Night. The laptop goes into a bag and out the door. On the desk, a small flat box stays behind, able to keep running agents and scheduled tasks. NVIDIA writes that a dedicated desktop can keep them going while a laptop travels or handles other work 1.
The official RTX Spark account announced that box this way: "RTX Spark, now in compact desktop form" 2. According to the post, it is a dedicated Windows PC for local models, always-on agents and creative work, with up to 128GB of unified memory 3.
RTX Spark, now in compact desktop form. ⚡
— NVIDIA RTX Spark (@NVIDIARTXSpark) November 18, 2026
A dedicated @Windows PC for local models, always-on agents and creative work, with up to 128GB of unified memory.
Available now from @ASUS, @Dell and @msiUSA, with more from @Acer, @GIGABYTEUSA, @HP and @Surface on the way. pic.twitter.com/EJd66c6gQW
What fits inside the small box
According to NVIDIA, the compact desktops carry a 20-core Grace CPU, a Blackwell RTX GPU with 6,144 CUDA cores and up to 128GB of unified memory, delivering up to 1 petaflop of AI performance 4.
NVIDIA's examples of what that enables include running massive large language models with 120B parameters locally 5 and rendering 90GB+ 3D scenes in real time 6. Those are NVIDIA's own claims, but it is hard not to picture the desk they imagine.
NVIDIA also points to a quieter benefit. Running supported models locally keeps locally processed code and files on the PC and avoids per-token cloud API charges for those tasks 7.
ASUS ProArt GR1X
ASUS's product page describes a 1.1L chassis measuring 150 x 150 x 51 mm 8. Memory goes up to 128GB of LPDDR5X-9400 with up to 300GB/s of bandwidth 9, and storage up to 4TB 10. The cooling is designed to sustain up to 140W TDP, ASUS says 11.
ASUS's US store lists it starting at $4,499, with a "Notify me" button rather than a buy button for now 12.
MSI EdgeMesa N AI+
The other small box here is MSI's EdgeMesa N AI+. According to MSI, the chassis is 1.3L 13, memory goes up to 128GB 14, storage up to 4TB 15, and the TDP is 140W 16. On the back are 4 USB Type-C ports, 3 of which support video output 17.
NVIDIA lists MSI among the makers whose compact desktops are available 18.
The GR1X is 1.1L 8 and the EdgeMesa 1.3L 13. Either one would fit in a corner of a bookshelf.
Dell, and the boxes still to come
NVIDIA also lists Dell among the makers with compact desktops available 18. When it comes to Dell's small machines, there is the Dell Creator Edition Desktop, described as a compact, always-on Windows device for creators running demanding AI workloads 19. Memory goes up to 128GB 20, and Dell says it supports models up to 200B parameters 21.
Mini PCs from Acer, GIGABYTE and HP, plus the Surface RTX Spark Dev Box, are on the way 18. For how a GIGABYTE box fits with local coding AI, see a local Copilot drafting a repository cleanup plan.
Turning a sketch into animation on the desk
What should the box do? NVIDIA's showcase links Photoshop and ComfyUI through Hermes Agent 22. With natural-language instructions, creators can turn sketches into rendered images and animations, or animate an image entirely on device, from isolating a subject and defining its motion to generating the finished video, NVIDIA writes 23.
At Adobe MAX, attendees used natural-language instructions to turn sketches into rendered images and animations for a shared digital Miami landscape 24. Inside ComfyUI, a generated clip can be expanded to 60 frames per second and converted to HDR for color grading, all in the same workflow on RTX Spark 25.
NVIDIA Switchyard, which automatically routes agent tasks between local and cloud models, is now available in Hermes 26. NVIDIA also notes that Substance 3D Painter 12.2 is optimized for RTX Spark in public beta 27, that the native Illustrator delivers up to 30% higher performance than its emulated version 28, and that Blender 5.3 integrates DLSS 4.5 Ray Reconstruction into the Cycles viewport 29.
Another box on the desk: DGX Spark units in a ring
This is a different line from the RTX Spark Windows machines. The DGX Spark HPC Developer Kit that NVIDIA announced at SC26 is a cluster of 4 DGX Spark systems, each connected directly to its neighbors in a ring 30, with no network switch to buy or set up 31. NVIDIA describes the kit as putting the GB10 Grace Blackwell Superchip on researchers’ desks 32.
Together they provide 512GB of combined memory and 16TB of NVMe storage 33. With NVFP4, NVIDIA says, researchers can deploy models with up to 700B parameters locally across the cluster 34. NVIDIA says the kit is on display at SC26 and available for sale 35. NVIDIA pitches it as a kit that turns a desk into a supercomputing lab.
If you are choosing, start with memory
Decide first whether the box sits next to your laptop or replaces it. That choice narrows the field quickly.
On memory, ASUS also offers a 64GB configuration 36, with up to 128GB at the top end 37. If big models are the goal, check how much memory a machine can carry before anything else. Before buying, confirm in the official documentation that the tools you want to run are supported.
A box that runs all the time tends to want its model files and assets on network storage. NVIDIA writes that 10GbE on the compact desktops enables fast transfers of large datasets, model files and creative assets between the desktop and networked storage 38.
Ad: the “Search on Amazon” buttons are affiliate links, and we may earn a commission. They open a search, not a specific listing.
Read next
Torn on memory? See a deeper look at choosing memory. For handing work to a local agent, see how much to let a local AI agent touch.
