GIGABYTE
GIGABYTE ATAGB10-9002 AI edge system 128GB LPDDR5X 1TB NVMe
- Thin Client
- 20 core Arm, 10 Cor…
3.217,99 $
0.0399651BTC
Auf Lager
GIGABYTE
GIGABYTE ATAGB10-9000 AI edge system 128GB 4TB
Compact desktop AI system with a 20-core Arm CPU, NVIDIA GB10 Blackwell GPU, 128GB LPDDR5X memory and a 4TB PCIe 5.0 NVMe SSD
- TypeThin Client
- CPU Type20-core Arm, 10 Cortex-X925 + 10 Cortex A725
- Back I/O Ports1 x USB 3.2 Gen 2x2 Type C (PD IN) 3 x USB 3.2 Gen 2x2 Type-
- Installed Memory Size128 GB
- Memory TypeLPDDR5x
- Memory FeaturesMemory Interface: 256-bit Memory Bandwidth: 273 GB/s
Capacity 4TB PCIE5.0 SSD
- 1TB PCIE4.0 SSD
- 4TB PCIE5.0 SSD
- 1 petaFLOP AI performance with 128GB coherent unified memory
- NVIDIA GB10 Grace Blackwell Superchip with 20-core Arm CPU and fifth-gen Tensor Cores
- 4TB PCIe 5.0 NVMe SSD plus 273 GB/s memory bandwidth
- NVIDIA ConnectX-7 10 GbE and Wi-Fi 7 for scalable clustering
- Compact 150 x 150 x 50.5 mm desktop form factor at 1.3 kg
Personal AI supercomputer
The GIGABYTE AI TOP ATOM ATAGB10-9000 is a compact desktop system built for local AI development. It combines a 20-core Arm processor with an NVIDIA GB10 Blackwell GPU to deliver petaflop-scale compute in a 150 mm square chassis.
Unified memory and high-speed I/O
The system ships with 128GB of LPDDR5X memory on a 256-bit interface providing 273 GB/s bandwidth, plus a 4TB PCIe 5.0 NVMe SSD with self-encryption. Rear I/O includes three USB 3.2 Gen 2x2 Type-C ports at 20Gbps, one USB-C with power delivery input, HDMI 2.1a, 10 GbE RJ-45, and an NVIDIA ConnectX-7 SmartNIC, alongside Wi-Fi 7 and Bluetooth 5.3.
Scales to 405B parameter models
Two units can be linked via NVLink-C2C and ConnectX-7 to support models up to 405 billion parameters, extending the single-system 200B limit. The 240W power adapter and NVIDIA DGX OS provide a ready-to-run stack for prototyping, fine-tuning and inference without datacentre infrastructure.
Besonderheiten
- 1 petaFLOP AI performance with 128GB coherent unified memory
- NVIDIA GB10 Grace Blackwell Superchip with 20-core Arm CPU and fifth-gen Tensor Cores
- 4TB PCIe 5.0 NVMe SSD plus 273 GB/s memory bandwidth
- NVIDIA ConnectX-7 10 GbE and Wi-Fi 7 for scalable clustering
- Compact 150 x 150 x 50.5 mm desktop form factor at 1.3 kg
Technische Daten
| Brand | Gigabyte |
|---|---|
| Series | AI TOP ATOM |
| Model | ATAGB10-9000 |
| Type | Thin Client |
| CPU Type | 20-core Arm, 10 Cortex-X925 + 10 Cortex A725 |
| Back I/O Ports | 1 x USB 3.2 Gen 2x2 Type C (PD IN) 3 x USB 3.2 Gen 2x2 Type-C (Up to 20Gbps) 1 x NVIDIA ConnectX-7 1 x RJ-45 connector, 10 GbE 1 x HDMI 2.1a |
| Installed Memory Size | 128 GB |
| Memory Type | LPDDR5x |
| Memory Features | Memory Interface: 256-bit Memory Bandwidth: 273 GB/s |
| SSD (Installed) | 4TB NVME.M.2 PCIE5.0 with self-encryption |
| GPU Type | NVIDIA GB10 Blackwell Architecture |
| LAN Ports | 1 x 10 GbE LAN Port |
| WiFi Generation | Wi-Fi 7 |
| Bluetooth | Bluetooth 5.3 |
| Operating System | NVIDIA DGX OS |
| Power Characteristics | 240W ADP |
| Dimensions | 150 x 150 x 50.5 mm |
| Weight | 1.3 kg |
| Versandgewicht | 2.47 kg |
| Paketgröße | 216 × 178 × 117 mm |
Andere Versionen dieses Modells
1 weitere aus dieser Serie
Gleiche Familie, unterschiedlicher Speicher, Kühler oder Takt — vergleichen Sie, bevor Sie sich entscheiden.
Fragen zu diesem Artikel
What power supply is required for this system?
The unit ships with a 240W ADP adapter and requires no additional power supply.
Can two of these systems be linked for larger models?
Yes, NVIDIA NVLink-C2C and the ConnectX-7 SmartNIC allow two units to scale to models with up to 405 billion parameters.
What operating system comes pre-installed?
NVIDIA DGX OS is installed, providing the full NVIDIA AI software stack for development and inference.
Ebenfalls in dieser Kategorie
Kunden haben diese ebenfalls verglichen
Gleiche Abteilung, gleiche Kasse — acht Kryptowährungen und ein für 30 Minuten fixierter Kurs.
LG Electronics
LG Electronics CQ600N-6N AI edge system, 4GB DDR, 16GB eMMC
272,99 $314,99 $−13%
0.00339034BTC
Auf Lager
