NVIDIA DGX Spark 1x10GbE 2x200GbE 4TB NVMe
NVIDIA DGX Spark
940-54242-0005-000
- Equipped with the GB10 Superchip
- 1 petaFLOP of AI performance using FP4
- NVIDIA Blackwell GPU with fifth-generation Tensor Core technology
- High-performance NVIDIA Grace CPU with a 20-core Arm architecture
- 128 GB of coherent unified memory
- 4 TB of NVMe storage
- Support for large language models with up to 200 billion parameters
- NVIDIA Connect-X network connectivity for linking two DGX Spark units and working with models of up to 405 billion parameters
- Compact design with a minimal desktop footprint
- Connects to a standard wall outlet
NVIDIA professional desktop computing solutions for AI development
Desktop AI Computing Requirements
The increasing size and complexity of generative AI models are making development on local systems more challenging. Prototyping, fine-tuning, and running inference on large models locally require substantial memory capacity and significant computing performance. As companies, software vendors, government agencies, startups, and researchers expand their AI teams, the need for AI computing resources continues to grow.
200B-Parameter Models on Your Desktop
The NVIDIA DGX Spark™ is part of a new class of computers designed from the ground up to build and run AI. Powered by the NVIDIA GB10 Grace Blackwell Superchip and based on the NVIDIA Grace Blackwell architecture, the NVIDIA DGX Spark™ delivers up to 1000 TOPS of AI performance for demanding AI workloads. With 128 GB of unified system memory, developers can experiment with, fine-tune, or run inference on models with up to 200B parameters. In addition, NVIDIA ConnectX™ connectivity can link two NVIDIA DGX Spark™ supercomputers to enable inference on models with up to 405B parameters.
To provide developers with a familiar experience, the NVIDIA DGX Spark™ uses the same software architecture that powers industrial-scale AI factories. With NVIDIA DGX OS based on Ubuntu Linux and pre-configured with the latest NVIDIA AI software stack, along with access to the NVIDIA NIM™ developer program and NVIDIA Blueprints, developers can quickly begin using familiar tools such as PyTorch, Jupyter, and Ollama to prototype, fine-tune, and run inference on the NVIDIA DGX Spark, then deploy seamlessly to the data center or cloud.
By delivering large-scale performance and capabilities in a compact form factor, the NVIDIA DGX Spark™ enables developers, researchers, data scientists, and students to continue pushing the boundaries of generative AI.
Built on NVIDIA Grace Blackwell
At the heart of the NVIDIA DGX Spark™ is the new NVIDIA GB10 Grace Blackwell Superchip, based on the NVIDIA Grace Blackwell architecture and optimized for a desktop form factor. The GB10 features a powerful NVIDIA Blackwell GPU with fifth-generation Tensor cores and FP4 support, delivering up to 1000 TOPS of AI compute performance. The GB10 includes a high-performance 20-core Grace Arm CPU to accelerate data preprocessing and orchestration, improving model fine-tuning and real-time inference. The GB10 Superchip uses NVLink™-C2C to provide a coherent memory model between the CPU and GPU, with 5 times the bandwidth of PCIe Gen 5.
Work with Large AI Models
With 128 GB of unified system memory and support for the FP4 data format, the NVIDIA DGX Spark™ supports AI models with up to 200B parameters, enabling AI developers to prototype, fine-tune, and run inference on large models at their desktops. With built-in NVIDIA ConnectX networking technology, two NVIDIA DGX Spark™ systems can be linked to work with even larger models, such as Llama 3.1 405B.
Develop Locally, Deploy Anywhere at Scale
NVIDIA AI Software Stack
- NVIDIA CUDA-X™ libraries for accelerated deep learning and machine learning
- NVIDIA AI Enterprise, providing cloud-native tools, frameworks, and microservices, such as NVIDIA NIM™ and NeMo™, for streamlined AI development and deployment
- NVIDIA AI Workbench, providing a unified, user-friendly environment for managing, developing, and scaling AI projects across local and cloud resources
- Support for leading frameworks such as TensorFlow, PyTorch, and JAX, all optimized for NVIDIA hardware
The NVIDIA DGX Spark™ provides organizations and developers with a powerful, cost-effective environment for model prototyping, freeing valuable computing resources in their cluster environments for the more suitable tasks of training and deploying production models. By using the NVIDIA AI platform software architecture, NVIDIA DGX Spark™ users can move their models seamlessly from the desktop to DGX Cloud or any accelerated cloud or data center infrastructure with virtually no code changes, making it easier than ever to prototype, fine-tune, and iterate.
NVIDIA GB10 Superchip
The NVIDIA Grace Blackwell architecture delivers a powerful system-on-a-chip (SoC) solution combining NVIDIA Blackwell GPU technology with Arm CPU cores to deliver supercomputer-level performance in an efficient, compact SoC.
CPU Cores
The GB10 provides 20 Arm cores, 10 Cortex-X925 cores, and 10 Cortex-A725 cores.
Blackwell GPU
The GB10 Blackwell GPU provides 6144 CUDA cores to accelerate graphics and FP32 compute workloads, 5th-generation Tensor cores that provide up to 1,000 AI TOPS1 of FP4 performance, and 4th-generation RT cores that provide real-time ray-tracing acceleration. The GPU also includes 5th-generation NVDEC and 9th-generation NVENC engines for accelerated encoding and decoding of MPEG-2, VC-1, H.264 (AVCHD), H.265 (HEVC), VP8, VP9, and AV1 video formats, including support for 4:2:2 H.264 and H.265 video.
128GB LPDDR5x Memory
The DGX Spark features 128GB of coherent, unified LPDDR5x system memory. Accessible to both CPU and GPU, DGX Spark enables users to work with large workloads, including AI models of up to 200 billion parameters.
ConnectX-7 NIC
Equipped with a ConnectX-7 Smart NIC, the DGX Spark provides up to 200 gigabit Ethernet (GbE). Connecting two DGX Spark systems via the ConnectX-7 networking port enables users to work with AI models of up to 405 billion parameters.
NVIDIA DGX OS
With NVIDIA DGX OS pre-installed, the DGX Spark development environment will be familiar to DGX developers. Based on Ubuntu Linux, the DGX OS system software makes it easy to move work to DGX Cloud or any accelerated data center or cloud infrastructure. The installed software also includes RTX toolkits and libraries supporting NVIDIA RTX visualization technologies.
NVIDIA AI Software
In addition to DGX OS, DGX Spark includes the NVIDIA AI software stack, which provides CUDA and CUDA-X toolkits and libraries to enable AI developers to quickly bring up their AI projects and workflows. Users can install their preferred AI frameworks, SDKs, and services, such as PyTorch, TensorFlow, or Jupyter Notebooks, or take advantage of NVIDIA platforms and frameworks such as Isaac, Metropolis, or Holoscan, along with NVIDIA NIMs and AI Blueprint example workflows.
NVIDIA DGX Spark™ Founders Edition Product Design
The NVIDIA DGX Spark is a compact desktop system, measuring 150 mm L x 150 mm W x 50.5 mm H, designed to fit on any desktop with minimal space requirements. The system is designed to resemble the original DGX 1 system, the original AI supercomputer first delivered to customers in 2016, making it an attractive addition to any workspace.
| Architecture | NVIDIA Grace Blackwell |
| GPU | Blackwell Architecture |
| CPU | 20-core Arm, 10 Cortex-X925 + 10 Cortex-A725 |
| CUDA Cores | Blackwell Generation |
| Tensor Cores | 5th Generation |
| RT Cores | 4th Generation |
| Tensor Performance 1 | 1,000 AI TOPS |
| System Memory | 128GB LPDDR5x, unified system memory |
| Memory Interface | 256-bit |
| Memory Bandwidth | 273 GB/s |
| OS | DGX OS |
| Storage | 4 TB NVME.M2 with self-encryption |
| USB | 4x USB Type C |
| Ethernet | 10 GbE 1x RJ-45 connector |
| NIC | ConnectX-7 NIC @ 200 Gbps |
| Wi-Fi | Wi-Fi 7 |
| Bluetooth | BT 5.4 |
| Power Consumption | 240 W |
| Display Connectors | 1x HDMI 2.1a |
| NVENC | NVDEC | 1x | 1x |
| System Dimensions | 150 mm L x 150 mm W x 50.5 mm H |
| System Weight | 1.2 kg |