NVIDIA DGX Spark™ - NVIDIA -Macnica

NVIDIA DGX Spark™ Product Brief

Features

Powered by the NVIDIA GB10 Grace Blackwell Superchip, NVIDIA DGX Spark™ delivers petaFLOP1 of AI performance in a power-efficient, compact form factor. With the NVIDIA AI software stack and 128 GB of pre-installed memory, developers can locally access up to 200 billion parameters to prototype, fine-tune, and infer the latest generation of reasoning AI models from DeepSeek, Meta, NVIDIA, Google, Qwen, and others.

Lineup/Product Information

DGX Spark is available from NVIDIA authorized OEM partners as well as from NVIDIA authorized OEM partners.
OEM partners are integrating the same GB10 and NVIDIA AI software stack technologies used in DGX Spark into their product families. Macnica offers DGX Spark from MSI, Dell Technologies, and ASUS in addition to the original NVIDIA product.

Comparison table of specifications of each company

	NVIDIA	MSI		Dell Technologies		ASUS
product name	DGX Spark	EdgeXpert 1TB	EdgeXpert 4TB	Dell Pro Max with GB10 2TB	Dell Pro Max with GB10 4TB	Ascent GX10 (1TB)
architecture	NVIDIA Grace Blackwell	NVIDIA Grace Blackwell		NVIDIA Grace Blackwell		NVIDIA Grace Blackwell
CPU	20-core Arm 10 Cortex-X925 + 10 Cortex-A725 Arm	20-core Arm 10 Cortex-X925 + 10 Cortex-A725 Arm		20-core Arm 10 Cortex-X925 + 10 Cortex-A725 Arm		20-core Arm 10 Cortex-X925 + 10 Cortex-A725 Arm
GPUs	Blackwell generation (5th generation Tensor/4th generation RT)	Blackwell generation (5th generation Tensor/4th generation RT)		Blackwell generation (5th generation Tensor/4th generation RT)		Blackwell generation (5th generation Tensor/4th generation RT)
AI performance	Up to 1 PFLOP (FP4)	Up to 1 PFLOP (FP4)		Up to 1 PFLOP (FP4)		Up to 1 PFLOP (FP4)
system memory	128GB LPDDR5x integrated system memory	128GB LPDDR5x integrated system memory		128GB LPDDR5x integrated system memory		128GB LPDDR5x integrated system memory
storage	4TB	1TB	4TB	2TB	4TB	1TB
Network	1xRJ-45 10GbE + NVIDIA ConnectX-7 Smart NIC	1xRJ-45 10GbE + NVIDIA ConnectX-7 Smart NIC		1xRJ-45 10GbE + NVIDIA ConnectX-7 Smart NIC		1xRJ-45 10GbE + NVIDIA ConnectX-7 Smart NIC
wireless	Wi-Fi 7、Bluetooth 5.4	Wi-Fi 7、Bluetooth 5.4		Wi-Fi 7、Bluetooth 5.4		Wi-Fi 7、Bluetooth 5
Video Output	1xHDMI 2.1	1xHDMI 2.1		1xHDMI 2.1		1xHDMI 2.1
USB	4xUSB TypeC	4xUSB TypeC		4xUSB TypeC		4xUSB TypeC
size	150 × 150 × 50.5 mm	151 × 151 × 52mm		150 × 150 × 50.5mm		150 × 150 × 51 mm
OS	NVIDIA DGX OS (Ubuntu-based)	NVIDIA DGX OS (Ubuntu-based)		NVIDIA DGX OS (Ubuntu-based)		NVIDIA DGX OS (Ubuntu-based)
Connectivity	Supports up to 405B parameters when connecting two devices (using ConnectX)	Supports up to 405B parameters when connecting two devices (using ConnectX)		Supports up to 405B parameters when connecting two devices (using ConnectX)		Supports up to 405B parameters when connecting two devices (using ConnectX)
power cable	attached	attached		attached		attached
Product warranty	1 year	1 year, 2 years, 3 years		1 year, 2 years, 3 years		1 year, 2 years, 3 years
Notices	No maintenance	Warranty extension available only at time of purchase		Maintenance subscription required		Maintenance subscription required
Connected DAC	can be	can be		can be		can be
Datasheet URL	NVIDIA Datasheet	MSI Datasheet		Dell Technologies Datasheet		ASUS Datasheet

DGX Spark's Placement in NVIDIA DGX™ Systems

NVIDIA In 2016, the company AI To provide researchers with their own supercomputers NVIDIA DGX-1^™ After that, DGX The platform continues to grow at a remarkable pace and is currently TOP500 and Green500 Many of the world's most powerful supercomputers are registered on NVIDIA DGX It is built on a platform. NVIDIA DGX New to the platform is a new class of AI It is a supercomputer DGX Spark is. DGX Spark To all developers AI It was developed to deliver the next wave of breakthroughs by delivering computing power of petaflops in a compact desktop form factor. AI Performance and 128GB This allows developers to 2,000 With billions of parameters AI Run inference on the model and get up to 700 You can locally fine-tune models with billions of parameters.

Visit the NVIDIA DGX™ Systems product page

How to use DGX Spark

What can you do?

- Prototyping: Develop, test, and validate applications and solutions using the NVIDIA AI software stack and a range of open source software.

- Fine-tuning: Fine-tuning a trained AI model can improve inference accuracy.
128GB of integrated system memory enables fine-tuning of AI models with up to 70 billion parameters.

-Data Science: Maximize performance for large-scale, complex data analysis.

-Inference: The combination of 5th-generation Tensor Cores supporting FP4 and 128GB of system memory delivers up to 1 petaFLOP of high-speed computing power, accelerating inference for cutting-edge AI models.

Edge Applications: Develop edge applications using a variety of NVIDIA AI frameworks, including NVIDIA Isaac™ and NVIDIA Metropolis.

Usage scene

One supercomputer for each AI researcher/developer

Finally, every AI researcher and developer has their own supercomputer. DGX Spark delivers up to 1 petaFLOP of AI performance while remaining quiet and small enough to fit on a desktop. With the NVIDIA AI software stack pre-installed, AI researchers and developers can get started on their AI projects right away.

Demonstration experiment of a system using generative AI

With up to 1 petaFLOP of AI performance and 128GB of CPU-GPU coherent memory, advanced generative AI models, including large-scale language models (LLMs), can be run locally. This enables rapid implementation and verification of systems using generative AI, even in areas where privacy and security are important.

education

Any AI software from around the world, from NVIDIA to the open source community, can be used on DGX Spark. You can learn anything about AI on DGX Spark without relying on the cloud. DGX Spark is also an ideal platform for AI education, from educational institutions to corporate training.

demo video

Running NVIDIA AI Blueprint for video search and summarization (VSS) on DGX Spark

I tried building RAG Chat on DGX Spark with GPT-OSS 120B

Macnica original purchase bonus

Starter Guide (free)

This guide explains everything from setting up DGX Spark and its OEM products to building a Japanese-speaking RAG chatbot. Please consider this guide to be a supplement to the official documentation published by NVIDIA. It also carefully explains points of caution that experts often overlook, with consideration for those unfamiliar with NVIDIA GPU environments. The chatbot example assumes user prompt input in Japanese.

Table of contents (※Subject to change without notice)

Purpose of this book
Setup Method
Items to be prepared separately
Initial Setup Steps
Japanese input
DGX Dashboard
Jupyter Lab
Visual Studio Code
NVIDIA Sync
Japanese RAG chatbot using GPT-OSS-120B
Future exploration
Inquiries/Support Information

Ticket-based support (paid)

We provide email-based Q& A and support for questions about the specifications, features, and actual operation of DGX Spark (NVIDIA, MSI, Dell Technologies, ASUS) purchased from us, as well as technical questions related to the NVIDIA SDK.

*For inquiries regarding the NVIDIA SDK, a ticket will be opened once the problem has been resolved or information has been provided by NVIDIA (via the NVIDIA Developer Forums).

Response policy

We will respond based on the materials provided by each company, our own knowledge, and the results of verification using our own equipment, and will also consult with the manufacturer if necessary.

Reception

Please send us the terms and conditions separately. Please contact us using the quote form at the bottom of the page.
* Manufacturer name of the product you want [NVIDIA CorporationPlease select
For inquiries and requests regarding estimates, please contact us atDGX SparkPlease write "I would like to use the paid ticket-based support service."