# Nvidia Articoli collegati

Il Centro Notizie HTX fornisce gli articoli più recenti e le analisi più approfondite su "Nvidia", coprendo tendenze di mercato, aggiornamenti sui progetti, sviluppi tecnologici e politiche normative nel settore crypto.

DeepSeek Secretly Builds AI Chip, Specializing in Inference, Project Started a Year Ago with No Public Recruitments

DeepSeek, the Chinese AI company known for its algorithmic models, is secretly developing its own AI chip to reduce dependence on Nvidia, according to a Reuters report. The chip is designed specifically for AI inference, not training, and the project began approximately a year ago. Currently in early stages, DeepSeek is reportedly in discussions with chip design firms, foundries, and memory suppliers. The company, historically focused on algorithmic efficiency, has been discreetly hiring chip design engineers without public job postings. This move aligns with a global trend where major AI model companies like OpenAI and Anthropic are also pursuing custom chip development. DeepSeek founder Liang Wenfeng previously highlighted chip shortages as a challenge. While the company initially trained models on Nvidia H800s and later adapted to Huawei's Ascend chips, it now seeks greater control over its hardware foundation. Designing a competitive AI chip is a significant challenge, requiring years and substantial investment with no guarantee of success. However, DeepSeek's efforts are backed by a recent major funding round of approximately 51 billion RMB (about $7.4 billion) raised in June 2026. The funds are designated for expanding data centers based on domestic chips, developing proprietary AI chips, and recruiting top global talent. Infrastructure plans are also advancing, with job postings for data center design engineers, including projects in locations like Ulanqab, Inner Mongolia. The company remains characteristically low-key, with sources speaking anonymously and no official comment from DeepSeek itself. Nevertheless, this initiative marks a strategic expansion from software algorithms into the hardware layer that powers its AI systems.

marsbit3 h fa

DeepSeek Secretly Builds AI Chip, Specializing in Inference, Project Started a Year Ago with No Public Recruitments

marsbit3 h fa

When a League of Legends World Champion Uses 'Pro Gamer Moves' in Stock Trading...

From League of Legends world champion to stock market success: the story of 'imp' Former League of Legends world champion, 'imp' (Gu Seung-bin), is reportedly achieving financial success in the stock market after retiring from esports. Known for his stellar ADC performance with Samsung White, winning the S4 World Championship in 2014, imp's post-retirement life has been marked by personal struggles, including a costly divorce that reportedly left him with significant debt and depression. However, the narrative has shifted. Fellow former pro player Doinb recently claimed imp has achieved financial freedom through savvy investments in Korean and US stocks, particularly in semiconductor companies like NVIDIA and SK Hynix. While imp has not publicly confirmed this, he previously hinted at his market activities during livestreams, suggesting he invested in NVIDIA before the AI boom and that his assets grew tenfold over a seven-month period. This potential financial turnaround mirrors a dramatic comeback for the player once nicknamed the 'Rat King,' suggesting he may have found a new arena in which to excel.

Odaily星球日报Ieri 03:09

When a League of Legends World Champion Uses 'Pro Gamer Moves' in Stock Trading...

Odaily星球日报Ieri 03:09

One Megawatt Sustains 60,000 Agents, NVIDIA GB300 Crushes Previous Generation by 20x

NVIDIA's latest GB300 NVL72 system achieves a 20x improvement in AI agent throughput per megawatt compared to its predecessor, the H200, according to a new industry benchmark called AA-AgentPerf. Where the H200 could handle roughly 2,600 concurrent agents per megawatt, the GB300 NVL72 can support approximately 61,400. The significance lies less in raw chip performance and more in the new benchmark itself. AA-AgentPerf, created by the independent firm Artificial Analysis, is the first benchmark designed specifically for "AI agent" workloads. Traditional benchmarks measure single, fixed-length requests, but AI agents operate in long, complex chains involving dozens of model calls, tool use, and ever-growing context. These create unique system pressures that older tests cannot capture. AA-AgentPerf replays real programming agent trajectories with lengthy sessions and varying input lengths. Its key metric is "agents per megawatt," measured under strict Service Level Objectives (SLOs) that guarantee a minimum token output speed per agent. It also allows real-world optimizations like KV cache reuse and speculative decoding, which older benchmarks often disable. The results highlight two key trends: rack-scale systems like the 72-GPU GB300 NVL72 are inherently more efficient than single nodes, and the architectural leap from Hopper to Blackwell (H200 to GB300) represents a systemic, not just incremental, performance gain. The GB300's advantage stems from its high-bandwidth NVLink fabric connecting all GPUs, allowing large MiE models to be efficiently distributed and parallelized. Important caveats include that the 61,400 figure represents simulated concurrent sessions, not independently running full models, and that benchmark results are a snapshot that will improve with software optimization. AA-AgentPerf is a new standard whose industry adoption remains to be seen.

marsbit2 giorni fa 01:03

One Megawatt Sustains 60,000 Agents, NVIDIA GB300 Crushes Previous Generation by 20x

marsbit2 giorni fa 01:03

Token Devours One-Third of Payroll, Silicon Valley's AI Bill is Spinning Out of Control

The article discusses the dual reality of AI token costs in Silicon Valley. While research firm SemiAnalysis reports spending 30% of its employee salary budget on internal LLM tokens—translating to massive productivity gains like converting complex Excel models in minutes—other giants are struggling with ballooning, uncontrolled AI bills. Uber exhausted its annual AI budget in months after rapid engineer adoption, and Microsoft is cutting third-party AI tools due to high costs. NVIDIA's CEO argues tokens are becoming "means of production" and plans substantial AI budgets per engineer. Despite current cost concerns, the analysis emphasizes that cost collapse is just beginning. Through software optimizations (like 14x throughput boosts) and next-gen hardware (e.g., GB300 NVL72 with 17-32x H100 performance), real token costs can fall far below list prices. Anthropic's gross margins reportedly soared as token prices dropped. Gartner predicts a >90% inference cost drop by 2030. The piece highlights a split: massive AI capex ($740B announced) contrasts with tech layoffs and minimal measured economic impact so far. The transition mirrors past infrastructure shifts—investment precedes widespread productivity. For early adopters like SemiAnalysis, tokens already deliver high leverage; for others, the choice is to adopt now or risk falling behind.

marsbit2 giorni fa 00:15

Token Devours One-Third of Payroll, Silicon Valley's AI Bill is Spinning Out of Control

marsbit2 giorni fa 00:15

Both OpenAI and Anthropic are 'Developing Their Own Chips' — Beyond Cost, the Control Over Computing Power is Paramount

OpenAI and Anthropic are both advancing plans to develop custom AI chips, driven by the need to control computing power and reduce costs. According to reports, Anthropic is in early-stage development of its own chips and in talks with Samsung for manufacturing, while OpenAI is collaborating with Broadcom and TSMC, aiming to deploy its first inference chip by late 2026. The primary motivation extends beyond just lowering expenses. For these large model companies, chips are core production assets. By designing specialized hardware (ASICs) tailored to their specific model architectures—OpenAI's being more sparse and Anthropic's more dense—they aim to achieve deeper software-hardware co-design. This synergy can significantly improve inference speed, energy efficiency, and overall unit economics, offering advantages that off-the-shelf GPUs cannot. This move does not signify an immediate replacement for suppliers like Nvidia. The process from design to deployment takes 18-24 months, and Nvidia's GPU ecosystem remains deeply entrenched. Instead, custom chips provide a strategic alternative and negotiating leverage, allowing companies to use them for specific, high-volume workloads like inference while still relying on external GPUs and TPUs for other tasks. The trend reflects a broader industry shift where AI competition is evolving from pure algorithmic prowess to integrated control over the entire software-hardware stack. Companies like Google, Amazon, Meta, and Microsoft are already on this path. For foundries like Samsung, securing orders from AI leaders like Anthropic represents a significant opportunity to expand its footprint in the advanced semiconductor market for AI. Ultimately, the race for "computing sovereignty" is now a central battleground for major AI players.

marsbit07/03 13:38

Both OpenAI and Anthropic are 'Developing Their Own Chips' — Beyond Cost, the Control Over Computing Power is Paramount

marsbit07/03 13:38

While Semiconductor Stocks Plunge, Anthropic Plans to Develop a 2nm Chip

Anthropic, the AI company behind Claude, is exploring the development of its own custom AI chip, according to a report from The Information. The company is in early discussions with Samsung Electronics to manufacture the chip using Samsung's most advanced 2-nanometer process and packaging technology. While the project is still in preliminary stages, including defining chip specifications, and could be abandoned, it marks a strategic step for Anthropic. The move comes as the company seeks greater control over its computing costs and hardware optimization, particularly for inference tasks to run its models more efficiently and cheaply. Samsung's potential involvement follows its participation as a strategic investor in Anthropic's recent $65 billion funding round. For Samsung, partnering with a major AI lab represents a significant opportunity for its foundry business to compete with market leader TSMC in advanced semiconductor manufacturing. Anthropic's CEO, Dario Amodei, has previously highlighted the immense financial challenge of securing enough computing power for anticipated growth, making cost-effective inference a critical focus. The company would join other tech giants like Google, Amazon, Microsoft, Meta, and OpenAI in pursuing custom AI silicon. However, analysts note this trend creates deeper interdependencies rather than independence, as US AI labs become more tightly woven into Asian semiconductor supply chains. Despite this move, Anthropic remains heavily reliant on a multi-cloud, multi-vendor strategy for its immediate computing needs. It has secured massive, long-term commitments for capacity from Amazon Web Services (Trainium chips), Google (TPUs), and even leased a large GPU cluster from xAI. For now, Nvidia continues to dominate the AI chip market, with its share reportedly growing to 74%.

链捕手07/03 09:54

While Semiconductor Stocks Plunge, Anthropic Plans to Develop a 2nm Chip

链捕手07/03 09:54

NVIDIA's Annual 'Most Dangerous' Paper: AI Self-Replicating Code, Unlimited Leveling and Evolution

NVIDIA's "Red Queen Gödel Machine" (RQGM) paper proposes a potentially groundbreaking AI self-evolution framework. It breaks from the long-stalled concept of the "Gödel Machine," which required mathematically proven beneficial self-modifications, by adopting an evolutionary approach. The core, and most striking, innovation is that the AI does not just evolve its own code in a static environment. Instead, it co-evolves both the "student" (the task-performing agent) and the "examiner" (the evaluation system that judges it). This creates a dynamic, recursive self-improvement loop inspired by the biological "Red Queen Hypothesis"—where continuous adaptation is needed just to maintain relative fitness. The mechanism operates in epochs. Within an epoch, a fixed examiner evaluates all candidate code variants. At epoch boundaries, a new, potentially more rigorous examiner can replace the old one, but only if it proves statistically superior on a held-out "ground truth" dataset. This "controlled utility evolution" aims to ensure progress is measurable and grounded. The paper demonstrates RQGM's effectiveness across three domains: 1. **Code Generation:** It achieved a 71.7% test-set pass rate (improving over a 69.9% SOTA) while using 1.35-1.72x fewer computational tokens. 2. **Paper Writing:** In a subjective task, the co-evolved writer and reviewer achieved a 40.5% acceptance rate by a fixed human panel, up from 21.8%. 3. **Math Proofs:** It evolved more accurate graders (at 3x lower cost) and higher-scoring provers. Notably, RQGM also mitigated a known LLM bias where AI reviewers favor AI-generated content. By specifically rewarding reviewers that correctly rejected AI-written papers from a historical pool, the evolved system achieved impartiality while maintaining 80% accuracy. The research has sparked significant discussion about the acceleration of Recursive Self-Improvement (RSI). Some, like Anthropic's Jack Clark, have predicted a high probability of highly autonomous, self-evolving AI emerging by 2028. The paper suggests that when an AI begins to design its own evaluators and push itself toward ever-higher standards in a recursive loop, it may be taking a fundamental step toward redefining intelligence and autonomy.

marsbit06/28 07:50

NVIDIA's Annual 'Most Dangerous' Paper: AI Self-Replicating Code, Unlimited Leveling and Evolution

marsbit06/28 07:50

The Eight Fathers of Transformer: Where Are They Now?

The founding authors of the seminal 2017 paper "Attention Is All You Need," which introduced the Transformer architecture, have all departed from Google and are now pioneering diverse ventures across the AI landscape. Their current paths highlight the field's evolution and fragmentation. Ashish Vaswani co-founded Essential AI after a stint at Adept, but recent reports suggest he and his team are being recruited by Nvidia. Noam Shazeer left Google twice, first to found Character.AI—later acquired by Google—before recently joining OpenAI. Niki Parmar, after co-founding both Adept and Essential AI with Vaswani, is now at Anthropic. Jakob Uszkoreit is the CEO of Inceptive, applying AI to RNA and drug design. Llion Jones co-founded Sakana AI in Tokyo, exploring collective intelligence from smaller models. Aidan N. Gomez is the CEO of Cohere, focusing on enterprise AI solutions. Łukasz Kaiser remains in research at OpenAI, contributing to models like GPT-4 and o1. Illia Polosukhin co-founded the NEAR Protocol blockchain, positioning it for an AI-agent economy. Despite their divergent paths—spanning biotech, enterprise software, blockchain, and foundational research—a shared sentiment persists: the Transformer is not the final answer. Each founder continues to seek the next significant architectural breakthrough, driven by the same innovative spirit that created the modern AI era.

marsbit06/28 05:30

The Eight Fathers of Transformer: Where Are They Now?

marsbit06/28 05:30

NVIDIA's New Open-Source MoE: One Line of Import, Fine-Tuning Accelerated by 3.7x

NVIDIA has open-sourced NeMo AutoModel, a tool designed to significantly accelerate the fine-tuning of Mixture-of-Experts (MoE) large language models. By adding just one import line to existing code based on Hugging Face Transformers v5, users can achieve a 3.4x to 3.7x increase in training throughput and reduce GPU memory usage by 29% to 32% without altering their API. The key innovations include Expert Parallelism (EP) to distribute expert weights across GPUs, lowering memory pressure; DeepEP to fuse computation and communication; and TransformerEngine kernels for accelerated core operations. Benchmarks on models like Qwen3-30B-A3B show training throughput per GPU jumping from 3075 to 11340 tokens per second. The solution also enables the fine-tuning of very large models, such as the 550B parameter Nemotron 3 Ultra, which would exceed memory limits with the standard Transformers v5. Code and benchmarks are available on GitHub.

marsbit06/26 07:29

NVIDIA's New Open-Source MoE: One Line of Import, Fine-Tuning Accelerated by 3.7x

marsbit06/26 07:29

Report Analysis: TSMC's AI Revenue to Double by 2027, CoWoS Capacity Remains a Bottleneck

Morgan Stanley's report on June 23 forecasts explosive growth for TSMC's AI-related revenue. It predicts revenue will surge to $86.3 billion in 2027, a 218% increase from 2026's $27.1 billion. While GPUs (led by Nvidia) remain the primary driver, significant new demand is emerging from AMD's Venice CPUs and MI400 GPUs, and Google's TPUs. A critical bottleneck is CoWoS advanced packaging capacity. Global demand for CoWoS is projected to reach 2.694 million units in 2027, a 93% year-on-year increase. Even with TSMC's planned expansion to 200k wafers/month and non-TSMC capacity, a shortage is expected, particularly for high-end CoWoS-L used by Nvidia. This scarcity grants TSMC pricing power. Key near-term catalysts include improved ABF substrate supply, rising CPU demand, and Nvidia's Rubin Ultra production. The report highlights winners across the supply chain, including MediaTek (Google TPU partner) and ASE Group.

marsbit06/25 03:56

Report Analysis: TSMC's AI Revenue to Double by 2027, CoWoS Capacity Remains a Bottleneck

marsbit06/25 03:56

活动图片

Indepth Researchright-arrow

Project Updatesright-arrow