> ## Content Index
> Fetch the complete content index at: https://www.notatechguy.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# NVIDIA RTX Spark desktops launch with 128GB unified memory
- URL: https://www.notatechguy.com/nvidia-rtx-spark-desktops-launch-with-128gb-unified-memory/
- Published: 2026-11-18T16:36:04.000Z
- Updated: 2026-11-18T16:36:04.000Z
- Description: NVIDIA RTX Spark compact desktops from ASUS, Dell and MSI ship today with up to 128GB unified memory and a 20-core Grace CPU for local AI.
- Author: Marcello Babbili
- Tags: Technology & AI, Nvidia, AI Chips

NVIDIA is shipping compact desktops today that carry up to 128GB of unified memory and a claimed 1 petaflop of AI performance [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). According to NVIDIA, that is enough to run a 120-billion-parameter language model locally, a workload that has lived almost entirely on cloud servers [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

**My read:** The 128GB unified memory figure is the number that matters. Running a 120-billion-parameter model locally crosses a threshold for developers who want to prototype agents without per-token API costs. But every spec carries "up to" language, and NVIDIA has not published pricing, base configurations, or independent benchmarks for the 1 petaflop claim. I would wait for third-party reviews before treating the peak numbers as typical performance.

ASUS, Dell and MSI have RTX Spark compact desktops on sale now [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). Mini PCs from Acer, GIGABYTE and HP are listed as coming soon, along with a Microsoft Surface RTX Spark Dev Box [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). RTX Spark laptops are already on shelves [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). Jensen Huang first unveiled the line at COMPUTEX 2026 [P⁴](https://www.nvidia.com/en-us/geforce/news/gfecnt/20266/computex-2026-nvidia-geforce-rtx-announcements/?ref=notatechguy.com).

### What is inside the box

Each desktop pairs a 20-core NVIDIA Grace CPU with a Blackwell RTX GPU carrying 6,144 CUDA cores [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). The Grace CPU and the GPU share a single pool of up to 128GB of unified memory, meaning data does not have to be copied between separate CPU and GPU memory banks [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). Ten-gigabit Ethernet handles file transfers [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

NVIDIA says the combination can render 90GB-plus 3D scenes in real time and smoothly edit 4:2:2 video, a high-bitrate broadcast format, without proxy files [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

No one outside the company has verified these claims.

### The software story

Adobe has shipped native Windows on Arm versions of its creative apps for RTX Spark [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). Blender gains NVIDIA's DLSS Ray Reconstruction, which uses AI to improve ray-traced image quality [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). ComfyUI, an image-generation workflow tool, runs locally [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

For developers, the desktops support native CUDA through Windows Subsystem for Linux, so existing GPU code written for Linux servers should port with minimal changes [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). NVIDIA's Hermes Agent and NemoClaw blueprints let developers set up AI agents, and Unsloth, a fine-tuning library, is supported for adapting models on the machine itself [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com). NVIDIA Switchyard, now available in Hermes, rounds out the agent stack [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

A GitHub project, devjeonghwan/vllm-for-rtx-spark, appeared on October 27 with 17 stars and a single contributor, adapting the vLLM inference engine for RTX Spark on Windows on Arm [P³](https://github.com/devjeonghwan/vllm-for-rtx-spark?ref=notatechguy.com). It is early, but it signals that the open-source community is already moving to support the platform.

### Who this is for

A small AI studio that currently pays per-token API costs to run a large model for prototype agents could, if the 120-billion-parameter claim holds, run that workload on a desktop with no recurring inference bill. A video production house editing 4:2:2 footage could skip the proxy-file workflow that adds time to every project. NVIDIA demonstrated both scenarios at Adobe MAX [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

What is missing: pricing, Australian availability, power consumption figures, and base-tier specifications. Every number NVIDIA has published reflects the maximum configuration. The entry-level memory, core count, and performance could be substantially lower.

Acer, GIGABYTE and HP have not announced release dates for their Mini PCs, nor has Microsoft for the Surface RTX Spark Dev Box [S¹](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com).

---

*Sources: [S1 — NVIDIA RTX Spark Compact Desktops Out Now](https://blogs.nvidia.com/blog/rtx-spark-desktop-local-ai-available-now/?ref=notatechguy.com) · [P2 — Slim Laptops & Small Desktops | NVIDIA RTX Spark](https://www.nvidia.com/en-us/products/rtx-spark/?ref=notatechguy.com) · [P3 — devjeonghwan/vllm-for-rtx-spark](https://github.com/devjeonghwan/vllm-for-rtx-spark?ref=notatechguy.com) · [P4 — NVIDIA RTX Spark Unveiled](https://www.nvidia.com/en-us/geforce/news/gfecnt/20266/computex-2026-nvidia-geforce-rtx-announcements/?ref=notatechguy.com) · [P5 — NVIDIA RTX Spark PCs Available Today — A New Beginning for Windows PCs](https://nvidianews.nvidia.com/news/nvidia-rtx-spark-pcs-available-today-a-new-beginning-for-windows-pcs?ref=notatechguy.com)*

---

*Written from 5 sourced items, 5 of them primary.*

## More from Not A Tech Guy

- [Nvidia, AMD chip demand booms despite OpenAI's frontier pause](https://www.notatechguy.com/nvidia-amd-chip-demand-booms-despite-openai-s-frontier-pause/)
- [Xing 4: China's 285B AI model on 8,192 Ascend cards](https://www.notatechguy.com/xing-4-china-s-285b-ai-model-on-8-192-ascend-cards/)
- [Mastra TypeScript AI framework trends on GitHub with 40+ providers](https://www.notatechguy.com/mastra-typescript-ai-framework-trends-on-github-with-40-providers/)