Menu
Your Cart
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
Label
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7
GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7

GIGABYTE AI TOP ATOM ATAGB10-9000 – NVIDIA GB10 Grace Blackwell AI Supercomputer, 128GB Unified LPDDR5X, 4TB PCIe 5.0 NVMe, 1 PetaFLOP FP4, Wi-Fi 7

34,999.00
VAT INCLUSIVE

Customers also bought

GIGABYTE / AI TOP SERIESATAGB10-9000 · PERSONAL AI SUPERCOMPUTER

AI SUPERCOMPUTING. AT YOUR DESK.

An ultra-compact personal AI supercomputer powered by NVIDIA GB10 Grace Blackwell, 128GB LPDDR5X unified system memory, and 4TB PCIe 5.0 NVMe storage. Designed for local model development, inference and data-science workflows.

EXPLORE TECHNICAL SPECS ↗
NVIDIA GB10 · GRACE BLACKWELLGIGABYTE AI TOP ATOMAI TOP SERIES / 4TB PCIe Gen5
128GBLPDDR5X Unified Memory
1 PFLOPUp to FP4 AI Compute
4TBNVMe PCIe 5.0 SSD
20 COREArm Processor
01 / AI ARCHITECTURE

ENGINEERED FOR AI WORKLOADS

GB10 · BLACKWELL

Accelerated AI Architecture

NVIDIA Grace Blackwell integrates a 20-core Arm CPU and Blackwell GPU architecture for AI-centric development.

UNIFIED MEMORY

Large Unified Memory

128GB LPDDR5X coherent unified system memory with 273GB/s bandwidth; this is not a standalone GPU with 128GB of dedicated VRAM.

LOCAL DEVELOPMENT

Build and Run Locally

NVIDIA DGX OS and AI TOP Utility support local AI development, inference and RAG workflows.

02 / AI APPLICATIONS

QUAD AI PERFORMANCE MATRIX

A grounded overview of AI workloads supported by GIGABYTE AI TOP ATOM — without invented FPS or tokens-per-second benchmarks.

INFERENCE

Local AI Inference

Run AI models locally without continuously relying on cloud inference. The platform supports models up to 200B parameters subject to model, quantization and software constraints.

FINE-TUNING

Model Fine-tuning

Adapt pretrained AI models for organization-specific data and workflows; memory and training time vary by model and method.

AI TOP UTILITY

Retrieval-Augmented Generation

AI TOP Utility provides model download, inference, retrieval-augmented generation (RAG), and machine-learning workflows.

DATA SCIENCE

Data Science

For local data exploration, modeling, predictive analysis, and experimentation by developers and researchers.

EDGE AI

Edge AI Applications

Compact 1-litre form factor for edge AI prototyping, computer vision and local deployment experiments.

CONNECTX-7

Two-System Scaling

GIGABYTE advertises support for up to 405B-parameter models with two compatible units connected via ConnectX-7; a QSFP cable is sold separately and software configuration is required.

Workload descriptions reflect GIGABYTE's published capabilities, not verified numerical benchmarks. The advertised 1 PetaFLOP is FP4 compute, not FP16/FP32 performance or token-generation speed. The 200B/405B figures depend on model format, system configuration, interconnect and software.

03 / OFFICIAL DATASHEET

FULL OFFICIAL TECHNICAL SPECIFICATIONS

Compute & AI

Model numberATAGB10-9000
SuperchipNVIDIA GB10 Grace Blackwell Superchip
CPU20-core Arm · 10 × Cortex-X925 + 10 × Cortex-A725
GPU architectureNVIDIA Blackwell Architecture
AI tensor performanceUp to 1 PetaFLOP · FP4
Chip interconnectNVIDIA NVLink-C2C
Unified system memory128GB LPDDR5X
Memory interface256-bit
Memory bandwidth273 GB/s

Storage & Connectivity

SSD4TB M.2 NVMe PCIe 5.0 SSD
Ethernet10GbE · RJ-45
WirelessWi-Fi 7 · Bluetooth 5.4
Interconnect NICNVIDIA ConnectX-7
High-speed interconnect2 × QSFP · 200 Gbps (ConnectX-7)
Power input1 × USB-C · USB 3.2 Gen 2x2 · PD IN
USB data ports3 × USB-C · USB 3.2 Gen 2x2 · up to 20Gbps
Display output1 × HDMI 2.1a
Security slot1 × Kensington Nano slot

Physical & Software

Operating systemNVIDIA DGX OS / Ubuntu Linux
External adapter240W adapter
Chassis volume1 Litre
Dimensions (H × W × D)50.5 × 150 × 150 mm
Weight1.2 kg
Included in the boxAI TOP ATOM + 240W adapter + power cord (regional)

OFFICIAL SOURCES & PRODUCT NOTES

Specifications apply to ATAGB10-9000 with 4TB PCIe 5.0 SSD. Do not confuse with 9001 or 9002 storage variants. The platform uses an Arm/Linux environment; check your application compatibility before purchasing. Verify imagery, warranty and package details with the seller.

Official GIGABYTE Saudi Arabia specifications·Product Features
GIGABYTE AI TOP ATOM | ATAGB10-9000 | Personal AI Supercomputer
No reviews yetBe the first to share your opinion about this product.
تواصل معنا