H100 GPU
PulseAugur coverage of H100 GPU — every cluster mentioning H100 GPU across labs, papers, and developer communities, ranked by signal.
5 day(s) with sentiment data
-
New AR diffusion methods boost long video generation quality and speed
Two new research papers propose novel autoregressive (AR) methods for generating long videos using diffusion models. Trajectory Forcing (TraF) leverages the denoising trajectories of previous video chunks as context, im…
-
New framework "Any Forcing" boosts autoregressive video generation stability
Researchers have introduced "Any Forcing," a novel framework designed to enhance autoregressive video generation. This approach aims to provide flexibility in sampling steps for balancing quality and efficiency, while a…
-
New EDU-Jev models offer faster, more accurate educational decisions
Researchers have developed EDU-Jev, a new family of single-pass decision models designed for educational applications. These models, available in 0.8B, 2B, and 4B parameter sizes, output probability distributions over p…
-
New "Know, then Draw" paradigm boosts visual generation efficiency
Researchers have introduced a novel learning paradigm called "Know, then Draw" that combines autoregressive pretraining for semantic understanding with diffusion post-training for high-fidelity rendering in visual gener…
-
AI's hidden environmental cost: Lifecycle of hardware revealed · 2 sources tracked
A new analysis reveals that the environmental cost of AI extends far beyond data center energy consumption, encompassing the entire lifecycle of hardware. The production of a single H100 GPU by TSMC involves significant…
-
LidarVerse introduces unified real-time LiDAR simulation model
Researchers have introduced LidarVerse, a novel real-time unified LiDAR world model designed for sensor simulation across various tasks and sensors. This model integrates both reconstruction-based and generative simulat…
-
LidarVerse: Unified LiDAR World Model for Real-Time Sensor Simulation
Researchers have introduced LidarVerse, a novel real-time unified LiDAR world model designed for sensor simulation across various tasks and sensors. This model integrates both reconstruction-based and generative simulat…
-
SpaceX taps NVIDIA for ambitious orbital AI data center project
SpaceX plans to launch its ambitious Starmind project, an orbital data center designed to harness solar energy for AI computations, utilizing NVIDIA's advanced GPUs. The project aims to overcome terrestrial limitations …
-
Mingxin Technology unveils GPU platform acceptance framework beyond benchmarks
Mingxin Technology has developed a comprehensive GPU compute platform acceptance framework that goes beyond standard benchmark testing. This framework addresses the gap between benchmark performance and real-world clust…
-
Depth-Aware Analysis Reveals Sensitivity in Qwen MoE Model Layers
Researchers have developed a depth-aware sensitivity analysis method for Mixture-of-Experts (MoE) models, specifically applied to the Qwen3.6-35B-A3B model. Their findings indicate that early and middle layers are highl…
-
Avatar-Forever framework enables real-time infinite avatars with parallel training
Researchers have introduced Avatar-Forever, a novel framework designed for high-quality, real-time infinite interactive avatars. This system decouples the training of generation efficiency and long-horizon robustness in…
-
New pipeline automates industrial device configuration using LLMs and ontologies
Researchers have developed SysName, a pipeline designed to automate the configuration of industrial fieldbus devices. This system uses a hybrid retrieval index combined with an ontology graph derived from ECLASS, AAS, a…
-
New ELMOD language model offers efficient German inference on mobile devices
Researchers have developed ELMOD, a 2.7 billion parameter German language model optimized for efficient on-device inference. The model was trained using publicly available data and specialized pre-processing techniques …
-
FlashDecoder: Transformer-based video decoder achieves real-time generation
Researchers have developed FlashDecoder, a novel Transformer-based video decoder designed for real-time generation. Unlike existing models that use slow and memory-intensive 3D convolutional decoders, FlashDecoder proce…
-
RynnWorld-Teleop uses generative world models for robotic data synthesis
Researchers have introduced RynnWorld-Teleop, a novel system that utilizes generative world models for digital teleoperation in robotics. This approach replaces physical robot interaction with synthesized training data,…
-
Sakana AI's Fugu-Ultra agent autonomously optimizes ML training and text reconstruction
Sakana AI has developed Fugu-Ultra, an AI agent that autonomously improves machine learning training recipes. In one experiment, Fugu-Ultra iteratively edited training code and ran 123 experiments over 14 hours on a sin…
-
StreamChar framework enables real-time character audio-video generation
Researchers have introduced StreamChar, a novel framework for generating character audio and video in real-time streaming scenarios. This system decouples long-horizon orchestration from short-window audio-video denoisi…
-
New frameworks enable real-time streaming audio-video generation and narration
Researchers have developed two new frameworks, StreamChar and FlowNar, designed for real-time audio-video generation and narration for long-form videos. StreamChar utilizes an LLM-based orchestrator and a joint audio-vi…