Overview
The GIGABYTE AI TOP ATOM brings datacenter-class AI performance to the desk. Built around the NVIDIA GB10 Grace Blackwell Superchip, this compact personal AI supercomputer is designed for local AI development, inference, fine-tuning and experimentation with large models, without depending on cloud infrastructure.
With up to 1 PFLOP of FP4 AI performance, 128GB of unified LPDDR5x memory and fast NVMe storage, AI TOP ATOM gives developers, researchers and AI teams a strong local platform for large language models, generative AI, computer vision and private AI workflows. It ships with NVIDIA DGX OS and the NVIDIA AI software stack, so it is ready to build on from the first boot.
AI TOP ATOM is available in two configurations: 1TB (ATAGB10-9002) and 4TB (ATAGB10-9000) NVMe storage. Both share the same GB10 platform, memory and connectivity.
Key Features
- NVIDIA GB10 Grace Blackwell Superchip: Blackwell GPU and 20-core Arm CPU on one package
- Up to 1 PFLOP FP4 AI Performance: 5th generation Tensor Cores with FP4 precision support
- 128GB Unified LPDDR5x Memory: 256-bit interface, 273 GB/s, shared coherently between CPU and GPU
- Models up to 200B Parameters: Run inference on large models locally, fine-tune mid-size models on device
- 1TB or 4TB NVMe Storage: Choose the configuration that fits your datasets and model library
- 10GbE + NVIDIA ConnectX-7: High-speed networking for data pipelines and multi-node setups
- Wi-Fi 7 and Bluetooth 5.4: Flexible connectivity for any workspace
- NVIDIA DGX OS: Ubuntu-based operating system with the NVIDIA AI stack preinstalled
- Compact Desktop Form Factor: 150 x 150 x 50.5 mm, 1.2 kg
Built for Local AI Development
AI TOP ATOM is made for teams and professionals who want to prototype, fine-tune, validate and deploy AI workflows locally. From generative AI and agentic systems to computer vision and embedded AI pipelines, the platform delivers enterprise-grade AI capability in a desktop-sized system.
The unified memory architecture lets larger workloads run efficiently while minimizing data movement between compute resources. Combined with high memory bandwidth and fast storage, the system enables rapid experimentation and iteration.
Technical Specifications
| Specification | Details |
|---|---|
| Architecture | NVIDIA Grace Blackwell |
| Processor | NVIDIA GB10 Grace Blackwell Superchip |
| GPU | NVIDIA Blackwell architecture, 5th generation Tensor Cores, 4th generation RT Cores |
| CPU | 20-core Arm (10x Cortex-X925 + 10x Cortex-A725) |
| AI Performance | 1 PFLOP (FP4) |
| Memory | 128 GB LPDDR5x unified system memory |
| Memory Interface | 256-bit |
| Memory Bandwidth | 273 GB/s |
| Storage | 1 TB NVMe M.2 (ATAGB10-9002) or 4 TB NVMe M.2 (ATAGB10-9000) |
| Networking | 1x 10GbE RJ-45, NVIDIA ConnectX-7 SmartNIC |
| Wireless | Wi-Fi 7, Bluetooth 5.4 |
| Display | 1x HDMI 2.1a (multichannel audio output) |
| USB | 4x USB Type-C |
| Operating System | NVIDIA DGX OS (Ubuntu Linux) |
| Power Adapter | 240W |
| Dimensions | 150 x 150 x 50.5 mm (L x W x H) |
| Weight | 1.2 kg |
Ideal Use Cases
- Local LLM development and inference
- AI fine-tuning and model validation
- AI workstation environments for developers and data scientists
- Edge AI prototyping before deployment
- Computer vision workloads
- Data science and research projects
- Private AI environments without cloud dependency
Why Choose This Product?
AI TOP ATOM packs the same NVIDIA GB10 Grace Blackwell platform found in DGX Spark into a GIGABYTE-built desktop system, with a choice of 1TB or 4TB storage. It runs models that previously required cloud infrastructure with zero latency, full data privacy and no recurring cloud costs, and the unified 128GB memory makes it a practical everyday machine for AI teams.
Interested? Contact us for pricing, bulk orders for teams and enterprise deployment support.
