BreakpointNVIDIA announces a 64GB DGX Spark configuration for running AI agents, inference, fine-tuning and data science workloads locally. The post explains how two systems can pool 128GB of memory through NVIDIA Sync Cluster Assistant, delivering up to 1.7x performance for larger models and workloads.
Nvidia engineering blog
NVIDIA announces 25 games joining GeForce NOW throughout October, including six available this week and The Witcher 3: Wild Hunt – Remastered. The update also highlights a CONTROL Resonant reward and RTX 5080-class cloud performance for Ultimate members.
NVIDIA describes how Blackwell GPUs and inference optimizations accelerate OpenAI’s GPT-6 Astra Ultrafast mode, delivering up to 8x faster token generation than Astra Standard. The announcement highlights benefits for coding agents, tool-use workflows, and more flexible reuse of infrastructure across AI workloads.
The post presents productivity, durability, and fungibility as the three drivers of AI-factory return on investment. It explains how full-stack co-design, CUDA compatibility, workload flexibility, and extended hardware lifecycles can improve throughput, utilization, and demand across NVIDIA systems.
NVIDIA is accepting worldwide applications for its 2027–2028 Graduate Fellowship Program, offering doctoral students grants, mentorship and technical support. Awards reach $60,000, and applicants must complete an NVIDIA research internship in summer 2027.
NVIDIA and CoreWeave announce production availability of Vera Rubin NVL72 systems, Vera CPUs, and Forge for training, evaluating, and improving AI agents. The post reports benchmark results including 4.8x higher token throughput, more than 3x faster sandbox startups, and lower-cost serverless reinforcement learning.
NVIDIA and research partners released AlphaFold2-predicted 3D structures for protein complexes from more than 2,800 viruses through the AlphaFold Database. The post explains how BioNeMo accelerates large-scale GPU inference and introduces an open structure-prediction pipeline for researchers studying future pandemic threats.
NVIDIA announces the arrival of CONTROL Resonant and eight other games on GeForce NOW, alongside a limited-time Ultimate membership bundle. The post also previews upcoming support for Googlebooks and highlights cloud-streaming features such as RTX 5080-class performance, DLSS 4, ray tracing, and HDR streaming.
NVIDIA validation engineer Sakeena Fiza explains how data-center systems are brought up, tested, and debugged before mass production. The profile highlights failure analysis across hardware, firmware, software, thermal behavior, signaling, and manufacturing conditions at rack and cluster scale.
NVIDIA’s AI Day Singapore highlights how Southeast Asian governments, enterprises and research organizations are deploying region-specific AI. The showcase covers Nemotron model fine-tuning, agentic AI, multimodal video search, localized datasets, smart-city applications and Sea Limited’s adoption of the Vera Rubin platform.
NVIDIA Isaac ROS 5.0 adds agent-ready workflows, ROS Lyrical and Ubuntu 24.04 support, and GPU-accelerated tools for robotics development. The release includes FoundationPose improvements, reusable perception and manipulation skills, and Jetson support spanning development through edge deployment.
NVIDIA highlights Egypt’s expanding AI ecosystem and Africa’s shift toward locally hosted accelerated computing. The post covers new AI factories, GPU infrastructure investments, developer-training growth and startups applying AI across healthcare, robotics, finance and language technologies.
NVIDIA outlines a layered safety model for deploying physical AI in autonomous vehicles and robots, spanning hardware, software, AI behavior, operating environments and lifecycle validation. The post describes Halos components, simulation and synthetic-data testing, safety standards, certification processes and industry integrations.
The post presents AI-agent security as a layered engineering discipline, covering identity, authorization, sandboxing, network controls, protected logging and human approval. It also emphasizes repeatable security testing, accountable ownership and evidence-based validation across the agent lifecycle.
NVIDIA introduces DSX Ready, a qualification program for power and cooling products used in AI factories. The launch covers battery energy storage systems and cooling distribution units, with defined testing, review, and self-qualification processes that reduce integration risk while preserving the need for site-level engineering.
NVIDIA highlights five companies applying AI, digital twins, accelerated computing and advanced energy systems to clean-energy challenges. The examples span grid interconnection, nuclear operations, recycled-battery power for AI data centers, advanced fission and fusion, including a reported reduction in grid-connection evaluations from 30–45 days to two minutes.
NVIDIA highlights 11 games joining GeForce NOW, led by the launch of the creature-catching adventure Aniimo. The update also covers path tracing and DLSS 4.5 enhancements for 007 First Light, supported devices, and a limited-time game sale.
Researchers at the University of Manchester adapted NVIDIA Earth-2 generative models to forecast UK-wide air pollution at 2–3 kilometer resolution. The workflow trained in two days on Isambard-AI and supports inference on a desktop DGX Spark, enabling faster scenario modeling and potential real-time health applications.
NVIDIA reports MLPerf Inference v6.1 results for Vera Rubin and GB300 NVL72 systems, including up to 3.7x higher throughput and 99% multi-rack scaling efficiency. The post explains how NVFP4 precision, disaggregated serving, expert parallelism, high-speed interconnects and software optimization contribute to inference performance.
Emerald AI, Google and NVIDIA introduce the AI Energy Management Alliance, a coalition focused on grid-responsive AI data centers. The post outlines how workload shifting, storage, generation and standardized performance requirements could improve interconnection speed, reliability and grid capacity utilization.