NVIDIA RTX Spark Compact Desktops Out Now
NVIDIA RTX Spark compact desktops from ASUS, Dell, and MSI are now available, offering 128GB unified memory and up to one petaflop of local AI computing power.
By Dillip Chowdary • Nov 19, 2026 • Source: NVIDIA Blog

At a glance
- Product
- NVIDIA RTX Spark compact desktops
- Processor
- 20-core NVIDIA Grace CPU
- Graphics
- Blackwell RTX GPU with 6,144 CUDA cores
- Unified memory
- Up to 128GB
- AI performance
- Up to 1 petaflop
- Availability
- ASUS, Dell, and MSI desktops available now
In this post
NVIDIA launched RTX Spark compact desktop systems built by hardware partners ASUS, Dell, and MSI for local artificial intelligence computing. According to NVIDIA Blog's report, these dedicated Windows personal computers combine high-capacity unified memory with graphics hardware to run generative models without per-token cloud charges. The machines keep persistent background agents and scheduled tasks operational locally while laptops travel or handle separate mobile assignments.
This reporting examines system hardware configurations, partner release timing, creative software optimizations, and developer tools across the RTX Spark desktop platform. The updates target software engineers, digital creators, video editors, and technical enthusiasts executing local multi-agent systems or rendering complex media. It details operational memory limits, platform driver requirements, native Arm application support from Adobe, and newly supported open-weight models.
01Grace CPU and Blackwell RTX silicon
The compact desktops feature a 20-core NVIDIA Grace central processor combined with an NVIDIA Blackwell RTX graphics processing unit containing 6,144 CUDA cores. Configured with up to 128GB of unified memory, each system delivers up to 1 petaflop of artificial intelligence computing performance. Onboard 10-gigabit Ethernet enables rapid transfers of model checkpoints, heavy training datasets, and creative project assets between desktop units and networked storage pools.
That hardware configuration allows technical teams to execute massive 120-billion-parameter language models directly within memory while avoiding per-token cloud application programming interface charges. Creative studios can simultaneously preview 90GB and larger 3D scenes in real time or edit uncompressed 4:2:2 video streams. Running these memory-heavy processes on a standalone local machine keeps proprietary application code, customer records, and media files securely on premises.
Note: Hardware options from ASUS, Dell, and MSI are available for purchase immediately, joining the existing family of RTX Spark laptops. Additional mini PCs from Acer, GIGABYTE, and HP alongside the Microsoft Surface RTX Spark Dev Box are scheduled to arrive in subsequent hardware releases.
02Agent orchestration and hybrid model routing
Software developers build local intelligence pipelines using native CUDA acceleration alongside Linux development tools through Windows Subsystem for Linux. Engineers can deploy autonomous agents with Hermes Agent or NVIDIA NemoClaw blueprints, while model fine-tuning runs on supported open architectures using Unsloth. These software foundations let personal digital agents maintain continuous background operations without monopolizing a primary laptop or travel workstation.
NVIDIA Switchyard is integrated into Hermes to automatically route computational tasks between local silicon and remote cloud endpoints. Everyday tasks execute locally on the desktop graphics processing unit, while advanced reasoning queries draw on cloud models to balance subscription expenses. Creators can also link Adobe Photoshop and ComfyUI through Hermes Agent to turn rough sketches into rendered imagery and video animations using natural-language instructions.
At the Adobe MAX event, attendees used these agent workflows to generate rendered images and animations for a shared digital Miami landscape. The media pipeline is further extended by RTX Video Frame Generation and RTX Video HDR directly within ComfyUI. These tools allow creators to expand generated video clips to 60 frames per second and apply high dynamic range color profiles inside one local session.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
03Native Arm acceleration across Adobe software
Adobe Premiere integrates NVIDIA TensorRT-RTX acceleration for automated video editing tools including Enhance Speech, Object Mask, Scene Edit Detection, and Auto Reframe. Acceleration also supports Film Impact transition plug-ins and professional video formats including Canon RAW, Red RAW, ARRI RAW, ARRICORE, ARRI HDE, ProRes RAW, long-GOP, and MXF containers. Video editors processing standard H.264 deliverables achieve more than 50 percent faster project turnarounds.
Adobe Photoshop adds TensorRT-RTX acceleration for its Remove Tool alongside native Camera Raw support that operates without binary emulation on Windows on Arm. At Adobe MAX, Adobe previewed a next-generation image processing engine featuring graphics processor acceleration for interactive Gaussian blur and Light Leaks filters. Photographers can retouch high-resolution raw imagery and remove unwanted subjects quickly without relying on software translation layers.
Adobe Substance 3D Painter 12.2 introduces its first native public beta for Windows on Arm, letting visual effects artists texture 3D assets natively. Adobe Illustrator and InDesign now provide native Arm versions optimized for RTX Spark, with Illustrator delivering up to 30 percent higher performance than emulated editions. These native compilations ensure designers avoid emulation performance penalties during intensive vector illustration, texturing, and multi-page publishing workflows.
04Real-time Cycles viewport Ray Reconstruction
Blender 5.3 integrates NVIDIA DLSS 4.5 Ray Reconstruction directly into the Cycles viewport rendering engine. Instead of navigating noisy previews that take several seconds to resolve, artists can evaluate lighting, reflections, and materials in real time with near-final ray-traced visual fidelity. This interactive feedback helps animators and 3D specialists adjust scene lighting and shaders without submitting time-consuming test renders.
Note: Enabling DLSS 4.5 Ray Reconstruction requires Blender 5.3 alongside the latest NVIDIA Game Ready Driver or NVIDIA Studio Driver. Creators can use the NVIDIA App to receive automated driver notifications, manage GPU settings, and discover new developer tools.
05Enterprise server platforms and open weights
At Microsoft Ignite, the companies announced that the NVIDIA DGX Station for Windows powered by the GB300 Grace Blackwell Ultra Desktop Superchip is adding support for Windows Server 2028. This evolution targets organizations running artificial intelligence models near on-premises enterprise software and corporate datastores. Microsoft is expanding hardware support across Windows silicon to provide scalable platforms for local inference, multi-agent development, and intensive training workloads.
Windows Machine Learning now allows developers to run GGUF model files from Hugging Face using an integrated llama.cpp backend. An experimental DxCGC backend allows game developers to compile and execute machine learning models through DirectX 12 alongside standard graphics rendering. Application programming interfaces in Windows add GPU acceleration for AION Instruct and VSR, while developers can compile high-performance models using the command-line interface.
Open-source tooling also advanced with llama.cpp v1.0.0, introducing API versioning, expanded model compatibility, and CUDA optimizations for NVFP4 inference. Ricursive Intelligence released Ripple-65B Preview, a 65-billion-parameter model trained on B300 GPUs and refined with Nemotron datasets for local execution on DGX Spark. Runway introduced Praxis-1, a 27-billion-parameter world action model running locally on GeForce RTX 5090, RTX PRO 6000 Blackwell, DGX Spark, and DGX Station.
Frequently asked questions
Which manufacturers offer NVIDIA RTX Spark compact desktops today?
Desktops from ASUS, Dell, and MSI are available for purchase immediately. Additional mini PCs from Acer, GIGABYTE, and HP alongside the Microsoft Surface RTX Spark Dev Box are coming soon.
What are the hardware specifications of RTX Spark compact desktops?
Each desktop includes a 20-core NVIDIA Grace CPU, an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores, and up to 128GB of unified memory. The systems deliver up to 1 petaflop of AI performance and include 10-gigabit Ethernet connectivity.
Which Adobe creative applications run natively on RTX Spark?
Adobe Illustrator, InDesign, and Substance 3D Painter 12.2 run natively on Windows on Arm for RTX Spark. Adobe Premiere and Photoshop add TensorRT-RTX acceleration for AI editing features and raw image processing.
What software is required to use DLSS Ray Reconstruction in Blender?
Enabling DLSS 4.5 Ray Reconstruction requires Blender 5.3 alongside the latest NVIDIA Game Ready Driver or NVIDIA Studio Driver.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement