# RTX Spark compact desktops from ASUS, Dell and MSI go on sale with up to 128 GB of unified memory

> OnPrem World · https://onpremworld.com/en/news/2026-11-nvidia-rtx-spark-compact-desktops-available · CC BY 4.0

Published: 2026-11-18 · Hardware

NVIDIA said on November 18 that RTX Spark compact desktops from ASUS, Dell and MSI are available to buy, joining the RTX Spark laptops already on sale. The desktops pair a 20-core Grace CPU with a Blackwell RTX GPU with 6,144 CUDA cores, up to 128 GB of unified memory and 10-gigabit Ethernet, and NVIDIA says they can run 120-billion-parameter language models locally. Mini PCs from Acer, GIGABYTE and HP and the Microsoft Surface RTX Spark Dev Box are listed as coming soon. NVIDIA gave no prices or regions; the same post notes that Windows ML can now run GGUF models from Hugging Face through a llama.cpp backend.

## Why it matters locally

A small Windows desktop with up to 128 GB of memory the GPU can use is now something you can actually order, not just a launch announcement.

## For the market

Three brands shipping at once, with four more named as coming, suggests RTX Spark desktops will be priced by competition between PC makers rather than by NVIDIA alone, which could work in buyers' favour. Without published prices it is too early to compare them with DGX Spark, which our data lists at US$3,999 at launch and reports at US$6,950 for the 128 GB Founders Edition since October 2026.

## In plain words

Small desktop computers that can run AI assistants entirely on your own desk are now in shops from three well-known brands. Because the work happens on the machine, your files stay with you and there are no per-use cloud fees for those tasks.

## For practitioners

These are Windows on Arm machines with native CUDA and Linux tooling through Windows Subsystem for Linux, so test your Python packages and native extensions on Arm64 first. Our data lists RTX Spark at about 300 GB/s of memory bandwidth, so the 128 GB is best used for mixture-of-experts models or several mid-size models rather than one large dense model; NVIDIA names Unsloth for fine-tuning, and 10GbE makes the box usable as a shared inference host.

## Source

[NVIDIA Blog: NVIDIA RTX Spark Compact Desktops Out Now](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/)

## On this site

- Hardware: [RTX Spark (N1X) laptops & compact desktops](https://onpremworld.com/en/hardware/nvidia-rtx-spark-n1x.md)
- Hardware: [DGX Spark](https://onpremworld.com/en/hardware/nvidia-dgx-spark.md)

[All news](https://onpremworld.com/en/news.md)

Summaries are written by OnPrem World from the official source. We link to the original and never republish it.
