AI/News

Nvidia's DGX Spark 64GB goes on sale 23 October at $4,999: models to 100 billion parameters on one box, two boxes pool to 128GB

Nvidia's own post: the same GB10 Grace Blackwell chip and software as the 128GB model, sold only through Acer, ASUS, Dell, Gigabyte, HP and MSI; two units cabled together reach 200-billion-parameter models and "up to 1.7x the performance" in Nvidia's Qwen 3.8 27B test. The page does not say what the 128GB model now costs, and coverage disagrees.

By Daily Aletheia · Checked against the primary source · 5 October 2026 · 2 min read


Image: Nvidia - the DGX Spark 64GB image from "NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI", blogs.nvidia.com, 2 October 2026

Nvidia announced a 64GB DGX Spark on 2 October, in a post by Allen Bourgoyne: "Starting Friday, Oct. 23, the new configuration will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI", at a price "starting at $4,999". The 64GB unit is "available exclusively from manufacturer partners" and keeps "the GB10 Grace Blackwell Superchip, DGX OS and full NVIDIA AI software stack - same as the 128GB model".

What one box runs. "It supports up to 100-billion-parameter models and the agentic applications built on them, fully on device." Nvidia's three suggested uses: "Run an AI agent around the clock" - a coding or research agent kept running on the box; serve a language or image model to laptops and desktops on the same network; and scale to two units when "a larger model, a longer context window or concurrent agent requests" outgrow one.

What two boxes run. Every unit ships with a ConnectX-7 network port; two "can connect directly with a QSFP cable, pooling their memory to 128GB and expanding model support to up to 200 billion parameters while delivering twice the memory bandwidth and up to 1.7x the performance". The 1.7x is from "NVIDIA's Qwen 3.8 27B test". A Sync Cluster Assistant "detects connected units, validates device configuration and configures the ConnectX-7 network"; a Sync Model Launcher, "coming at the end of the month", downloads and launches Qwen3.8 27B on one box or a pair and sets up OpenCode against it.

Software out of the box. The post lists the Agent Toolkit, CUDA-X libraries, Nemotron open models, and "Ollama, vLLM, and PyTorch with CUDA"; the getting-started list adds llama.cpp and LM Studio. Blender is named as "among the first major creator application providers to support the platform", with an installer "coming soon".

What the page does not say. It gives no price for the 128GB model and no comparison to it, only that the 64GB "keeps the platform at an accessible price point". Coverage filled the gap two ways: VideoCardz reports the 64GB costs more than the original 128GB launch price, while Hardware Busters reports the 128GB has risen to $6,950. Neither figure is on Nvidia's page.

Every figure above is Nvidia's own, read off its post.

The Check

What was said
"DGX Spark drops to 64GB memory but costs more than the original 128GB version" (VideoCardz); "the 128GB model jumps to $6,950" (Hardware Busters)
What the Source Says
Nvidia's post gives one price, "starting at $4,999" for the 64GB, and says the new configuration "keeps the platform at an accessible price point". It names no price for the 128GB model, old or new.
The Difference
The $4,999 holds. Every comparison to the 128GB model is coverage, and the two outlets disagree on what that price is. Verdict on the comparison: their words, not Nvidia's.

Sources & further reading

  1. 01Nvidia - NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI, Allen Bourgoyne, 2 Oct 2026 ↗Primary
  2. 02VideoCardz - NVIDIA DGX Spark drops to 64GB memory but costs more than the original 128GB version (the comparison as it travelled) ↗Attributed
  3. 03Hardware Busters - Nvidia DGX Spark 64GB Arrives at $4,999 as the 128GB Model Jumps to $6,950 (the comparison as it travelled) ↗Attributed