NVIDIA DGX Spark, 20-core Arm Cortex CPU, 4TB Grace Blackwell AI Supercomputer, 4TB NVME.M2, 128GB Memory, 1 PFLOPS of FP4 AI Performance, ConnectX-7 Smart NIC | 940-54242-0006-000
Model : NVIDIA DGX Spark
SKU : NVIDIA DGX Spark 4TB
NVIDIA DGX Spark - 4TB
A Grace Blackwell AI Supercomputer on your desk.
> NVIDIA GB10 Grace Blackwell superchip
> 1 PFLOPS of FP4 AI performance
> 128GB of coherent, unified system memory
> ConnectX-7 Smart NIC
> 4TB NVME.M2 with self-encryption
> 150mm L x 150mm W x 50.5mm H
Description
GeForce RTX 50 Series - Game Changer! The Most Advanced AI platform for Gamers, Creators and beyond. Multiply performance with NVIDIA DLSS 4.5, generate images at unprecedented speed, and unleash your creativity with NVIDIA Studio.
The NVIDIA DGX Spark is a revolutionary AI Supercomputer that brings the power of the data center to your desk. It is built around the cutting-edge NVIDIA GB10 Grace Blackwell superchip, delivering 1 PFLOPS of FP4 AI performance. This ultra-compact system features 128GB of coherent, unified system memory, a rapid 4TB NVMe.M2 SSD with self-encryption, and a ConnectX-7 Smart NIC providing blazing 200 Gbps network connectivity. The DGX Spark is an essential tool for developers and researchers in AI, deep learning, and simulation, offering immense power in a form factor measuring just 150mm x 150mm.
Spark Something Big
Powered by the NVIDIA GB10 Grace Blackwell Superchip, NVIDIA DGX Spark™ delivers one petaFLOP1 of AI performance in a power-efficient, compact form factor. With the NVIDIA AI software stack preinstalled and 128 GB of memory, developers can prototype, fine-tune, and inference the latest generation of reasoning AI models from DeepSeek, Meta, NVIDIA, Google, Qwen and others with up to 200 billion parameters locally.
Workloads
Accelerate All AI Workloads
Delivering the power of an AI supercomputer in a desktop-friendly size, NVIDIA DGX Spark is ideal for AI developer, researcher, and data scientist workloads.
Desktop AI Compute Demands
The increasing size and complexity of generative AI models are making local development more challenging than ever. Prototyping, fine-tuning, and inferencing large models on desktop systems now require massive memory capacity and significant compute performance.
As enterprises, software providers, government agencies, startups, and researchers continue to scale their AI initiatives, the demand for powerful and accessible AI compute resources keeps growing.
200B Parameter Models on Your Desk
NVIDIA DGX™ Spark represents a new class of AI-first computers, purpose-built to develop and run modern AI workloads.
Powered by the NVIDIA GB10 Grace Blackwell Superchip, and based on the NVIDIA Grace Blackwell architecture, DGX Spark delivers up to 1 petaFLOP¹ of AI performance, enabling developers to handle demanding AI workloads with ease.
With 128 GB of unified system memory, developers can experiment, fine-tune, and run inference on models of up to 200 billion parameters² directly on their desktop. Using NVIDIA ConnectX™ networking, two DGX Spark systems can be connected to support inference on models as large as 405 billion parameters².
Enterprise-Grade Software, Desktop Simplicity
NVIDIA DGX Spark provides a familiar and seamless development experience by mirroring the same software architecture used in industrial-scale AI factories.
It ships with:
- NVIDIA DGX OS based on Ubuntu Linux
- A preconfigured NVIDIA AI software stack
- Developer access to NVIDIA NIM™ and NVIDIA Blueprints
Developers can immediately start working with popular tools such as PyTorch, Jupyter, and Ollama to prototype, fine-tune, and run inference locally—then easily scale workloads to the data center or cloud.
Built on NVIDIA Grace Blackwell
At the core of NVIDIA DGX Spark is the NVIDIA GB10 Grace Blackwell Superchip, optimized specifically for a desktop form factor.
The GB10 Superchip combines:
- A powerful NVIDIA Blackwell GPU with fifth-generation Tensor Cores and FP4 support
- A high-performance 20-core NVIDIA Grace Arm CPU for accelerated data preprocessing and orchestration
Using NVIDIA NVLink™-C2C, the GB10 delivers a coherent CPU+GPU memory model with 5× the bandwidth of PCIe Gen 5, significantly accelerating model tuning and inference.
Key Features
- Built on NVIDIA GB10 Grace Blackwell Superchip
- NVIDIA Blackwell GPU with fifth-generation Tensor Core technology
- NVIDIA Grace CPU with 20-core high-performance Arm architecture
- Up to 1 petaFLOP¹ of AI performance using FP4
- 128 GB unified, coherent system memory
- Supports AI models up to 200 billion parameters²
- NVIDIA ConnectX™ networking to link two systems for models up to 405 billion parameters²
- 4 TB NVMe storage
- Compact desktop form factor
Specification
| Architecture | NVIDIA Grace Blackwell |
| GPU | Blackwell Architecture |
| CPU | 20 core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm |
| CUDA Cores | Blackwell Generation |
| Tensor Cores | 5th Generation |
| RT Cores | 4th Generation |
| Tensor Performance | 1 PFLOP (Theoretical FP4 TOPS using sparsity feature) |
| System Memory | 128 GB LPDDR5x, coherent unified system memory |
| Memory Interface | 256-bit |
| Memory Bandwidth | 273 GB/s |
| Storage | 4 TB NVME.M2 with self-encryption |
| USB | 4x USB Type-C |
| Ethernet | 1x RJ-45 connector, 10 GbE |
| NIC | ConnectX-7 NIC @ 200 Gbps |
| Wi-Fi | WiFi 7 |
| Bluetooth | BT 5.4 |
| Audio Output | HDMI multichannel audio output |
| Power Supply | 240 Watts |
| Display Connectors | 1x HDMI 2.1a |
| NVENC | NVDEC | 1x | 1x |
| Operating System | NVIDIA DGX™ OS |
| System Dimensions | 150 mm (L) x 150 mm (W) x 50.5 mm (H) |
| System Weight | 1.2 kg |
| Product Name | NVIDIA DGX Spark |
| Product Code | 940-54242-0006-000 |
| Product Description | NVIDIA GB10 Grace Blackwell Superchip, 20 core Arm: 10 Cortex-X925 + 10 Cortex-A725 Arm, 128GB LPDDR5x, 4 TB NVME.M2 |
| Declared Mean A-weighted Sound Power Level (LWA,m) | 35 dB (Operating Mode), 19 dB (Idle) |
| Declared Mean A-weighted Emission Sound Pressure Level (LpA,m) | 29 dB (Operating Mode), 13 dB (Idle) |
| Statistical Adder for Verification (Kv) | 3 dB (Operating Mode), 3 dB (Idle) |
| Compliance | Measurements in conformance with ECMA-74 and declared with ECMA-109 (June 2025) |
| Notes | a) LWA,m computed as the arithmetic average of measured A-weighted sound power levels. b) LpA,m computed as the arithmetic average at the defined operator position (C.4b). c) Kv ensures 95% probability of acceptance under ECMA-109 verification. d) Operating mode under max GPU stress in 25°C ambient. Note 1: LWA,c = LWA,m + Kv. Note 2: dB = decibels. *Preliminary specifications, subject to change. |
