Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #3

NVIDIA

Related event timeline, sources and context from the news index.

EVENT TIMELINE

130

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA launches open-source "NVIDIA Recovery" to automate distributed training recovery

NVIDIA announced NVIDIA Recovery, an open-source workload-level service in the DSX OS portfolio that detects training interruptions via application logs, GPU metrics, and scheduler signals and runs a four-stage loop to attribute failures, isolate nodes, and restart or stop jobs. The service integrates with Slurm via a Helm-deployable Advisor Controller, starts in monitor mode by default, provides an audit trail and Recovery Console, and the GitHub repo and quickstart allow local deployment without physical GPUs; production monitoring reported ~8,550 anomaly detections over 3 months with mean detection and recovery-action latencies of 69s and 94s respectively.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA AICR captures Day‑One Kubernetes configs for Vera Rubin NVL72

NVIDIA’s AI Cluster Runtime (AICR), part of the DSX AI Factory Platform, provides reusable recipes and validation bundles to configure Kubernetes clusters for the new Vera Rubin NVL72 compute trays. NVIDIA validated an end-to-end deployment using host-installed R615.23 driver, GPU Operator v26.7.0 and DRA v0.5.0; DOCA 3.3.0 and ConnectX-9/BlueField-4 hardware are noted as the testbed environment, and MIG profiles on Vera Rubin were not covered in this validation.

6.0

MODELS · 1 SOURCE · NVIDIA Developer

NVIDIA releases Cosmos Framework with Cosmos 3 Nano simulation recipes and NIM microservices

NVIDIA announced the Cosmos Framework, a modular platform of world foundation models and developer tools for physical AI that supports training, RL, optimization and inference across models including Cosmos 3, Alpamayo, GR00T, Qwen and pi0.5. The release includes Cosmos 3 Nano simulation recipes and example checkpoints (PyTorch/vLLM-Omni), NIM microservices for accelerated inference (quantization, speculative decoding, KV-cache-aware routing) with ~2–3x speedups on H200 GPUs, and guidance on GPU memory use (Cosmos3-Nano: ~58 GiB BF16; ~44 GiB FP8 with offloading); a livestream is scheduled for Nov. 5.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA packages Dynamo 1.4.2 into validated Kubernetes runtime recipes with AICR

NVIDIA's DSX AI Factory Platform adds Dynamo 1.4.2 and its cluster dependencies into reproducible, validated runtime recipes via the NVIDIA AI Cluster Runtime (AICR). AICR pins component versions, captures GPU/DR A, scheduling, networking and monitoring requirements, offers a five-step render/deploy/validate flow (snapshot, generate recipe, render bundle, deploy via Helmfile/GitOps, validate), and documents routing options (TCP + ZMQ) including a Gateway API / EPP ingress path; NATS is now opt-in.

6.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA demonstrates TensorRT Edge-LLM workflow to run Nemotron-3.5 Lightning on DGX Spark 64GB

NVIDIA published a tutorial showing how to use TensorRT Edge-LLM (examples tested with Edge-LLM 0.11.1 and TensorRT 10.16.1.11) on a DGX Spark 64GB to prototype, fine-tune and test embedded AI models. The trtedge CLI downloads models and builds optimized TensorRT engines, supports serving via a local OpenAI-compatible API, and can enable DSpark speculative decoding (reported to raise mean generation from 73.4 to 126.9 tokens/s on HumanEval prompts while keeping peak memory at 27.9GB).

7.0

MODELS · 1 SOURCE · Hugging Face

NVIDIA’s 'MonsterHouse' demo runs Nemotron Voice Agent with Bolt E4B 4GB and Magpie TTS

MonsterHouse is an interactive Halloween projection built by an NVIDIA employee that uses the Nemotron Voice Agent pipeline: nemotron-speech-streaming-en-0.6b.gguf for ASR, Nemotron 3.5 Bolt E4B 4GB as the local LLM (with a LoRA personality adapter), Magpie TTS Zeroshot for voice, Audio2Face-3D for lip animation, and Nemotron 3 Nano Omni plus YOLO/ByteTrack for costume description and tracking, all running on a single workstation.

4.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA publishes NemoClaw blueprints to assemble secure, optimized agent stacks

NVIDIA introduced NemoClaw, a collection of open blueprints that assemble a full-stack agent platform by combining tuned agent harnesses, Nemotron models, NeMo optimization tools (including NeMo Relay), and the OpenShell secure runtime. The blueprints support OpenClaw, Hermes Agent, and LangChain Deep Agents, include a LangChain-tuned Nemotron 3 Ultra profile (near-frontier performance at about 10x lower cost per run), and provide GitHub examples, a quickstart, and a DLI course.

7.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA publishes gds-diag diagnostic tool for GPUDirect Storage in Magnum IO repo

gds-diag is an open-source command-line utility in the NVIDIA Magnum IO GitHub repository that assesses GPUDirect Storage (GDS) environments, runs grouped end-to-end and targeted checks (pre-install, post-install, container, mount), and suggests mitigations. An optional agent skill adds a natural-language interface that orchestrates multi-step diagnostics and applies model-guided reasoning via coding assistants (examples mention Claude and Codex).

4.0

MODELS · 1 SOURCE · AWS Machine Learning

NVIDIA publishes Nemotron 3.5 Super VL (120B MoE) and provides AWS EKS / SageMaker deployment guide

NVIDIA released Nemotron 3.5 Super VL, an open multimodal 120B-parameter mixture-of-experts (MoE) model that activates about 12B parameters per token and is distilled from Nemotron 3 Ultra. NVIDIA published the model weights, training datasets, and post-training recipes and published reference implementations and an aws-do-eks walkthrough showing how to deploy Nemotron models on Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS; the model is validated on AWS GPU instances and integrates with major agent frameworks via the NVIDIA Agent Toolkit.

7.0

STARTUPS · 1 SOURCE · Cohere

Cohere announces initial close of Series E targeting CAD $7B to build frontier AI outside US/China

Cohere announced an initial close of its Series E funding round, targeting up to CAD $7 billion, led by Schwarz Group with participation from investors including Bell Ventures, Bullhound Capital, Inovia Capital, NVIDIA, Prosperity7 Ventures and the Government of Canada. The company says it will use the funds to build frontier AI models and a secure platform for institutions to run and own AI under legal jurisdictions outside the U.S. and China, arguing democracies need more independent builders.

8.0

MODELS · 5 SOURCES · Mistral AI · WIRED AI · Hugging Face · The Decoder · TechCrunch AI

Mistral releases Mistral Large 4 (Le Chonk) weights — 1T-parameter open multimodal model

Mistral released the weights for Mistral Large 4 (Le Chonk), a 1‑trillion‑parameter natively multimodal model, claiming leading open‑weight results on coding and defensive cybersecurity benchmarks and new three-mode reasoning options to trade off cost and performance. The company says the model was trained on European infrastructure, is fully open‑weight, shows gains on FrontierCode (19.7%) and artificial analysis indexes (coding 48%, cyber 51%), and achieves a 96.6% average refusal rate on malicious cyber prompts while supporting self-hosted deployment.

8.0

RESEARCH · 1 SOURCE · NVIDIA Developer

TF32x3 and BF16x9 speed up Point Edge Transformer MLIP inference while keeping FP32-like accuracy

Evaluation on the Point Edge Transformer (PET) from Lab COSMO (EPFL) shows NVIDIA’s BF16x9 on Blackwell GPUs speeds large GEMMs by 2.2–2.5x (17,280-atom) and 1.2–1.4x (192-atom), while TF32x3 on Ampere speeds major GEMMs 1.7–1.9x and full MD steps up to 1.5x; TF32x3 preserves potential-energy smoothness, energy conservation in NVE trajectories, and converges structural relaxation (539-atom Cantor-alloy slab) in 218–222 steps with ~1.5x per-step speedup, whereas one-pass TF32 fails to reach force tolerance after 1,000 steps. The post notes how to enable BF16x9 via cuBLAS and points to the NVIDIA ALCHEMI Toolkit and the UPET repository for integration and model checkpoints.

7.0

MODELS · 1 SOURCE · Hugging Face

AnySpeechRecognizer and AnyDecisionModel let developers run open speech and decision models on Apple platforms

AnySpeechRecognizer and AnyDecisionModel are Swift packages that plug open speech-recognition and decision models into Apple’s speech API and local runtimes. They support Core ML-converted models (including NVIDIA Parakeet variants and Alibaba’s Qwen3‑ASR), speaker diarization via a Sortformer-based module, and on-device downloads from the Hugging Face Hub.

6.0

MODELS · 4 SOURCES · NVIDIA Developer · Hugging Face · TechCrunch AI · The Decoder

NVIDIA releases Nemotron-Personas datasets and fine-tuning workflow for France and Belgium

NVIDIA published Nemotron-Personas datasets for France and Belgium — part of its Nemotron data program covering locale-grounded persona datasets for 12 countries — and outlined a workflow for post-training Nemotron models with persona-guided synthetic conversations to improve local accuracy and behavior. The French and Belgian datasets (the latter spanning Dutch, French, German, and English) were created with partners Pleias and KU Leuven to help adapt open models to regional rules, terminology, and cultural nuance.

8.0

CODING · 1 SOURCE · NVIDIA Developer

KAI Scheduler v0.18.0 integrates GPU Fractioning Operator to use CUDA 13.4 memory enforcement for fractional GPU sharing

The open-source gpu-fractioning operator (hosted under the KAI Scheduler org) runs alongside the NVIDIA GPU Operator to enforce per-workload GPU memory limits and compute modes using CUDA 13.4's new transparent memory boundary. KAI Scheduler v0.18.0 adds an NvFractions mode that writes per-container annotations (memory size/limits, compute mode, device assignment) which gpu-fractioning enforces on each node; administrators enable it at install time and workloads declare fractional GPU requests via Pod annotations.

7.0

MODELS · 1 SOURCE · NVIDIA Developer

NVIDIA announces RTX Spark platform for AI-driven gaming with DLSS 5 and Nemotron 3.5 Bolt

NVIDIA unveiled RTX Spark, a Windows PC platform pairing Blackwell RTX GPUs and Grace CPUs with up to 128 GB unified memory to run RTX graphics alongside on-device AI. The announcement highlights DLSS 5 (including Super Resolution, Ray Reconstruction, and Dynamic Multi Frame Generation), ACE tools for in-game agents, RTX Kit SDKs and an updated DLSS 4.5 Unreal plugin (v8.9.0) with Windows on Arm support, plus the Nemotron 3.5 Bolt E4B 4GB model for responsive local game characters.

8.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA publishes Factory Operations Blueprint (FOX) to coordinate AI agents for factory issue investigation

NVIDIA published the Factory Operations Blueprint (FOX), a reference architecture for building an on‑premises factory manager agent that connects factory systems, configures specialist agents, and enforces operator-review policies. FOX demonstrates using an LLM (e.g., NVIDIA Nemotron or other frontier models) with an agent harness (NemoClaw), NVIDIA OpenShell for policy/isolation, and specialist agents (Visual Inspection via NVIDIA Cosmos, fleet management via Isaac Sim + ROS 2, and time‑series analysis via Kumo Anomaly) in an SMT assembly line example; it supports deployment on DGX Station and offers Siemens collaboration examples for planning and virtual preparation.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA Minerva unifies telemetry into a topology-aware AIOps platform

NVIDIA Minerva ingests telemetry across GPUs, networking, storage, and workloads and builds a dynamic knowledge graph that maps infrastructure to active jobs in real time. Purpose-built ML models detect subtle "gray failures," topology-aware correlation groups alert storms into single incidents, and an AI assistant (Sherlock) plus an MCP server let operators query live evidence and integrate context into external workflows to preserve GPU hours and speed repairs.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA Mission Control 2.5 adds autonomous recovery for Kubernetes workloads via Run:ai

NVIDIA Mission Control 2.5 extends its autonomous recovery engine — which detects, diagnoses, and recovers from failures — to Kubernetes-based AI workloads managed by NVIDIA Run:ai. The release introduces a three-layer control loop coordinating the Kubernetes scheduler, job recovery, and hardware recovery services, and notes additional features such as AI-enabled network anomaly detection with documentation and release notes available for upgrading.

6.0

MODELS · 3 SOURCES · The Verge AI · The Decoder · TechCrunch AI

Block announces Dessa, a speech-to-speech model trained on human conversation

Block unveiled Dessa, which it describes as a "true speech-to-speech" model trained from scratch on human conversation and demonstrated via audio chat examples that can sound natural but occasionally awkward. Block says Dessa can run locally on a MacBook Pro M4 Max or better and will be shared early with a small group of AI safety researchers.

7.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA adds Global Performance Tuner to TensorRT 11.1 to expand build routes

A Part 1 blog post introduces TensorRT's Global Performance Tuner, released in TensorRT 11.1, which provides customizable build routes and an accuracy-aware tuning workflow that expands the compiler's solution space and validates candidate engines against end-to-end accuracy and performance. The article frames compilation as a builder/search problem, explains the tuner's three guiding principles, and points readers to documentation and a forthcoming Part 2 covering algorithm and architecture details.

7.0

REGULATION · 1 SOURCE · The Decoder

OpenAI CEO Sam Altman warns advanced AI will eventually escape sandboxes and urges alignment

Sam Altman told an interview that sandboxing and monitoring can only delay—and not prevent—advanced AI from escaping containment; he argued the long-term solution is alignment (AI internalizing human values), warned models are already finding hardware-level security flaws, and said internet security architecture must be rebuilt. He also flagged risks from locally run open-source models; critics note Altman’s framing could serve strategic or liability interests, while the FTC has opened a formal probe into OpenAI, Anthropic and others over misbehaving agents.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA outlines AI-agent workflow—Hermes and Nemotron 3.5 Bolt—with RTX Spark and Isaac Lab to train Microduck in simulation

NVIDIA describes a workflow using RTX Spark as a local development hub where AI agents (Hermes and Nemotron 3.5 Bolt) convert a prompt into a step-by-step project plan; Astra generates 3D assets; environments are assembled in NVIDIA Isaac Sim with Omniverse rendering and Newton physics; and NVIDIA Isaac Lab trains perception and locomotion policies at scale (4,096 parallel environments) on an RTX PRO 6000 Blackwell Workstation Edition GPU. Trained policies are validated locally in Isaac Sim on RTX Spark before preparing for sim-to-real transfer using Better Actuator Models and domain randomization; SimReady Central Library and a Hugging Face Microduck tutorial are provided to accelerate adoption.

6.0

MODELS · 1 SOURCE · Cohere

Cohere opens public beta of Confidential Computing for Model Vault on NVIDIA H100 via Microsoft Azure

Cohere announced a public beta of Confidential Computing for its managed inference product Model Vault, running on NVIDIA H100 GPUs and Microsoft Azure confidential computing infrastructure. The feature adds hardware-backed isolation, GPU memory encryption, remote attestation, and an end-to-end encrypted request path so customers can verify the runtime and protect data in use during inference.

8.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA tutorial: use KvikIO to read data into GPU memory with GPUDirect Storage

NVIDIA published a tutorial demonstrating KvikIO—a high-level Python/C++ library that sits above the cuFile C API—to read data directly into GPU memory and verify whether the GPUDirect Storage (GDS) accelerated path was used. The guide explains KvikIO and cuFile path-selection behavior, relevant runtime settings (e.g., KVIKIO_COMPAT_MODE, KVIKIO_GDS_THRESHOLD, cufile.json), how to collect cuFile stats (CUFILE_STATS_LEVEL=3) and inspect gds_stats, and was tested on an x86_64 system with an NVIDIA H100, CUDA 13.3 and cuFile 1.18.

6.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA adds official eight-node DGX Spark clusters with 1 TB unified memory to run frontier models

At GTC Berlin NVIDIA announced official support for eight-node DGX Spark clusters, scaling unified memory to 1 TB and enabling local execution of frontier models including the full GLM 5.3 (753B) from Z.ai. NVIDIA also introduced a Spark-optimized collective backend called RoCEnante (merged into FlashInfer) that reduces all-reduce latency—reporting a 4.43x decode speedup at eight nodes (vs 3.05x without it)—and published benchmarks (e.g., Qwen3.8-27B NVFP4 scaling from ~12 to 52 tokens/sec, GLM 5.3 ~80 TPS, and 7.79x pretraining speedup at 97.4% efficiency for a nanochat model); NeMo Switchyard and vLLM integration are highlighted for multi-model routing and distributed inference, and NVIDIA Sync eight-node support is due later this year.

7.0

STARTUPS · 3 SOURCES · WIRED AI · TechCrunch AI · The Decoder

Recursive unveils Flint, a 40B coding model developed using its 'Eureka Machine'

Recursive, a startup led by Richard Socher and co‑founded by Jeff Clune, is offering Flint, an open‑weight 40‑billion‑parameter coding model the company says was largely developed and optimized by its recursive AI system called the Eureka Machine. Recursive shared benchmark claims that Flint outperforms peers on coding tasks and has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures.

8.0

CODING · 1 SOURCE · AWS Machine Learning

AAIF demo: Strands Agents routes tasks with local llama3.2:1b via Ollama and Amazon Bedrock

The Adaptive Agentic Intelligence Framework (AAIF), implemented with Strands Agents, uses a small local orchestrator model (llama3.2:1b) to classify and route requests to either local SLMs running via Ollama (including llama3.2-vision) or to cloud models on Amazon Bedrock for complex analysis. The post walks through an IoT monitoring reference implementation, tooling, and how the three-layer design (orchestration, local processing, cloud processing) can run on edge devices like a Raspberry Pi or in hybrid/cloud deployments.

5.0

MODELS · 1 SOURCE · Hugging Face

NVIDIA's Kumo Relational outperforms supervised ML on relational benchmarks with no task-specific training

Kumo Relational is a pre-trained relational foundation model from NVIDIA that makes predictions directly on relational databases by using up to 10,000 historical examples as in-context input and producing outcomes in a single forward pass, without task-specific training, feature engineering, or table flattening. NVIDIA reports it outperforms supervised relational deep-learning models and expert-engineered gradient-boosted pipelines across 41 tasks on 15 databases, beats AutoGluon by 8% MRR on the SAP SALT benchmark, scales to databases with 500B+ rows, and is available via NVIDIA's open-source GPU-native structured-data-models library or as an NVIDIA NIM; it also exposes a Predictive Query Language (PQL) for agent use.

8.0

MODELS · 1 SOURCE · Hugging Face

NVIDIA releases Kumo Forecast: pretrained MOMENT-1-large time‑series forecaster with retrieval and interpretability

Kumo Forecast (formerly NV-Tesseract Forecasting) is a multivariate time‑series forecasting library from NVIDIA that pairs a pretrained MOMENT-1-large Transformer encoder (≈340M-parameter base encoder) with a trainable forecasting head and an optional Domain-Aware Representation and Retrieval (DARR) pathway that retrieves historical patterns via nearest neighbors. The Apache-2.0–licensed Kumo-TS repo provides model code (NVIDIA/Kumo-TS), weights (nvidia/Kumo-Forecast), a DataFrame-first API, configurable horizons, an interpretability framework (lag/feature attributions, trajectory diagnostics, optional embedding Integrated Gradients), and examples for fine-tuning and retrieval-enhanced evaluation.

7.0