NVIDIA DGX Spark AI Supercomputer - 4TB
PRODUCT DESCRIPTION
NVIDIA DGX Spark AI Supercomputer 4TB (940-54242-0007-000) - 1 Year Local Warranty
NVIDIA DGX Spark — A Petaflop of AI Performance on Your Desk
The NVIDIA DGX Spark puts up to 1 petaFLOP of FP4 AI compute in a 1.2 kg desktop box. Powered by the NVIDIA GB10 Grace Blackwell Superchip with 128 GB of coherent unified memory, it runs inference on models up to 200 billion parameters and fine-tunes models up to 70 billion parameters — locally, on your desk, with no cloud dependency.
Enterprise AI Without the Cloud — Privacy, Compliance, Control
DGX Spark brings the NVIDIA DGX platform to enterprises, research labs and advanced developers who need on-premise AI for data privacy, compliance and IP protection — finance, healthcare, research, defense and government. The full NVIDIA AI software stack comes preloaded on DGX OS: CUDA, TensorRT, PyTorch and TensorFlow containers, NVIDIA NIM and NGC-certified containers.
Scale to 405B Parameters with Two Linked Sparks
The built-in NVIDIA ConnectX-7 SmartNIC (200 Gbps, RDMA-capable) lets you link two DGX Spark units into a single system that handles models up to 405 billion parameters — pair yours with a second unit and the NVIDIA QSFP112 DAC cable (X0101G00400A).
Deployment-Ready in Singapore
In stock at SourceIT with GST invoice, 1 year local warranty, and our NVIDIA DGX Spark Setup and Enterprise Deployment Guide to get you running from day one.
NVIDIA DGX Spark Datasheet — Technical Specifications
General
| Model | NVIDIA DGX Spark AI Supercomputer — 4TB |
| Part Number | 940-54242-0007-000 |
| GTIN | 812674029197 |
| Category | Compact Desktop AI Supercomputer |
| Warranty | 1 Year Local Warranty (Singapore) |
Processor — NVIDIA GB10 Grace Blackwell Superchip
| Architecture | CPU + GPU Unified (Grace Blackwell), NVLink-C2C Interconnect |
| CPU | 20-Core Arm — 10 × Cortex-X925 Performance + 10 × Cortex-A725 Efficiency |
| GPU | NVIDIA Blackwell Architecture |
| Tensor Cores | 5th Generation (FP4 / FP8 Optimised) |
| RT Cores | 4th Generation |
| GB10 TDP | 140 W |
AI Performance
| Peak AI Compute | Up to 1 PFLOP (1,000 TOPS) FP4 with Sparsity |
| Inference | Models Up to 200 Billion Parameters (Single Unit) |
| Fine-Tuning | Models Up to 70 Billion Parameters |
| Clustered (2 Units) | Models Up to 405 Billion Parameters via ConnectX-7 |
Memory & Storage
| System Memory | 128 GB LPDDR5x Coherent Unified Memory (Shared CPU–GPU) |
| Memory Interface | 256-bit, 273 GB/s Bandwidth |
| Storage | 4 TB NVMe M.2 Self-Encrypting SSD |
Networking & I/O
| SmartNIC | NVIDIA ConnectX-7 — Up to 200 Gbps, RDMA Capable, Multi-Node Clustering |
| Ethernet | 1 × 10 GbE RJ-45 |
| Wireless | Wi-Fi 7, Bluetooth 5.4 |
| USB | 4 × USB Type-C |
| Display | 1 × HDMI 2.1a; Up to 3 × DisplayPort via USB-C Alt Mode |
| Video Engines | 1 × NVENC, 1 × NVDEC |
Software
| Operating System | NVIDIA DGX OS (Ubuntu LTS-Based) with Preloaded AI Stack |
| AI Stack | NVIDIA AI Enterprise, CUDA Toolkit, TensorRT, NVIDIA NIM, PyTorch & TensorFlow Containers, NGC-Certified Containers |
| Security | Secure Boot (UEFI), TPM-Based Security, Self-Encrypting SSD |
Power & Physical
| Power Supply | 240 W External Adapter |
| Dimensions | 150 mm × 150 mm × 50.5 mm |
| Weight | Approximately 1.2 kg |
| Power Cord | US Type-B Cord Included (Factory Sealed); UK Power Cord Provided Separately by SourceIT |
Workloads & Use Cases
Primary: LLM training, fine-tuning and large-scale inference — faster experimentation, shorter training cycles and real-time deployment for business-critical workloads.
Also ideal for: generative AI (image, video, multimodal), RAG systems and vector databases, agentic AI development, data science and embeddings, computer vision, scientific computing/HPC research, and edge AI with low-latency local inference.
Regulated industries: keeps sensitive data on-premise for finance, healthcare, research, defense and government deployments where public cloud is not an option.
Ordering Information — AI Supercomputers at SourceIT
| NVIDIA DGX Spark 4TB | 940-54242-0007-000 (This Product) |
| ASUS Ascent GX10 1TB (GB10 Platform) | GX10-GG0007BN — In Stock |
| ASUS Ascent GX10 2TB (GB10 Platform) | GX10-GG0030BN — In Stock |
| ASUS Ascent GX10 4TB (GB10 Platform) | GX10-GG0031BN — In Stock |
| NVIDIA QSFP112 400G DAC Cable (Links 2 Units) | X0101G00400A — In Stock |
| MSI XpertStation WS300 (DGX Station GB300, 748GB) | In Stock — Enquire |
| HP ZGX Fury G1n GB300 AI Workstation | Enquire |
Datasheet, Downloads & Setup Guide
Frequently Asked Questions — NVIDIA DGX Spark
What AI models can the DGX Spark run?
A single DGX Spark handles inference on models up to 200 billion parameters and fine-tuning up to 70 billion parameters — think Llama, Qwen, DeepSeek, Mistral and Flux-class models locally. Two linked units scale to 405 billion parameters.
How do I connect two DGX Spark units?
Via the built-in ConnectX-7 SmartNIC using the NVIDIA QSFP112 400G DAC cable — a direct connection that turns two Sparks into one system for larger models and distributed training.
What is the difference between the NVIDIA DGX Spark and the ASUS Ascent GX10?
Both are built on the same NVIDIA GB10 Grace Blackwell platform with 128 GB unified memory and up to 1 PFLOP FP4. The DGX Spark is NVIDIA's first-party unit with 4 TB storage; the ASUS GX10 offers 1 TB, 2 TB and 4 TB tiers at different price points with onsite warranty.
Do I need special power or cooling?
No. The DGX Spark draws just 240 W from a standard wall socket with an external adapter and enterprise-grade active cooling — it runs on a desk, not a data centre. Note: the factory-sealed box includes a US power cord; SourceIT provides a UK/Singapore cord separately.
What software comes preinstalled?
NVIDIA DGX OS with the full NVIDIA AI stack: CUDA, TensorRT, NVIDIA NIM, PyTorch/TensorFlow containers and NGC-certified containers — ready for development on day one. See our setup and deployment guide.
Is the DGX Spark covered by local warranty in Singapore?
Yes. Every DGX Spark sold by SourceIT comes with 1 year local warranty and a GST invoice. Volume and enterprise procurement enquiries are welcome.
