Skip to content

VersusGPU

RTX Spark vs DGX Spark: Which NVIDIA Local AI Mini PC Should You Choose?

RTX Spark desktops and DGX Spark both pair a Grace CPU with a Blackwell GPU, up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance. RTX Spark machines are Windows 11 on Arm PCs with 10Gb Ethernet; DGX Spark runs Ubuntu-based DGX OS and can cluster up to four units over ConnectX-7. Neither is a sensible gaming buy.

By , Editorial team

4 min read

RTX Spark vs DGX Spark: Which NVIDIA Local AI Mini PC Should You Choose?

NVIDIA now sells two kinds of compact local AI computer with a Grace CPU, a Blackwell GPU and up to 128GB of memory shared between them. DGX Spark is NVIDIA's AI developer box running Linux. RTX Spark compact desktops, which went on sale from ASUS, Dell and MSI on 18 November 2026, are Windows 11 PCs. On paper the headline numbers look alike, so the choice is mostly about software, networking and what else you want the machine to do. This comparison is based on NVIDIA's product pages and launch coverage; Forge PC hasn't tested either system.

The short version

  • Choose an RTX Spark desktop if you want a Windows PC that runs local AI alongside creative apps such as Adobe's and Blender, with WSL for Linux tools.

  • Choose DGX Spark if you work in Linux and NVIDIA's AI developer stack, or you want to link several units together for bigger models.

  • Choose neither if your main use is gaming. A conventional desktop with a graphics card is far better value for that.

Specifications compared

RTX Spark compact desktops

DGX Spark

Chip

RTX Spark: 20-core NVIDIA Grace CPU and Blackwell RTX GPU with 6,144 CUDA cores

NVIDIA GB10 Grace Blackwell Superchip

Memory

Up to 128GB unified memory

64GB or 128GB unified memory

AI performance (NVIDIA's figure)

Up to 1 petaflop

Up to 1 petaFLOP at FP4

Operating system

Windows 11 (on Arm)

DGX OS, an Ubuntu-based Linux distribution

Networking

10Gb Ethernet

NVIDIA ConnectX-7 with two QSFP ports, up to 200Gb/s each

Clustering

Not a headline feature

Up to four units

Makers

ASUS, Dell and MSI now; Acer, GIGABYTE, HP and a Microsoft Surface Dev Box to follow

NVIDIA, plus GB10 systems from partner OEMs

The 1 petaflop figures are peak FP4 numbers, a low-precision format. Real-world speed depends on the model, its precision and the software. Our FP4 vs FP8 vs FP16 explainer explains why these headline figures aren't directly comparable with gaming GPU specs.

Software is the real difference

RTX Spark runs Windows 11 on Arm. NVIDIA says it supports native CUDA and Linux development tools through Windows Subsystem for Linux, and it's pushing native Arm versions of creative software. NVIDIA says Adobe is expanding native Windows on Arm support across its apps, with Adobe Illustrator up to 30% faster than its emulated version on RTX Spark by NVIDIA's figures. Apps that aren't yet native to Arm run under emulation. The Verge notes that Windows on Arm game support is improving, but at these prices gaming isn't a realistic reason to buy one.

DGX Spark ships with DGX OS, which NVIDIA describes as a Linux distribution based on Ubuntu that matches the software on its other DGX systems. It comes with or supports a large catalogue of AI tools, frameworks and models from NVIDIA NGC. NVIDIA says you can connect to it from a laptop or desktop to launch tools such as VS Code or Cursor remotely. That makes it a better fit as a headless AI server or a stepping stone to NVIDIA's data-centre systems.

Networking and scaling

DGX Spark has two QSFP ports, each rated at up to 200Gb/s. NVIDIA says up to four units can be linked: two or three can be connected directly with QSFP cables, while four need an external network switch. NVIDIA says a single 128GB DGX Spark can run inference on models of up to 200 billion parameters and fine-tune models of up to 70 billion parameters, and that linked systems handle larger models.

RTX Spark desktops have 10Gb Ethernet, which NVIDIA pitches for fast transfers of datasets and model files to network storage, not for clustering.

Memory and model sizes

Both top out at 128GB shared between CPU and GPU, much more than the 16GB or 32GB on desktop graphics cards. That's why they're interesting for large models. Dell claims its RTX Spark Creator Edition Desktop runs models of up to 200B parameters on-device, matching NVIDIA's DGX Spark claim. DGX Spark is also offered with 64GB, which NVIDIA positions for smaller agent workloads. Our guide to how much RAM you need for local AI explains how memory maps to model sizes.

Price

Neither is cheap. The Verge reports that ASUS's ProArt GR1X RTX Spark desktop starts at $4,499 in the US with 64GB of memory and 1TB of storage. UK prices for the RTX Spark desktops hadn't been announced when we checked. For the UK picture so far, see our RTX Spark desktops explainer and our report on Dell's Creator Edition Desktop.

How they compare with the alternatives

If you only want to run local models, compare both with AMD's Ryzen AI Max machines in our DGX Spark vs Ryzen AI Max+ 395 comparison, and with a conventional tower and graphics card in our best PC for local AI guide. A desktop GPU has far less memory but is much faster for models that fit in its VRAM, and it can play games properly.

Our verdict

RTX Spark and DGX Spark share the same core idea, a Grace CPU and Blackwell GPU with up to 128GB of shared memory, but they're aimed at different people. Pick an RTX Spark desktop if you want a Windows PC for creative work and local AI on one machine, and you're happy with Windows on Arm. Pick DGX Spark if you're an AI developer who works in Linux, wants NVIDIA's DGX software stack, or plans to cluster units for bigger models. If you mainly game, buy a conventional desktop instead.

The bottom line

RTX Spark and DGX Spark share the same core idea, a Grace CPU and Blackwell GPU with up to 128GB of shared memory, but they're aimed at different people. Pick an RTX Spark desktop if you want a Windows PC for creative work and local AI on one machine, and you're happy with Windows on Arm. Pick DGX Spark if you're an AI developer who works in Linux, wants NVIDIA's DGX software stack, or plans to cluster units for bigger models. If you mainly game, buy a conventional desktop instead.

Sources

  1. 01NVIDIA, NVIDIA RTX Spark Compact Desktops Out Now
  2. 02NVIDIA, DGX Spark
  3. 03The Verge, The first Nvidia RTX Spark mini PCs launch today from Asus, Dell, and MSI
  4. 04Dell, Dell Creator Edition Desktop: Advanced Local AI on the Windows You Already Know

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

Single-PC vs Dual-PC Streaming: Do You Still Need a Second PC to Stream?

Versus

Single-PC vs Dual-PC Streaming: Do You Still Need a Second PC to Stream?

A single PC is enough for most streamers: NVIDIA says the dedicated NVENC encoder on GeForce RTX cards lets you stream without impacting GPU or CPU performance. A dual-PC setup with a capture card and HDMI passthrough keeps the stream separate from the gaming PC and suits heavy productions, but costs more and is more complicated.

Forge PC Staff3 min

GPU Fans Not Spinning? How to Tell If It's Normal and What to Do If It Isn't

How-To

GPU Fans Not Spinning? How to Tell If It's Normal and What to Do If It Isn't

GPU fans that don't spin at idle are usually normal: MSI says its Zero Frozr cards keep fans still below 60°C, and ASUS says its new Dual cards stop their fans below 50°C. Check the GPU temperature and fan speed, then run a game. If the card gets hot and the fans stay at 0rpm, check power, fan settings and drivers.

Forge PC Staff3 min

Intel Arc Pro B60 vs RTX 5060 Ti 16GB for Local AI: 24GB vs CUDA

Versus

Intel Arc Pro B60 vs RTX 5060 Ti 16GB for Local AI: 24GB vs CUDA

Intel's Arc Pro B60 has 24GB of GDDR6 on a 192-bit bus with up to 456GB/s, aimed at AI inference workstations and multi-GPU setups. NVIDIA's RTX 5060 Ti 16GB has 16GB of GDDR7 on a 128-bit bus, 180W total graphics power and broad CUDA support. The B60 wins on capacity; the 5060 Ti on compatibility and gaming.

Forge PC Staff3 min

How to Use Speculative Decoding in LM Studio to Speed Up Local LLMs

How-To

How to Use Speculative Decoding in LM Studio to Speed Up Local LLMs

Speculative decoding in LM Studio pairs your main model with a much smaller draft model from the same family to speed up generation without changing output quality. Turn on Developer Mode, load the main model and pick a draft model in the chat sidebar. LM Studio says speed-ups vary by prompt and can vanish if the draft model is too big.

Forge PC Staff3 min