Nvidia DGX Spark GB10 1NVMe
- Part Number
- 940-54242-0006-000
- Condition
- New
- Manufacturer
- NVIDIA
- CPU
- 1x NVIDIA GB10 Grace CPU (10 cores Cortex-X925, 10 cores Cortex-A725)
- RAM
- 128GB (LPDDR5x)
- Power supply unit
- 1x 240W
- HDD Type
- NVMe
- HDD
- 4 TB NVME.M2 with self-encryption
- Network Card
- 1x RJ-45 connector 10 GbE
- GPU
- 1 x NVIDIA GB10
- OS
- NVIDIA DGX OS
- Accelerator manufacturer
- NVIDIA
- Accelerator model
- NVIDIA GB10 Grace Blackwell Superchip integrated GPU
- Installed GPU/APU count
- 1 integrated GPU
- Maximum GPU/APU count
- 1 integrated GPU
- Memory per accelerator
- 128GB unified system memory shared by the CPU and GPU
- Total accelerator memory
- 128GB
- GPU memory type
- LPDDR5X unified memory
- Accelerator memory bandwidth
- 273GB/s
- Accelerator form factor
- NVIDIA GB10 Grace Blackwell Superchip
- Accelerator interconnect
- Single integrated accelerator; dual-system scaling is provided through NVIDIA ConnectX-7 networking
- Accelerator-to-CPU interface
- On-package coherent interface within the NVIDIA GB10 Grace Blackwell Superchip
- Supported compute data types
- FP32, TF32, BF16, FP16, FP8, FP4 and INT8
- Accelerator virtualization
- Containerized CUDA application isolation through NVIDIA Container Toolkit
- Software ecosystem
- NVIDIA DGX OS, NVIDIA CUDA, CUDA-X, NVIDIA NGC, NVIDIA AI Enterprise, PyTorch and NVIDIA TensorRT-LLM
- Maximum system power
- 0.24kW
- Cooling
- Integrated air-cooled thermal management system
- Recommended workloads
- Local AI development and prototyping, inference with models up to 200 billion parameters, fine-tuning of compatible models up to 70 billion parameters, autonomous agents, data science and accelerated application development
- Dimensions
- 50.5 x 150 x 150 mm (H x W x D)
- Weight
- 1.2 kg
Need a technical consultation on choosing a server?
Submit a requestNVIDIA DGX Spark is a compact personal AI supercomputer designed for enterprise AI development, generative AI, agentic applications, machine learning, data science, and edge solution prototyping. Built around the NVIDIA GB10 Grace Blackwell Superchip, the platform brings an integrated Blackwell GPU, a 20-core Arm CPU, and 128GB of coherent unified memory to a desktop form factor.
Unlike a conventional workstation assembled from separate components, NVIDIA DGX Spark combines GPU acceleration, unified memory, local storage, high-speed networking, and the validated NVIDIA AI software stack in an integrated platform. It can operate as a standalone AI development node or be connected to a second DGX Spark system through NVIDIA ConnectX-7 networking for work with larger models.
GPU architecture for AI development and inference
NVIDIA DGX Spark is equipped with one NVIDIA GB10 Grace Blackwell Superchip. It combines a Blackwell-architecture GPU with fifth-generation Tensor Cores and fourth-generation RT Cores and a 20-core Arm CPU comprising 10 Cortex-X925 and 10 Cortex-A725 cores. The CPU and GPU share 128GB of coherent LPDDR5X unified system memory with 273GB/s of bandwidth.
The integrated Blackwell GPU delivers up to 1 PFLOP of AI performance at FP4 precision with sparsity and includes 6,144 CUDA cores. This architecture enables local development, testing, and inference with AI models of up to 200 billion parameters, as well as fine-tuning of supported models of up to 70 billion parameters.
NVIDIA ConnectX-7 networking provides high-speed connectivity for data movement and system-to-system workflows. Two DGX Spark systems can be connected to work with AI models of up to 405 billion parameters. The platform also includes 10GbE, Wi-Fi 7, Bluetooth 5.4, USB Type-C, and HDMI connectivity for integration into enterprise development environments.
This configuration includes one self-encrypting 4TB NVMe M.2 SSD for the operating system, models, datasets, and local project storage. NVIDIA DGX OS and the preinstalled NVIDIA AI software stack provide access to CUDA, NGC, PyTorch, TensorRT-LLM, container tools, optimized libraries, and frameworks used throughout enterprise AI workflows.
Primary use cases
- AI model development and prototyping. DGX Spark provides a local environment for creating, testing, and validating generative AI models, AI agents, machine learning applications, and enterprise proofs of concept.
- Model fine-tuning. The 128GB unified memory architecture supports fine-tuning of compatible models with up to 70 billion parameters for organization-specific datasets and business processes.
- Local inference. The platform can test, validate, and run inference with AI models of up to 200 billion parameters without sending every workload to external cloud infrastructure.
- Agentic and private AI applications. DGX Spark is suitable for developing autonomous agents, retrieval-augmented generation solutions, internal assistants, and applications that benefit from local data processing.
- Data science and edge development. The system supports accelerated analytics, computer vision, robotics, smart-city, and other edge application development using the NVIDIA AI ecosystem.
NVIDIA DGX Spark in enterprise infrastructure
NVIDIA DGX Spark is designed for organizations that need dedicated local AI resources for development, experimentation, validation, and selected production workloads without deploying a rack-scale GPU server for every project. Its compact form factor allows AI engineers, developers, researchers, and data scientists to work with the NVIDIA software ecosystem directly at the desktop.
Local execution can reduce dependence on cloud-based inference resources and helps teams keep selected models and datasets within the organization’s environment. For larger projects, two DGX Spark systems can be connected through ConnectX-7 networking, while validated applications can subsequently be migrated to NVIDIA DGX Cloud or NVIDIA-accelerated data center infrastructure.
With the GB10 Grace Blackwell Superchip, 128GB of coherent unified memory, one 4TB NVMe SSD, and the preinstalled NVIDIA AI software stack, DGX Spark provides a practical platform for organizations developing generative AI, autonomous agents, private AI services, data science solutions, and edge applications that require local control and a consistent path from prototype to deployment.
to Buy a Server
from Servermall
warranty
testing
We want to make sure that our equipment is fully reliable for you. That’s why all our equipment (even new) is subject to complete diagnostics, testing, and part replacement.
testing
We fully load our equipment with various utilities and software. Rest assured, it’ll work for a very long time.
Power supply
The components work at full capacity and are tested for several days. If we are not sure about something, we replace it.
of all equipment
Our regulations imply the same testing for both used and new equipment. Oh, and by the way, we complement the official manufacturer’s warranty with our own warranty.