Nvidia engineering blog

Loading
Ian Buck Kicks Off NVIDIA’s Week at SC26 as Science Enters the Age of Agents
NVIDIA's Ian Buck used SC26 to unveil Nemotron for Science, an open model and chemistry dataset for running research agents locally, along with the Rubin HPC GPU and expanded Vera CPU platform. The talk covers agent examples in GROMACS repair and WRF porting, plus quantum error-correction work including magic state cultivation and a CUDA-Q decoding algorithm. Most results are presenter claims without published methodology.
How the World’s Most Advanced AI Infrastructure Supply Chain Invents the Future
NVIDIA's operations leader Deb Shoquist describes how the company builds supplier relationships to scale AI infrastructure production. The post walks through a GB200 compute board test-handler throughput failure, the cross-team response that followed, and a traceability process that cut chip-to-system lead time from five weeks to two. It is written by NVIDIA about its own manufacturing operations.
CUDA at 20: How the Platform Transformed Scientific Computing
NVIDIA marks 20 years of CUDA at SC26 by recognizing researchers and developers whose work used GPU computing in molecular biology, deep learning, and protein structure prediction. The post recounts the platform's history from the 2006 launch and cites the more than 11 million developers who now use it, but offers little technical detail a practitioner could apply.
NVIDIA CUDA-X Libraries Give Agents the Tools to Accelerate Scientific Discovery
NVIDIA announces expanded CUDA-X libraries, including cuPhoton, cuPotential, cuSpa, Daviator, DAQIRI, rapids-singlecell and HMMER, for scientific workloads in astrophysics, fusion, materials science, drug discovery and single-cell biology. The post reports speedups for each, such as an Event Horizon Telescope correlation step cut from over a month to eight minutes, and it positions the libraries for use by AI agents.
NVIDIA Expands Slurm to Bring Quantum Computing Into Supercomputing
NVIDIA announces a Slurm release that adds support for quantum processors (QPUs) through the QRMI and QDMI interfaces, letting QPU builders expose one integration across supercomputing centers. The post also outlines CUDA-Q updates, including a quantum error-correction decoder reported as 13,000x faster than the industry standard.
Lambda Boosts AI Factory Resiliency and Efficiency With NVIDIA Photonics
Lambda is deploying NVIDIA Photonics co-packaged optics switches to connect GB300 NVL72 racks with more than 10,000 GPUs. The post reports about 4 to 10 percent latency reduction, a median optics temperature of 36°C versus 51°C for pluggable optics, and field-replaceable external light sources, with co-packaging removing retimer chips from the signal path.
NVIDIA Sweeps MLPerf Training and Endpoints as Vera Rubin Debuts With Leading Performance
NVIDIA reports that its Vera Rubin platform made its MLPerf debut, sweeping every MLPerf Training v6.1 benchmark and posting large throughput-per-watt gains in the first MLPerf Endpoints submission. The post cites a DeepSeek-V3 scaling result of about 97% efficiency from 256 to 1,024 GPUs, with the networking stack behind it. Many figures are provisional and vendor-reported, so readers should check them against MLCommons verification.
NVIDIA DGX Spark Developer Kit Turns a Desk Into a Supercomputing Lab
NVIDIA introduces the DGX Spark HPC Developer Kit, a four-node cluster of DGX Spark systems linked directly in a ring over ConnectX-7 networking without a switch, giving researchers local capacity for multi-node scientific and AI workloads. The post outlines the 512GB combined memory and 20TB NVMe storage, NVFP4 deployment of models up to 700B parameters, and the forthcoming HPC SDK playbook for building and debugging multi-node GPU applications. It also cites example workloads from academic...
How AI Is Helping Care Teams Get Patients the Right Care, Sooner
NVIDIA's roundup of healthcare AI deployments highlights Dyania Health's Synapsis AI, a clinical reasoning engine that analyzes patient records across notes, reports, and lab results to surface traceable clinical signals. The post describes on-premises processing inside a hospital firewall, built with NVIDIA NeMo, TensorRT, and TensorRT-LLM on local GPUs. It also summarizes a Cleveland Clinic study of over 7,800 aortic valve replacement patients, with clinicians retaining final decisions.
How AI Helps Recyclers Turn Mixed Waste Into Reusable Material
NVIDIA's blog profiles four Inception startups that use vision AI and robotics to sort recyclables, plastics, and used textiles on conveyor belts. The post covers edge inference on Jetson modules, GPU model training, and real-time constraints at belt speeds of up to 4 meters per second. Reported results include 95% recognition accuracy on a pilot line and claimed efficiency gains of 10% to 30% at facilities using Analyzer data.
NVIDIA Expands Open Source Commitment With OpenShell Donation and New AI Infrastructure
NVIDIA is contributing its OpenShell runtime for securing and governing autonomous agents to the Cloud Native Computing Foundation as a Sandbox project, and is contributing to Agent Substrate, a Google-led framework for running long-lived agent workloads on Kubernetes. The post also introduces NVIDIA Kubernetes Engine, Kubernetes Recipes, and Snapshot, where Snapshot restores checkpointed inference replicas without model reload or warmup. Design-partner benchmarks report startup reductions from...
NVIDIA Dynamo Recipes Bring Korea’s Leading Open Models to Production
NVIDIA announces Dynamo-optimized deployment recipes for four Korean open models: A.X K2, K-EXAONE 2.0, Solar Open 2, and Motif-3. The post gives a general description of multi-GPU inference routing and claims up to 2x performance on B200 GPUs, but it provides no architecture, implementation, or benchmark method.
Korea Advances Physical AI Across Industries With NVIDIA
NVIDIA outlines its collaborations with Korean industry leaders on physical AI, spanning semiconductor simulation with Samsung, level 4 autonomous driving with Kakao Mobility, and robot learning with LG Electronics. Samsung's results include a roughly 1,600x speedup on a chip-scale homogenization workflow by combining AI-predicted material properties, coarser grids, and GPU acceleration. The post is largely a partnership roundup written by the vendor.
From Chips to Skills, NVIDIA and Korea Expand Sovereign AI Collaboration
NVIDIA and Korea's Ministry of Science and ICT are expanding their sovereign AI collaboration across semiconductors, data center infrastructure, talent, startups, and models. The announcement covers NVIDIA training programs, NVLink Fusion ecosystem participation for Korean NPU and XPU companies, and Dynamo-based inference work with selected consortia and foundation model builders. The body describes partnerships and programs without technical implementation details.
NVIDIA Announces New Planned Investments in Korean Startups at the MSIT-NVIDIA AI Startup Partnership Forum
NVIDIA announced an investment in Korean interconnect startup Point2 and planned investments in Motif Technologies, RLWRLD, and Upstage during the MSIT-NVIDIA AI Startup Partnership Forum in Seoul. The post profiles the startups' AI work across foundation models, robotics, and enterprise document AI, but it covers corporate investment activity rather than engineering details.
Korea’s AI Ecosystem Advances and Secures Agentic AI With NVIDIA
NVIDIA reports that Hyundai Motor Group is exploring NemoClaw blueprints and the OpenShell runtime, and that AIM Intelligence is integrating its Starfort platform with OpenShell for secure enterprise agents. The post also notes that LG AI Research opened API access to K-EXAONE-2.5, which was trained on Nemotron datasets including the synthetic Korean persona dataset. Most of the content describes partnerships and integrations at a high level, with little technical detail.
Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in Simulation
NVIDIA and its partners describe how simulation is being used to train robots on precision manipulation tasks, ahead of CoRL. The post covers GR00T 2 world action model results, Isaac Lab 3.0 with the Newton physics engine, and a Success-Guided Sampling method that cut GPU time roughly 39x on a 20-task assembly benchmark. It also introduces the DexBench and VABAR evaluation benchmarks for industrial robot dexterity.