AI Assistant for Developers
Mainstream ecosystem support, with a one-stop workflow from model prototyping and fine-tuning to inference and AI agent deployment.
Mainstream ecosystem support, with a one-stop workflow from model prototyping and fine-tuning to inference and AI agent deployment.
Ensuring data privacy and accelerating productivity, applicable to scenarios such as predictive maintenance, medical image analysis, and financial forecasting.
Effortless AI teaching platform setup, delivering a "personal supercomputing" research experience with agile teaching and training.
End-to-end development of edge AI applications on the desktop, with real-time responsiveness and edge computing capability.
*FP4 precision (using the sparsity feature).
Two systems connected with ConnectX-7
*Using the sparsity feature.
*Requires an additional license.
*Requires a fee.
| Parameter | Description |
|---|---|
| Model | FusionXpark GB10 |
| Architecture | NVIDIA Grace Blackwell |
| GPU | Blackwell Architecture |
| CPU | 20-core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm |
| CUDA Cores | 6144 |
| Tensor Cores | Fifth Generation |
| Supported Data Formats | TF32, FP16, BF16, INT8, FP8, FP6, FP4 |
| RT Cores | Fourth Generation |
| Tensor Performance | 1 PFLOP |
| System Memory | 128 GB LPDDR5x, unified system memory |
| Video Memory Interface | Bandwidth | 256-bit | 273 GB/s |
| Storage | Self-encrypting 1 TB, 2 TB or 4 TB NVMe M.2 |
| USB | 4 USB4 Type-C |
| Ethernet | 1 RJ-45, 10 GE |
| NIC | ConnectX-7 Smart NIC |
| Wi-Fi | Wi-Fi 7 |
| Bluetooth | BT 5.4 |
| Audio Output | HDMI multi-channel audio output |
| Video Interface | 1x HDMI 2.1a |
| NVENC | NVDEC | 1x | 1x |
| Weight | 1.2 kg |
| Dimensions | 150 mm (L) x 150 mm (W) x 50.5 mm (H) |
| Operating System | Pre-installed NVIDIA DGX OS |
| AI Software Stack | Supports NVIDIA AI Enterprise |
| Custom Signature | Supports custom signature |