NVIDIA DGX Spark Desktop AI Computer — Grace Blackwell 1 PFLOP Superchip, 128 GB Unified Memory, 4 TB NVMe, Wi-Fi 7, AI-Ready Workstation
Key Features
- NVIDIA GB10 GPU
- ARM Cortex-X925 + ARM Cortex-A725 Processor
- 128GB LPDDR5X Unified RAM
- 4TB NVMe M2 SSD
- 10GbE, WiFi 7 (802.11be), Bluetooth 5.4
- NVIDIA DGX OS
🧠 Overview — What is DGX Spark
The NVIDIA DGX Spark is a desktop-class AI supercomputer built around the NVIDIA Grace Blackwell architecture (GB10 Superchip), combining high-performance ARM CPU + Blackwell GPU + unified memory + high-speed storage & networking. It’s designed for AI / ML developers, researchers, data scientists who want powerful local AI compute — able to run large models, inference, and fine-tuning — without relying on cloud datacenters.
This makes DGX Spark a unique product: a “AI PC” rather than a conventional gaming/desktop PC.
📦 Specifications — DGX Spark Desktop AI Computer
| Specification | Details |
|---|---|
| Product | NVIDIA DGX Spark Desktop AI Computer |
| Architecture | NVIDIA Grace Blackwell (GB10 Superchip) |
| CPU | 20-core ARM CPU: 10× Cortex-X925 + 10× Cortex-A725 |
| GPU / AI Engine | Blackwell GPU + 5th-gen Tensor Cores + 4th-gen RT Cores |
| Unified System Memory | 128 GB LPDDR5x, 256-bit, 273 GB/s bandwidth |
| Storage | NVMe M.2 SSD — up to 4 TB (self-encrypted) |
| AI Performance | Up to 1 petaFLOP (FP4 precision, sparse) with GB10 chip |
| Model Capacity (Local) | Handles AI models up to ~200 billion parameters |
| Networking / Connectivity | 10 GbE (RJ-45), ConnectX-7 Smart NIC, Wi-Fi 7, Bluetooth 5.4 |
| I/O Ports | 4 × USB-C, 1 × HDMI 2.1a, RJ-45 10 GbE, QSFP (via NIC) |
| OS & Software Stack | Pre-installed with NVIDIA DGX OS — full AI & ML software stack (CUDA, cuDNN etc.) |
| Form Factor / Dimensions | Compact desktop form: 150 mm × 150 mm × 50.5 mm, ~1.2 kg |
| Power Supply | External 240 W PSU (GB10 TDP 140 W + overhead) |
| Video / Media | NVENC / NVDEC (1 × each), suitable for video encoding/decoding tasks |
| Use Cases | AI model inference & fine-tuning, LLMs, data science, research, local AI development, edge-AI deployment |
⚙️ Why DGX Spark Is a Game-Changer
-
💡 Unified CPU-GPU memory: 128 GB LPDDR5x unified memory avoids data transfer overhead, letting large models run smoothly.
-
🚀 1 PFLOP AI compute: With 5th-gen tensor cores & optimized Blackwell GPU, Spark delivers powerful AI performance previously limited to datacenters.
-
🔒 Local AI with privacy & low latency: Run large models, fine-tune, experiment — all on-premise.
-
🧰 Full software stack out-of-the-box: Ships with NVIDIA DGX OS + CUDA / ML frameworks, ready for AI workloads.
-
🖥️ Desktop form-factor: Compact size (150 × 150 × 50.5 mm) — fits on a workstation desk.
-
🔌 Enterprise-grade connectivity: 10 GbE, ConnectX-7 NIC, Wi-Fi 7 — ideal for data heavy & multi-user setups
Reviews
There are no reviews yet.