NVIDIA DGX Spark

Key features of the NVIDIA DGX Spark 4TB mini computer

Data centre performance in a miniature body

The heart of the NVIDIA DGX Spark is the GB10 Grace Blackwell superchip – a unique combination of a 20-core ARM processor (10× Cortex-X925 + 10× Cortex-A725) and a Blackwell architecture GPU with 5th generation tensor cores and 4th generation RT cores. Both parts of the superchip are connected by NVLink-C2C technology, which provides 5× higher throughput than PCIe 5th generation and lets the CPU and GPU share a single 128GB pool of unified LPDDR5X memory. The result is immense performance that consumes a maximum of 240W and fits on any desk. This shared memory, without the traditional separation of GPU VRAM and system RAM, is the key advantage of the DGX Spark over standard desktop setups. Where a GeForce RTX 5090 graphics card offers 32GB of memory, the DGX Spark makes four times that available for AI models, without needing to quantise models or split them between multiple devices.

A personal supercomputer that fits on any desk

NVIDIA DGX Spark proves that you no longer need a server room or a rack full of hardware to run the largest AI models. With dimensions of 150×150×50.5mm and a weight of 1.2kg, the DGX Spark fits on any desk next to your laptop, yet offers performance that, just a few years ago, would have required an entire server full of graphics cards. Four USB-C 4.0 ports with speeds up to 40Gb/s, Wi-Fi 7, 10Gb Ethernet, and an HDMI output ensure you can connect all your accessories without needing an external hub or docking station. The DGX Spark is designed for developers, scientists, and AI professionals who want the full power of a local, personal AI supercomputer without relying on the cloud – right where they work.

Prototype and develop AI applications without limits

The complete NVIDIA AI software stack provides developers with a full-featured platform for creating AI models, AI agents, and AI-enhanced applications – all locally, without cloud dependency. Once a solution is ready for deployment or final fine-tuning, the DGX Spark enables direct and seamless migration of workloads to the NVIDIA DGX Cloud or other NVIDIA-accelerated infrastructure.

Fine-tune AI models with up to 70 billion parameters

Use the 128GB of unified memory in the NVIDIA DGX Spark to fine-tune pre-trained models with up to 70 billion parameters, right at your workstation. Training on your own data allows you to specialise AI models for specific needs, industry data, or particular use cases – without having to send sensitive data to the cloud or pay for expensive GPU instances. The result is a model tailored precisely to your scenario, created locally, securely, and under your full control.

NVIDIA DGX Spark 4TB delivers top-tier data science

The combination of 128GB of unified memory and 1 PFLOP of parallel throughput performance makes the NVIDIA DGX Spark the ideal workstation for demanding data analytics and machine learning projects. Large datasets, complex computational models, and intensive training and analysis processes that previously required a powerful cloud cluster or dedicated server now run right on your desk. Fast, local, and with no waiting for remote infrastructure.

Inference for models with up to 200 billion parameters

Fifth-generation Tensor Cores with FP4 format support achieve up to 1 PFLOP of performance and, combined with 128GB of system memory, enable you to run inference for state-of-the-art AI models with up to 200 billion parameters directly on your desk. Test, verify, and deploy models like Llama, Gemma, or Qwen locally and in real time – without cloud latency, shared infrastructure, or the risk of sensitive data leaks.

Application development for robotics, industry, and smart cities

The NVIDIA DGX Spark is an exceptional platform for developing robotics systems, smart city solutions, and computer vision applications. Pre-installed NVIDIA Isaac frameworks for robotics, Metropolis for video analysis, and Holoscan for real-time data processing let developers take full advantage of the DGX Spark's power to rapidly prototype and deploy applications – locally, without dependence on cloud infrastructure, and with full support from the NVIDIA ecosystem.