32 articles tagged with #gpu. AI-curated summaries with sentiment analysis and key takeaways from 50+ sources.
Nvidia is facing a certified class action lawsuit over alleged securities fraud related to under-disclosure of cryptocurrency mining revenue. A U.S. federal judge has certified the case against Nvidia and its CEO after years of legal proceedings.
Nvidia stock rises following three major developments: Arm's new AI CPU launch, a massive $50B+ AWS order for 1 million GPUs, and resumed chip sales to China potentially worth $32B annually. These combined deals represent approximately $82B in new revenue streams for the semiconductor giant.
Nvidia stock gains momentum following Arm's AI CPU launch, a massive 1 million GPU order from AWS, and the restart of China chip operations, collectively revealing over $82 billion in previously undisclosed revenue streams.
Justitia is a new scheduling system for task-parallel LLM agents that optimizes GPU server performance through selective resource allocation based on completion order prediction. The system uses memory-centric cost quantification and virtual-time fair queuing to achieve both efficiency and fairness in LLM serving environments.
Nvidia announced the Groq 3 LPU at GTC 2024, its first chip specifically designed for AI inference rather than training, incorporating technology licensed from startup Groq for $20 billion. The chip uses SRAM memory integrated within the processor to achieve 7x faster memory bandwidth than traditional GPUs, optimizing for the low latency required for real-time AI inference applications.
HIVE Digital is transitioning away from Bitcoin mining due to hostile Swedish tax rules and halving risks, instead quadrupling its Canadian AI data center capacity to focus on contracted GPU revenue. This shift represents an acknowledgment that the traditional Bitcoin-only mining model faces significant challenges.
Researchers conducted comprehensive benchmarks of LLM inference on AMD Instinct MI325X GPUs, testing models from 235B to 1 trillion parameters. The study reveals that architecture-aware optimization is critical, with different model types requiring specific configurations for optimal performance on AMD hardware.
Researchers developed ES-dLLM, a training-free inference acceleration framework that speeds up diffusion large language models by selectively skipping tokens in early layers based on importance scoring. The method achieves 5.6x to 16.8x speedup over vanilla implementations while maintaining generation quality, offering a promising alternative to autoregressive models.
IREN announced plans to expand processing capacity by 50% through ordering over 50,000 Nvidia GPUs and filed for a potential $6 billion at-the-market share offering. The news caused the company's stock to decline in pre-market trading despite the significant capacity expansion plans.
TAO is a new verification protocol that enables users to verify neural network outputs from untrusted cloud services without requiring exact computation matches. The system uses tolerance-aware verification with IEEE-754 bounds and empirical profiles, implementing a dispute resolution mechanism deployed on Ethereum testnet.
GeneZip is a new DNA compression model that achieves 137.6x compression with minimal performance loss by recognizing that genomic information is highly imbalanced. The system enables training of much larger AI models for genomic analysis using single GPU setups instead of expensive multi-GPU configurations.
The memory chip shortage is driven by massive AI demand for high-bandwidth memory (HBM), causing DRAM prices to surge 80-90% this quarter. While major AI companies have secured supply through 2028, other industries face scarce supply and inflated prices that won't normalize for years.
Nvidia's new Vera Rubin technology significantly reduces AI computing costs, potentially threatening decentralized GPU networks like Render that rely on expensive and underutilized computing resources. This development could disrupt the business model of crypto-based distributed computing platforms.
AMD and OpenAI announced a multi-year strategic partnership to deploy 6 gigawatts of AMD Instinct GPUs for OpenAI's AI infrastructure, starting with 1 gigawatt in 2026. This represents a significant expansion of AI computing capacity to support next-generation AI development and global innovation.
OpenAI, NVIDIA, and Nscale have launched Stargate UK, a sovereign AI infrastructure partnership that will deliver up to 50,000 GPUs and create the UK's largest supercomputer. This initiative aims to accelerate national AI innovation, enhance public services, and drive economic growth through dedicated AI infrastructure.
Bernstein analysts warn that Nvidia could severely impact Supermicro by reducing GPU supply access, despite Supermicro's success being tied to Nvidia's $4 trillion valuation. The dependency relationship gives Nvidia significant leverage to potentially devastate Supermicro's hardware business at any time.
A large-scale study of prompt compression techniques for LLMs found that LLMLingua can achieve up to 18% speed improvements when properly configured, while maintaining response quality across tasks. However, compression benefits only materialize under specific conditions of prompt length, compression ratio, and hardware capacity.
Andrej Karpathy has open-sourced 'Autoresearch', a minimalist 630-line Python tool that enables AI agents to autonomously conduct machine learning experiments on single NVIDIA GPUs. The tool is derived from the nanochat LLM training core and represents a streamlined approach to automated ML research.
OrbitFlow is a new KV cache management system for long-context LLM serving that uses adaptive memory allocation and fine-grained optimization to improve performance. The system achieves up to 66% better SLO attainment and 3.3x higher throughput by dynamically managing GPU memory usage during token generation.
Nvidia is investing $4 billion total in photonics technology, splitting $2 billion each between Lumentum and Coherent to develop optical components for AI data centers. This strategic investment aims to improve energy efficiency, data transfer speeds, and bandwidth for next-generation AI infrastructure.
Google has announced Gemma 3, positioning it as their most capable AI model that can run on a single GPU or TPU. This represents a significant advancement in making powerful AI models more accessible for individual developers and smaller organizations.
AMD has partnered with Hugging Face to provide out-of-the-box acceleration for Large Language Models on AMD GPUs. This collaboration aims to make AMD's GPU hardware more accessible for AI developers and researchers working with popular open-source AI models.
Hugging Face and AMD have announced a partnership to optimize and accelerate state-of-the-art AI models for both CPU and GPU platforms. This collaboration aims to improve performance and accessibility of AI models across AMD's hardware ecosystem.
A company has released highly-optimized GPU kernels for block-sparse neural network architectures that can run orders of magnitude faster than existing solutions like cuBLAS or cuSPARSE. These kernels have achieved state-of-the-art results in text sentiment analysis and generative modeling applications.
Niv-AI has emerged from stealth mode with $12 million in seed funding to develop technology that measures and manages GPU power surges. The company aims to optimize GPU power performance, addressing a critical infrastructure challenge in AI computing.
