graphics processing unit
PulseAugur coverage of graphics processing unit — every cluster mentioning graphics processing unit across labs, papers, and developer communities, ranked by signal.
- competes with central processing unit 70%
- used by H.1000 Gnome 70%
- competes with application-specific integrated circuit 70%
- used by Dohuk Polytechnic University 70%
- used by Vulkan 70%
- uses data processing unit 70%
- competes with Cerebras Systems 70%
- competes with Tensor Processing Unit 60%
- uses central processing unit 50%
- used by central processing unit 50%
- used by Cerebras Systems 50%
- competes with data processing unit 50%
18 天有情绪数据
-
Google将向部分客户出售其TPU，这些客户也青睐大G的GPU
Alphabet宣布将其2026年的资本支出指导大幅提高至1800亿至1900亿美元，这得益于对人工智能计算资源的空前需求。公司CFO强调了Google Cloud的强劲增长，其增长得益于人工智能解决方案和创纪录的积压订单，并表示Google将开始向特定客户销售其定制TPU。此举旨在实现收入来源多元化并资助未来的芯片研究，预计对资产负债表的影响将在2027年更为显著。
-
Visual explainers detail GPU's AI role and embedding vector meaning
A visual explainer details why Graphics Processing Units (GPUs) are highly effective for artificial intelligence tasks, highlighting their strengths in matrix multiplication, parallel processing, memory bandwidth, and b…
-
FloatSOM framework accelerates distributed Self-Organizing Maps with flexible topologies
Researchers have developed FloatSOM, a new framework designed for large-scale Self-Organizing Map (SOM) analysis that overcomes memory limitations on GPUs. This framework enables multi-GPU execution and supports out-of-…
-
GPU firmware lags behind hardware, throttling AI workloads
The article argues that current GPU firmware is outdated, relying on early 2000s logic to manage modern AI workloads. This outdated firmware is identified as a bottleneck, potentially throttling the performance of advan…
-
GPU shortage becomes AI's biggest bottleneck, spurring efficiency focus
The escalating demand for Graphics Processing Units (GPUs) has become the primary constraint for the advancement of artificial intelligence. In response, organizations are increasingly adopting strategies focused on dev…
-
AI era demands flexible data center investments, moving beyond old refresh cycles
The AI era is forcing a significant shift in data center infrastructure investments, moving away from traditional refresh cycles. Companies are now navigating multiple, often misaligned, technology lifecycles for comput…
-
Stanford researchers develop new hardware to efficiently process sparse AI models
Researchers at Stanford University have developed a novel hardware chip designed to efficiently process sparse AI models. Sparsity, where most AI model parameters are zero, offers significant computational savings but i…
-
AHASD architecture boosts LLM speculative decoding on mobile devices
Researchers have developed AHASD, a novel asynchronous heterogeneous architecture designed to optimize large language model (LLM) inference on mobile devices. This architecture employs task-level decoupling for parallel…
-
DeepSeek V4 prioritizes batch invariance, sacrificing GPU efficiency for stability
DeepSeek V4's technical report reveals a core design choice of "batch invariance" to ensure consistent outputs across different batch configurations and processing pipelines. This feature is crucial for maintaining repr…
-
New GPU framework accelerates quantum state calculations for complex systems
Researchers have developed QiankunNet-cuSCI, a novel framework that fully accelerates the NNQS-SCI method for solving complex quantum systems using GPUs. This new approach addresses the scalability limitations of previo…
-
MTServe system optimizes generative recommendation models with hierarchical caches
Researchers have developed MTServe, a new system designed to make generative recommendation models more efficient. These models, while powerful, are computationally expensive due to the need to process extensive user hi…
-
NVIDIA and Siemens Healthineers develop AI for adaptive ultrasound imaging
NVIDIA and Siemens Healthineers have developed a new AI model called NV-Raw2Insights-US that processes raw ultrasound data directly, rather than relying on traditional image reconstruction methods. This approach allows …
-
DeepSeek V4 release sparks surge in Chinese semiconductor stocks, boosting domestic AI computing power
DeepSeek V4's release has significantly boosted China's A-share semiconductor market, with sectors like GPU and semiconductor equipment experiencing a surge. This rally is attributed to V4's compatibility with Huawei's …
-
Google invests $10B in AI firm Anthropic; Singtel and Mistral AI plot GPU, AIaaS moves
Google is investing $10 billion into the AI firm Anthropic, a significant move in the competitive AI landscape. Additionally, Hershey is exploring the use of AI agents to address its business challenges, and Singtel is …
-
LiveRamp integrates NVIDIA GPUs for 15x faster AI model training
LiveRamp has integrated NVIDIA's GPU infrastructure into its clean room environments. This enhancement is designed to significantly accelerate model training and inference processes. The integration aims to provide bran…
-
New ML-based GPU caching algorithm LCR boosts LLM inference speed
Researchers have developed a new GPU caching algorithm called Learning-Augmented LRU (LALRU) designed to improve efficiency during AI inference. This algorithm integrates learned predictions with caching policies to ens…
-
New techniques ZipCCL and FlashOverlap accelerate LLM training by optimizing communication
Researchers have developed ZipCCL, a lossless compression library designed to accelerate the distributed training of large language models by addressing communication bottlenecks. The library utilizes novel techniques l…
-
AI chip startups challenge Nvidia in inference era, as Google dominates compute
The AI chip industry is seeing a resurgence of startups focusing on inference, a diverse workload that differs significantly from model training. Companies like Groq, Cerebras Systems, SambaNova, and Lumai are developin…
-
Qwen3.6-35B 模型量化显示 FP8 质量不如 INT8，NVFP4 是谎言
Reddit 的 LocalLLaMA 社区的一位用户分享了关于 Qwen3.6-35B 模型的研究结果，重点关注了 Kullback-Leibler (KLD) 散度指标在 INT8、FP8 和 NVFP4 等不同量化格式下的表现。使用修改后的 VLLM 框架进行的分析表明，FP8 和 NVFP4 格式虽然可能速度更快，但质量可能不如 INT8。用户强调，量化格式的选择应与具体用例相匹配，平衡准确性、速度和 GPU 兼容性。
-
New HGQ-LUT and da4ml methods speed up DNN training and FPGA deployment
Researchers have developed HGQ-LUT, a new method for training lookup-table (LUT) based neural networks that significantly speeds up the training process, making it over 100 times faster on modern GPUs. This approach int…