The auto-updated technology brief
LLM training workloads face GPU memory limits due to competing demands on high-bandwidth memory. This article explains how host offloading can alleviate these bottlenecks in JAX-based training.
Telecom operators have invested over $240B in wireless spectrum. AI-native RAN and NVIDIA AI Aerial aim to maximize spectral efficiency, enhancing network capacity and resilience.
This article explores GPU-initiated communication techniques to enhance molecular dynamics simulations, enabling researchers to efficiently observe atomic behavior in demanding computational workloads.
NVIDIA introduces a LangChain deep agents harness profile for the Nemotron 3 Ultra to enhance AI agent performance by balancing accuracy and cost through fine-tuning smaller open models.
NVIDIA demonstrates how autonomous coding AI agents can improve vision reasoning models to over 90% accuracy with minimal manual effort, streamlining adaptation to production video tasks.
Coding AI agents are becoming practical operators for long-running machine learning workflows, enabling tasks like repository inspection, runtime setup, issue resolution, experiment launching, monitoring, and result summarization, which is crucial for reinforcement learning research.
NVIDIA explores estimating probabilities of rare, high-impact events using guided generative models to improve efficiency over brute-force Monte Carlo sampling.
NVIDIA introduces NVLink as a high-performance scale-up network designed to meet the growing demands of AI factories, enabling faster deployment of AI compute infrastructure for larger, more complex workloads.
NVIDIA's GB300 NVL72 achieved a world record in pre-training the DeepSeek-V3 671B model using mixture of experts (MoE), reaching 1,648 TFLOPs per GPU and advancing large-scale AI training efficiency.
Healthcare robotics faces unique challenges due to limited data and experimentation opportunities. NVIDIA explores GPU-native medical physics simulation to advance robotic surgery development.
NVIDIA's Nemotron 3 Ultra model excels in accuracy and efficiency for agentic register transfer level (RTL) coding, accelerating chip design by leveraging AI agents and LLMs to improve code generation and verification.
NVIDIA introduces the Vera CPU featuring Olympus cores optimized for maximum single-thread performance to enhance agentic AI workloads, improving per-agent responsiveness and AI factory throughput.