NVIDIA DGX Spark and DGX Station are two cutting-edge AI computing platforms designed to accelerate AI development, training, and inference with powerful hardware and optimized software stacks. Both systems leverage NVIDIA’s latest Grace Blackwell architecture but target different use cases and scales of AI workloads. Here is a detailed overview of each:
NVIDIA DGX Spark
Overview:
The NVIDIA DGX Spark is a compact, power-efficient AI workstation built around the NVIDIA GB10 Grace Blackwell Superchip. It is designed for AI developers, researchers, and data scientists who need a portable yet powerful system to prototype, fine-tune, and run inference on large AI models locally before scaling to data centers or cloud environments.
Key Hardware Specifications:
- Processor: 20-core Arm CPU (10 Cortex-X925 + 10 Cortex-A725 cores) integrated in the GB10 chip
- GPU: NVIDIA Blackwell architecture GPU with 5th generation Tensor Cores and 4th generation RT Cores
- Memory: 128 GB LPDDR5x unified system memory shared between CPU and GPU, with 273 GB/s bandwidth
- Storage: 1 or 4 TB NVMe M.2 SSD with self-encryption
- Networking: 10 GbE Ethernet port, ConnectX-7 Smart NIC supporting 200 GbE RDMA clustering for scale-out configurations
- Connectivity: 4x USB4 Type-C ports (up to 40 Gb/s), 1x HDMI 2.1a output, Bluetooth 5.3, WiFi 7
- Power Consumption: Around 170W, making it highly energy efficient
- Dimensions & Weight: 150 mm x 150 mm x 50.5 mm, weighing about 1.2 kg
Performance:
- Delivers up to 1,000 AI TOPS (tera operations per second) using FP4 precision
- Supports AI models with up to 200 billion parameters locally
- Can be clustered with another DGX Spark system for handling models up to 405 billion parameters
Software:
- Runs NVIDIA DGX OS based on Ubuntu Linux, optimized for AI workloads
- Comes pre-installed with the NVIDIA AI software stack including frameworks like TensorFlow, PyTorch, MATLAB, and NVIDIA’s AI tools such as Holoscan, Riva, and Metropolis
- Supports NVIDIA AI Enterprise suite for enterprise-grade AI development and deployment
Use Cases:
- Ideal for AI prototyping, fine-tuning, and inference at the edge or on a developer’s desk
- Suitable for AI research, data science, and development teams needing local AI compute power without relying on large data centers
- Dell and other OEMs are adopting DGX Spark for edge computing applications like video surveillance and operational technology environments
NVIDIA DGX Station
Overview:
The NVIDIA DGX Station is a high-performance AI workstation designed to bring data center-class AI training and inference capabilities to the desktop. It is built for data science teams and researchers who require powerful multi-GPU compute in a workstation form factor without the need for a full server rack.
Key Hardware Specifications:
- Processor: 20-core Intel Xeon E5-2698 v4 CPU at 2.2 GHz
- GPUs: Four NVIDIA Tesla V100 GPUs interconnected via NVIDIA NVLink for high-bandwidth GPU-to-GPU communication
- GPU Memory: 128 GB total system GPU memory
- Tensor Cores: 2,560 Tensor Cores across GPUs
- CUDA Cores: 20,480 CUDA cores
- System Memory: 256 GB DDR4 RDIMM
- Storage: Three 1.92 TB SSDs configured in RAID 0 for fast local storage, plus an additional 1.92 TB SSD for OS
- Networking: Dual 10 GbE (10GBASE-T) Ethernet ports
- Display: Three DisplayPort connectors supporting up to 4K resolution displays
- Cooling: Advanced water-cooling system for GPUs
- Power Consumption: Up to 1,500 W, powered by a 1,600 W PSU
- Dimensions & Weight: 518 mm x 256 mm x 639 mm, weighing about 40 kg (88 lbs)
Performance:
- Provides up to 500 TFLOPS of mixed-precision AI performance
- Designed for training large neural networks and multi-GPU workloads with high throughput and low latency
Software:
- Runs Ubuntu Desktop Linux OS with NVIDIA CUDA Toolkit and GPU drivers optimized for AI workloads
- Supports popular AI frameworks and NVIDIA’s AI software ecosystem for seamless development and deployment
Use Cases:
- Suitable for data science teams needing powerful AI training on-premises
- Can serve as a personal AI supercomputer for advanced AI model development and experimentation
- Provides a quieter, more accessible alternative to data center servers with enterprise-grade GPU performance
NVIDIA DGX Spark and DGX Station represent two tiers of AI computing designed for different scales and environments. DGX Spark offers a revolutionary compact, energy-efficient AI platform ideal for developers needing powerful local AI inference and prototyping capabilities with the flexibility to cluster for scale. In contrast, DGX Station delivers data center-class multi-GPU training performance in a desktop workstation form factor, suited for intensive AI research and development by teams requiring high throughput and multi-GPU scalability.
Together, these platforms extend NVIDIA’s AI ecosystem from edge and desktop environments to large-scale data centers, enabling a broad spectrum of AI innovation and deployment scenarios.