https://vertex.rspruijt.com/ Updates to Vertex: a searchable, independently maintained reference covering NVIDIA's hardware and software product catalog. en-us Mon, 05 Oct 2026 11:37:27 +0000 https://vertex.rspruijt.com/#/versioning/1.8 vertex-v1.8 Mon, 05 Oct 2026 00:00:00 +0000 Added 6 products announced by NVIDIA in the past two weeks: the NVIDIA Open Agent Safety Platform and NVIDIA OpenShell (Core AI Software Platform); NodeWright (Virtualization & Infrastructure Management); and NVIDIA TensorRT Model Connect, NVIDIA NeMo Relay, and NVIDIA cuObject + NVIDIA SCADA Server SDK (Developer Tools).<br/>• New row: NVIDIA Open Agent Safety Platform (Tier 1, Limited Availability) — NVIDIA's flagship AI-agent governance initiative, pairing the OpenShell software runtime with NVIDIA Sentry, a BlueField-4 hardware watchdog. Status reflects that Sentry has not shipped: it remains a reference design with no public release date, while OpenShell is already GA.<br/>• New row: NVIDIA OpenShell (Tier 2, GA, open source) — kernel-level sandboxing runtime for autonomous AI agents, confirmed in production use at Cadence, Dassault Systèmes, Siemens, Synopsys, CrowdStrike, Palantir, SAP, ServiceNow, Microsoft, Canonical, Red Hat, and Gecko Robotics.<br/>• New row: NodeWright (Tier 2, GA, open source) — Kubernetes-native package manager for safely updating GPU-cluster host-OS settings without disrupting running workloads; open-sourced under Apache 2.0 after years of internal NVIDIA use under the codename "Skyhook."<br/>• New row: NVIDIA TensorRT Model Connect (Tier 2, public preview) — takes a Hugging Face model ID to a running native C++ TensorRT inference bundle in two commands, no PyTorch/Python dependency at runtime.<br/>• New row: NVIDIA NeMo Relay (Tier 3, GA, open source) — agent-observability framework capturing execution traces in ATOF, ATIF, and native OpenTelemetry formats.<br/>• New row: NVIDIA cuObject + NVIDIA SCADA Server SDK (Tier 2) — RDMA-accelerated object/file storage access. Status distinguishes the two halves: cuObject itself has been GA since January 2026 (now v2.0.0), while the newly introduced SCADA Server SDK is pre-GA, with IBM's Storage Scale as its first known prototype implementation.<br/>• Fact-check correction applied before publishing: an earlier internal draft of these six entries dated the Open Agent Safety Platform and OpenShell to Sept 28 (platform announcement date, correct) but implied OpenShell itself also shipped that day (it shipped Sept 25) and listed Slack as an OpenShell user, which could not be verified against NVIDIA's own announcement or press coverage and was dropped. TensorRT Model Connect's date was corrected from a claimed Sept 29 to its actual Aug 18 public-preview date. The cuObject + SCADA Server SDK entry was rewritten to stop implying cuObject itself was newly GA, when only the SCADA SDK is new. https://vertex.rspruijt.com/#/versioning/1.7 vertex-v1.7 Tue, 15 Sep 2026 00:00:00 +0000 Completed the software-category Technical Deep-Dive expansion promised after the hardware-spec pass (Version 1.1): 59 existing products across Networking Software, Core AI Software, Robotics & Physical AI, Omniverse, Domain-Specific Software, Cloud Services, and Developer Tools now carry the same concrete-spec depth as the hardware categories.<br/>• Expanded Technical Deep-Dive entries with concrete architecture details, version numbers, and spec-sheet/GitHub links for 59 products across Categories 6, 9, 10, 11, 12, 13, 14, and 15.<br/>• Notable corrections surfaced during expansion: Omniverse's mid-2026 licensing change (NVIDIA AI Enterprise subscription no longer required for dev/production/redistribution); Merlin's public development activity confirmed effectively in maintenance mode since mid-2024; DGX Cloud Lepton's full current partner-cloud roster confirmed directly from NVIDIA's own announcements.<br/>• Two products (NVIDIA Network Operator, NVIDIA Agent Toolkit) were left unchanged — already at full expanded depth from prior passes. https://vertex.rspruijt.com/#/versioning/1.6 vertex-v1.6 Tue, 15 Sep 2026 00:00:00 +0000 Added 13 products flagged as missing from the catalog: ConnectX-7, ConnectX-8, and BlueField-3 (Networking Hardware); NVIDIA Aerial/AI-RAN and Cumulus NetQ (Networking Software); DRIVE OS (Automotive); NVIDIA AI Data Platform and NVIDIA AI Blueprints (Core AI Software); Omniverse Kit (Omniverse); NVIDIA LaunchPad (Cloud Services); and DLSS, NVIDIA Reflex, and RTX Remix as their own rows in Gaming & Consumer.<br/>• New rows: ConnectX-7 SuperNIC and ConnectX-8 SuperNIC (Tier 1, GA) — the SuperNIC generations preceding ConnectX-9, filling a gap where only the newest, still-sampling generation was previously listed.<br/>• New row: BlueField-3 DPU (Tier 2, GA) — the current mainstream BlueField generation, previously only its successor (BlueField-4, still sampling) was listed.<br/>• New row: NVIDIA Aerial / AI-RAN (Tier 2, Limited Availability) — NVIDIA's AI-native 5G/6G radio access network platform.<br/>• New row: Cumulus NetQ (Tier 3, GA) — the telemetry/operations layer for Cumulus Linux and NVOS fabrics, distinct from Cumulus Linux itself.<br/>• New row: DRIVE OS (Tier 2, GA) — the safety-certified automotive operating system underlying DRIVE AGX Thor and Orin.<br/>• New row: NVIDIA AI Data Platform (Tier 2, Limited Availability/partner-dependent) — a reference architecture embedding GPUs, DPUs and AI software into partner storage systems.<br/>• New row: NVIDIA AI Blueprints (Tier 2, GA) — NVIDIA's curated catalog of pre-built reference AI workflows built on NIM.<br/>• New row: Omniverse Kit (Tier 2, GA) — the SDK/application-development framework underlying every first-party Omniverse app.<br/>• New row: NVIDIA LaunchPad (Tier 2, GA) — NVIDIA's free, time-limited hands-on infrastructure evaluation program.<br/>• New rows: DLSS, NVIDIA Reflex, and RTX Remix broken out as their own entries in Gaming & Consumer, each with full version/generation history.<br/>• Judgment calls applied: Isaac GR00T N1's Notes now flag its non-commercial research license despite NVIDIA's "open" framing; NVIDIA Edify's Status updated to reflect its June 2025 standalone-NIM deprecation and partner-embedded-only distribution. https://vertex.rspruijt.com/#/versioning/1.5 vertex-v1.5 Tue, 15 Sep 2026 00:00:00 +0000 Added NVIDIA NeMo Switchyard to Category 15 (Developer Tools): an open-source framework that routes AI agent workloads across multiple models to balance cost, latency, and quality.<br/>• New row: NeMo Switchyard (Tier 1, GA/open source) — routes each agent task to the model best suited for it instead of defaulting every request to the most capable/expensive one, via a provider-agnostic SDK (switchyard-libsy) with both tuning-free and tunable routers.<br/>• Tier 1: cost-optimized model routing is a near-universal talking point in current agentic-AI/enterprise conversations. Integrates with NIM, Dynamo, and NeMo Framework, plus LangChain and Kong partnerships. https://vertex.rspruijt.com/#/versioning/1.4 vertex-v1.4 Tue, 15 Sep 2026 00:00:00 +0000 Added NVIDIA SkillEvaluator to Category 15 (Developer Tools): an open-source framework for testing whether AI agent skills actually improve agent performance.<br/>• New row: SkillEvaluator (Tier 2, GA/open source) — a three-tier evaluation framework (static checks, embedding-similarity duplicate detection, live sandboxed "Skill Lift" measurement via the Harbor framework) for validating AI agent skills before they ship.<br/>• Cross-vendor tool: integrates with Claude Code, OpenAI Codex, and Cursor, not just NVIDIA's own stack — flagged as very new, with adoption evidence limited to NVIDIA's own published benchmarks so far. https://vertex.rspruijt.com/#/versioning/1.3 vertex-v1.3 Tue, 15 Sep 2026 00:00:00 +0000 Rewrote the About page's "Why this exists" and "Who it's for" sections and added "How to use it" and "Value" sections, so the training and customer/partner-meeting use case is stated directly instead of implied.<br/>• "Why this exists" now names the second problem Vertex solves directly: keeping a team's shared understanding from drifting out of date as NVIDIA ships new products most months, and giving training sessions and customer/partner meetings a common reference.<br/>• Added a "How to use it" section (scanning a category, filtering by tier, explaining an entry to someone else, checking maturity) — previously only in the source document, not on the website.<br/>• Added a "Value" section summarizing what Vertex delivers over NVIDIA's own scattered product pages.<br/>• Matching wording updates made to the source document's "Why this doc exists", "How to use it", and "Value and benefits" sections. https://vertex.rspruijt.com/#/versioning/1.2 vertex-v1.2 Fri, 11 Sep 2026 00:00:00 +0000 Reorganized the About page and added a feedback contact.<br/>• "About the name: Vertex" moved up to the top of the About page, right after the page intro.<br/>• Added a Feedback section to the About page and to the source document, with a contact address (vertex@rspruijt.com) for corrections and suggestions. https://vertex.rspruijt.com/#/versioning/1.1 vertex-v1.1 Fri, 11 Sep 2026 00:00:00 +0000 Technical Deep-Dive entries expanded with concrete hardware specs and spec-sheet links across the 8 hardware categories.<br/>• Added concrete specs (memory capacity/bandwidth, compute throughput, core/transistor counts, power envelopes, interconnect bandwidth) to the Technical Deep-Dive column for 37 products across Data Center GPUs & CPUs, Data Center Systems, Professional/Workstation GPUs, Gaming & Consumer, Networking Hardware, Edge & Embedded Hardware, Automotive & Robotics Hardware, and Inference Accelerators (non-GPU).<br/>• Each expanded entry links to an official NVIDIA (or, for Groq, manufacturer) spec sheet or product page.<br/>• Announced-but-not-yet-shipping products (Rubin GPU, Vera CPU, Vera Rubin Superchip, Vera Rubin NVL72) are explicitly flagged as pre-final specs.<br/>• Software-category Technical Deep-Dive entries (categories not covered above) are unchanged in this pass — a follow-up covering those is planned. https://vertex.rspruijt.com/#/versioning/1.0 vertex-v1.0 Fri, 11 Sep 2026 00:00:00 +0000 Initial public release of Vertex.<br/>• 117 products cataloged across all 16 categories: data-center silicon, systems and reference designs, professional and consumer GPUs, networking hardware and software, edge and embedded hardware, automotive and robotics, core AI software, physical AI and simulation, domain-specific and vertical software, cloud and SaaS services, virtualization and infrastructure management, developer tools, and non-GPU inference accelerators.<br/>• Every entry includes a plain-English explanation, a technical deep-dive, a maturity status, a release date, named competitors, and contextual notes.<br/>• Full-text search, category/tier/status filtering, and sorting across the catalog.<br/>• Sources tab listing every citation used to fact-check the catalog.<br/>• RSS feed for tracking future updates.