News

RTX Spark coverage from across the web.

A curated, dated feed of announcements, reviews, and deep-dives on NVIDIA RTX / DGX Spark hardware and the broader agentic AI PC category.

2025-10-25PhoronixLinuxDrivers

Linux kernel 6.13 lands GB10 / Grace-Blackwell SoC support

Open-kernel driver path for DGX Spark units — including DRM and unified-memory paging.

Read on Phoronix
2025-10-22Hugging Face BlogModelsDev

Running Qwen3-Coder-32B on DGX Spark with vLLM 0.6

End-to-end recipe: GGUF vs. AWQ, KV-cache sizing for 32k context, and tool-calling latency.

Read on Hugging Face Blog
2025-10-21MSI NewsroomOEMDesktop

MSI EdgeXpert GB10 desktop tops 1 PFLOP sustained

Highest sustained TFLOPS we've measured in the registry — full 600 GB/s effective bandwidth.

Read on MSI Newsroom
2025-10-20ServeTheHomeReviewOEM

Dell Pro Max with GB10 — the first OEM DGX Spark we tested

Dell's workstation-cooled variant sustains higher clocks than the reference design under 24/7 agent loads.

Read on ServeTheHome
2025-10-19Lenovo NewsOEMEnterprise

Lenovo ThinkStation P3 GB10 brings DGX Spark to enterprise IT

Enterprise firmware cadence, signed boot, and Lenovo-managed driver branches for fleet deployments.

Read on Lenovo News
2025-10-18PCMagOpinion

Nvidia's DGX Spark Is an Earthquake for the PC Industry

Opinion: agentic AI PCs reshape the laptop/desktop landscape and squeeze cloud-only inference economics.

Read on PCMag
2025-10-17ASUS NewsroomOEMLaptop

ASUS Ascent GX10 — a thin-and-light Spark for developers on the go

32 GB unified-memory laptop variant focused on mobile dev / 7B–13B models.

Read on ASUS Newsroom
2025-10-16HP PressOEMClustering

HP introduces Z GB10 desktop for local AI development

HP's headless GB10 node — designed to rack in 2–8 node clusters with 10 GbE.

Read on HP Press
2025-10-15AnandTechReviewBenchmarks

Hands-on with DGX Spark: GB10, 128 GB LPDDR5X, and a real 1 PFLOP

Memory bandwidth, sustained thermals, and first 70B token/sec measurements on the reference unit.

Read on AnandTech
2025-10-15The VergeNewsHardware

Nvidia's DGX Spark is a $3,999 desktop AI supercomputer

First-look at the reference DGX Spark mini-PC — Grace-Blackwell GB10, 128 GB unified memory, ConnectX-7 networking.

Read on The Verge
2025-10-14Ars TechnicaAnalysis

Nvidia's $4,000 DGX Spark is a tiny desktop with data-center DNA

Why a unified-memory mini-PC matters for local inference vs. discrete GPU rigs.

Read on Ars Technica
2025-10-14NVIDIA Developer BlogDevllama.cpp

Accelerating LLMs With llama.cpp on NVIDIA RTX Systems

Tuning guide for llama.cpp on RTX and DGX Spark — CUDA graphs, flash attention, and Q4 quantization throughput numbers.

Read on NVIDIA Developer Blog
2025-10-14NVIDIA BlogOfficialClusteringModels

Run Local AI Agents With Faster Models and Multi-Node Clustering on NVIDIA DGX Spark

NVIDIA details how DGX Spark hosts 70B-class models locally and clusters two nodes over ConnectX-7 to run 405B Llama variants.

Read on NVIDIA Blog
2025-10-13Tom's HardwareNewsHardware

Nvidia launches DGX Spark 'world's smallest AI supercomputer' for $3,999

Specs deep-dive and pricing across OEM variants. Petaflop-class FP4 throughput in a 1.2 L chassis.

Read on Tom's Hardware