搜索“NVIDIA”

增长最快的交换机细分市场：2026 年的销量比 2025 年增长 94%，随后收入将以每年 59% 的速度增长，到 2031 年达到 420 亿美元，其中共封装光技术占比 23%。 Yole表示，在人工智能集群中，数据传输的支出占比增长速度超过了计算支出占比。互连不再是数据中心预算中的辅助项目，而是成为了主要支出项目之一。 美国公司主导着这个充满活力的市场。NVIDIA在2025年和2026年位居全球第一大网络ASIC供应商，它率先推出了基于交换机的纵向扩展架构，并且是高性能AI千兆集群横向扩展的关键参与者。 从2026年起，博通、思科、Marvell、Astera Labs和华为等公司将占据重要地位，尤其是在AMD MI400 GPU方面。与此同时，AWS、Meta和微软等超大规模数据中心运营商也在开发各自的AI ASIC芯片。谷歌的TPU则独立于传统的纵向扩展和横向扩展框架，采用环形拓扑结构进行点对点连接，而非通过数据包交换机。 Yole表示，长期以来一直是 NVIDIA 优势的纵向扩展交换技术，随着开放标准的出现，为超大规模数据中心提供了一个可靠的替代方案，这一技术正在向 NVI

匹配字段正文

robots to acquire increasingly complex skills in factories, construction sites, homes and other physical environments without people having to demonstrate every action. Skild used Nvidia Isaac Sim to create the simulated environment, giving the model a single objective: score a goal. Rather than specifying individual behaviors such as dribbling, tackling or recovering after a fall, the researchers allowed the policy to compete against progressively more capable versions of itself. Skild says: “T

匹配字段正文

excessive relative to the rest of the credit stack, although the gap opens up further out the curve,” Deutsche’s Jim Reid said in an email. “The AA-rated AI names (Meta, Alphabet, Nvidia, Amazon), sit at or below the bottom of the IG group, borrowing at around 5–5.5% at the front end. The BBB rated AI names (Oracle and SpaceX) cluster at around 8–8.5% at 10–12 years’ duration. That is the top of the IG distribution, roughly 1.5–2pp above most similar-duration bonds. The exception is CoreWeave, r

匹配字段正文

家输出用归一化的路由器权重组合，并缩放 2.5 倍。 每个 token 只运行 8 个专家，因此模型用约 23B 参数计算每个 token。所有 256 个专家都保留在内存中，因此部署需要 501B 模型的内存，但以 23B 模型的速度运行。BF16 checkpoint 以 safetensors 格式约为 1 TB。NVFP4 checkpoint 为 NVIDIA Blackwell GPU 以 4 位精度存储大多数参数，部分参数保留为 BF16 以保证准确性。Hub 上还有 FP8 checkpoint，非常适合 Hopper 和 Ampere 系列芯片。 如果你想自己部署 Beam，请查看下面的 vLLM 或 SGLang 配方。 残差流 残差连接是最独特的部分。标准 transformer 以权重 1 将每个子层输出加到流中，因此流随深度增长。 Beam 将输入嵌入计为第 1 步。在第 k 步，它以权重 1/sqrt(k) 加上子层输出，并以权重 sqrt((k-1)/k) 保留现有流。两个权重的平方和为 1，因此流从第一层到最后一层保持恒定尺度。在所有 104 次加法之后

匹配字段正文

day-zero support across the ecosystem including: Transformers (Hugging Face), vLLM (Inferact), SGLang (RadixArk), Unsloth, Axolotl, TokenSpeed (LightSeek Foundation), and Dynamo (NVIDIA) so you can go from weights to a production-scale endpoint without writing custom plumbing. Figure 2: Reflection’s production inference dashboard, powered by SGLang serving Beam at scale on NVIDIA Blackwell infrastructure. Agentic development with Mirror CLI Beam and Mirror CLI are both open-source friendly. We

匹配字段正文

Muse Spark Open - Meta’s open-weight AI for coding and long-horizon agents 🗣️ GPT-Live-1.1 - OpenAI’s new voice AI with more natural multilingual accents 🤖 Nemotron 3.5 Super VL - Nvidia’s open multimodal AI for specialized AI agents 🖼️ FLUX 3 Image Klein - BFL’s small open model for 4K image generation, editing 📰 Everything else in AI today Amazon’s Bee launched two AI wearables starting at $74.99 that record conversations, generate summaries and to-dos, and integrate with Alexa and productivit

匹配字段正文

Jensen Hunag talks about a large revenue line of Nvidia, software licensing: charging companies a yearly license of roughly a couple of thousand dollars per GPU to run, secure and maintain Nvidia's AI software, and with about 20M GPUs shipped a year and around 50% going to enterprises, this is a huge recurring revenue stream. ---- Full video on "Bg2 Pod" YouTube channel, (link in comment)

匹配字段正文

examples of real-world applications, followed by the full consumer launch in Phase Three. The Nova Physical AI Reinforcement Learning Developer Program offers support based on the NVIDIA Isaac platform, with an end-to-end workflow running from simulation training to validation on physical hardware. Developers can conduct parallel training in simulation, tune reward functions, and apply domain randomization, then deploy trained policies on a physical FF Nova for validation and optimization. Liste

匹配字段正文

that moves “from the desk side to the data center, to the edge, and all the way out to the cloud,” as Grocott puts it, it’s making moves to meet these needs. In collaboration with Nvidia Corp., the company is helping enterprises that do not want to run their AI agents exclusively in the cloud, but rather locally. Deskside Agentic AI is helping businesses achieve this goal, offering several potential benefits. To start, companies can address some data security concerns by working with their infor

匹配字段正文

rk Open includes custom GPU kernels optimised for inference through SGLang. Meta said the kernels delivered speedups ranging from 1.03 times to 3.95 times in kernel-level tests on NVIDIA and AMD GPUs. The company also reported a 3% end-to-end performance improvement on AMD MI350 GPUs. The model supports several hardware platforms, including AMD, Apple, Intel and NVIDIA silicon. Developers can further fine-tune it using PyTorch's TorchTitan framework. Meta has released the weights under its Muse

匹配字段正文

ia MLX (AA, Meta). Ecosystem: llama.cpp had GGUF support on day 0 (Delangue), and Unsloth published quants, with UD-IQ1 S at 176GB (Unsloth). vLLM day-0 stack: vLLM support covers NVIDIA, AMD MI355X (FP8 and MXFP4) and TPU (vLLM). TPU throughput: Megakernels with DSpark speculation (acceptance length 6) exceed 650 tok/s on 4 TPU v7 chips and 960 tok/s on 8 (hardware). Speculator: Inferact trained a block-diffusion DSpark drafter that proposes up to 7 tokens per pass (speculator). LoRA RL: LoRA G

匹配字段正文

els. It runs on inference engines including SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, Apple MLX, llama.cpp, and Unsloth, with optimizations for AMD, Apple, Intel, and NVIDIA silicon, and developers can further tune the model using PyTorch’s TorchTitan. Weights ship in BF16 and NVFP4 versions; running locally requires a GPU with at least 700GB of VRAM for NVFP4 or 2TB for BF16. Meta says Muse Spark Open is the first model in its class to run on a desktop, using two Apple M5 Ultra Mac

匹配字段正文

ls at long-horizon tasks, complex software engineering, and multi-step reasoning.” The two U.S. models Muse Spark Open competes with are essentially Thinking Machines’ Inkling and Nvidia’s Nemotron 3 Ultra. In Meta’s own benchmarks, its new model beats them across the board — often by 20 points or more. Credit: Meta. The closer fight, however, is with Chinese open models that many developers already run today. Spark Open leads Kimi K3, GLM 5.3, and Qwen3.8 on the agentic and tool-use benchmarks,

匹配字段正文

ROSOFT 🔥: - GitHub Copilot updated HydraFusion's model pool and added a new HydraFusion Max option, both still in research preview. GOOGLE 🔥: - Google Messages is adding an enhanced protection mode against phone scams. - YouTube Music is rolling out labels for AI-generated songs. OPENAI 🔥: - Codex got another global usage-limit reset on Day 10. NVIDIA 🔥: - NVIDIA AI teased its own model launch coming next week. Used Grok to compose this brief, cherry-picking the news and doing some post-editing.

匹配字段正文

a couple hours.” Money and minds Akrites has had some serious industry muscle behind it from the get go, with founding members including Anthropic, OpenAI, Google, Microsoft, AWS, NVIDIA, Cisco, IBM, Red Hat, Citi, JPMorganChase, among others. What, perhaps, distinguishes Akrites from most other Linux Foundation projects is that its members aren’t just being asked to write cheques. While financial contributions are very much part of the deal, members are also expected to commit engineering resou

匹配字段正文

sodass auch LTS-Nutzer davon profitieren. Weiterlesen nach der Anzeige Entfernt wurde die Unterstützung für sowieso obsolete WEP-WLANs und die EGLStreams-Schnittstelle, die ältere Nvidia-Treiber nutzten. Für Computer mit Qualcomm-Snapdragon-Prozessoren steht eine neue Preview-Version bereit. Laut Release Notes macht Linux 7.3 auf bestimmten Snapdragon-X2-Elite-Notebooks allerdings noch Probleme mit der Tastatur. Im App Center lassen sich nun proprietäre Treiber nachinstallieren, was seit der Abk

匹配字段正文

e Associated PressThe Associated PressOctober 15, 2026, 10:16 AM ETTSMC, which produces over 90% of the world's most cutting-edge chips, supplies to U.S. technology giants include Nvidia and Apple and has pledged $265 billion of investments in Arizona.AP Photo/Daniel CengAdd Fortune on Google for similar content. Taiwan Semiconductor Manufacturing Co. is exploring potential investments in Texas to expand its chip production in the United States, the advanced chipmaker said after posting another

匹配字段正文

ucture needed to develop the technology. Stocks in some US technology companies dropped on Wednesday over concerns around investments in AI. Shares in chip makers - including AMD, Nvidia and Intel - fell sharply. In Asia, shares in Japanese tech investment firm Softbank were down by more than 9% on Thursday. South Korean semiconductor giants also dropped, with Samsung falling by 9.5% and SK Hynix 11% lower. Shares have been falling this week after remarks by Hock Tan, the chief executive of US t

匹配字段正文

stigators, but it had a massive impact on tech stocks. The Nasdaq index fell 1.25% and the shares of many individual firms, including Oracle Corp., CoreWeave Inc., Intel Corp. and Nvidia Corp., all fell more. Their fortunes are tightly linked to those of the AI giants, given that many have signed massive multiyear, multibillion-dollar contracts with them. Holger Mueller of Constellation Research said the number thrown by Anthropic at investors is all about making it look good ahead of its IPO, b

匹配字段正文

be restored with a recorded explanation. “It’s more like a dimmer switch,” Bardoliwalla said. “You monitor, you pause, you restrict, and eventually, if necessary, you deactivate.” Nvidia Corp. is a design and launch partner, linking AI Passport with its Open Agent Safety Platform, including the OpenShell secure runtime and Sentry security framework. The integration connects enterprise policies with enforcement within the infrastructure where agents operate. The central implementation challenge i

匹配字段正文