NVIDIA DGX Spark AI Supercomputer 4TB

NVIDIA DGX Spark AI Supercomputer 4TB (940-54242-0007-000) - 1 Year Local Warranty 

NVIDIA DGX Spark – Enterprise-Grade AI Supercomputer

The NVIDIA DGX Spark AI Supercomputer is a compact yet powerful on-premise AI system built for enterprises, research labs, and advanced developers.
It is designed to deliver high-density AI compute for machine learning, deep learning, generative AI, and data-intensive workloads.
As part of the NVIDIA DGX platform, it provides enterprise-grade reliability, optimized performance, and a production-ready AI infrastructure in a desktop form factor.
This makes it ideal for organizations that require secure, on-site AI computing without relying solely on the public cloud.

Built for Generative AI, LLM Training, and High-Performance Inference

DGX Spark is optimized for large language model training, fine-tuning, and high-throughput AI inference.
It enables faster experimentation, shorter training cycles, and real-time AI model deployment for business-critical workloads.
With support for modern AI frameworks such as PyTorch, TensorFlow, and NVIDIA CUDA-accelerated libraries, it accelerates end-to-end AI development pipelines.
This makes it ideal for generative AI, multimodal AI, computer vision, natural language processing, and advanced data science projects.

4TB High-Speed NVMe Storage for Data-Intensive AI Workflows

The 4TB ultra-fast NVMe storage is engineered for high-speed data access, rapid dataset loading, and accelerated model checkpointing.
It allows teams to work with massive training datasets, large AI models, and continuous experimentation without I/O bottlenecks.
This improves performance across data preprocessing, training, validation, and model deployment stages.
It is ideal for organizations handling large volumes of training data, synthetic data generation, and real-time AI pipelines.

Scalable, Secure, and Ready for Enterprise AI Deployment

DGX Spark integrates seamlessly with NVIDIA AI Enterprise software, providing stability, security, and long-term production support.
It supports secure AI operations with enterprise-grade system management, monitoring, and lifecycle tools.
The system is designed for continuous 24/7 workloads with optimized cooling and stable power delivery.
It also supports scaling across multiple systems, making it suitable for growing AI teams and expanding workloads.

Ideal for On-Prem AI, Edge AI, and Confidential Workloads

DGX Spark is well suited for organizations that require local AI processing for data privacy, compliance, and intellectual property protection.
It enables on-premise AI development for regulated industries such as finance, healthcare, research, defense, and government.
It is also suitable for edge AI environments, enabling real-time inference and low-latency AI deployment at the source of data.
This makes it a strong choice for enterprises building secure, high-performance AI infrastructure without cloud dependency.

Technical Specification – NVIDIA DGX Spark AI Supercomputer (940-54242-0007-000)

Model: NVIDIA DGX Spark AI Supercomputer
GTIN: 812674029197
Category: Compact Desktop AI Supercomputer

Chip Architecture

• NVIDIA GB10 Grace Blackwell Superchip (CPU + GPU unified architecture)

• Built for advanced AI development, LLM training, generative AI, and enterprise inference workloads

CPU

• 20-core ARM CPU:

  • 10 × Cortex-X925 high-performance cores

  • 10 × Cortex-A725 efficiency cores

• Optimized for parallel compute, AI preprocessing, and mixed workload efficiency

GPU / AI Acceleration

• Integrated NVIDIA Blackwell GPU

• 5th Generation Tensor Cores (AI acceleration, FP4/FP8 optimized)

• 4th Generation RT Cores for graphics, simulation, and rendering acceleration

• Ideal for LLM workloads, multimodal AI, diffusion models, and inference pipelines

AI Performance

• Up to 1 PFLOPS FP4 AI compute

• Supports next-generation generative AI, RAG systems, HPC simulations, and enterprise fine-tuning

Unified Memory System

• 128GB LPDDR5X Unified System Memory (shared CPU–GPU)

• Memory Bandwidth: Up to 273 GB/s

• Unified coherent memory model via NVLink-C2C, eliminating CPU-GPU bottlenecks

• Excellent for training/fine-tuning LLMs, embeddings, and large transformer models

Storage

• 4TB NVMe M.2 Self-Encrypting SSD

• High-speed read/write performance for datasets, checkpoints, and enterprise workflows

Networking & Interconnect

• 10GbE RJ-45 Ethernet (standard networking)

• NVIDIA ConnectX-7 SmartNIC:

  • Up to 200Gbps high-speed interconnect

  • RDMA / InfiniBand capable

  • Supports multi-node AI clusters and distributed training

• Wireless: Wi-Fi 7 + Bluetooth 5.x

I/O Ports

• 4 × USB-C (USB 3.x high-bandwidth connectivity)

• HDMI 2.1a video output

• Supports modern displays and remote-access console workflows

Video Encode / Decode

• NVENC ×1 (hardware video encoder)

• NVDEC ×1 (hardware video decoder)

• Suitable for video AI, streaming, and real-time media workloads

Software Platform

• NVIDIA DGX OS (Ubuntu LTS-based enterprise AI OS)

• Full NVIDIA AI software stack:

  • NVIDIA AI Enterprise Suite

  • CUDA Toolkit

  • TensorRT

  • PyTorch & TensorFlow optimized containers

  • NGC-certified AI containers

Cooling & Power

• Enterprise-grade active cooling system for sustained high-load performance

• External desktop-class power adapter

• Efficient thermal management for long-duration AI compute tasks

Form Factor

• Ultra-compact desktop AI supercomputer

• Dimensions: ~150 mm × 150 mm × 50.5 mm

• Weight: ~1.2 kg

• Designed for lab environments, R&D, on-prem AI, and edge AI deployment

Workload Suitability

• Primary:

  • LLM training

  • LLM fine-tuning

  • Large-scale inference

• Secondary:

  • Generative AI (image, video, multimodal)

  • Data science & vector embeddings

  • Scientific computing & HPC research

  • Enterprise AI development

Security

• Secure Boot (UEFI)

• TPM-based security and encryption protections

• Self-encrypting SSD for data-at-rest protection

Accessories & Power Cord

• Included: US Type-B Power Cord (factory sealed)

• UK Power Cord will be provided separately