发布的预期进一步强化了这一叙事，市场关注苏姿丰是否会给出更激进的2027年Helios出货指引。推理时代，软件或是护城河硬件层面AMD已逼近甚至部分超越英伟达，但软件仍是关键变量。华尔街见闻提及，此前半导体研究机构SemiAnalysis高度评价英伟达在vLLM推理引擎上的性能优化，同时指出AMD在部分模型支持上仍有明显差距。在软件层面，英伟达推出的Dynamo分布式推理框架将vLLM深度集成，专门针对MoE模型实现了预填充与解码分离（Disaggregated Serving）、高效KV缓存传输以及双批次重叠等优化。

六 | 这套框架在NVL72上能够充分发挥硬件潜力，而AMD方面目前仍主要依赖标准vLLM与DISAGG版本，针对超大MoE模型与宽并行场景的深度优化尚未跟上。英伟达凭借二十年CUDA积累、主流框架优先适配及TensorRT等优化库构建的软件壁垒，在推理时代仍形成持久竞争优势。Counterpoint Research分析师Neil Shah同样指出：关键在于软件和优化，英伟达凭借CUDA拥有更为成熟庞大的生态系统。