EdgeXpert is a compact AI supercomputer built on NVIDIA Grace Blackwell architecture, delivering up to 1 petaFLOP of AI performance. With 128GB unified memory and 4TB NVMe storage, it is ideal for AI development, model training, and inference workloads.
Talk to an expert
After sale services
Architecture: NVIDIA® Grace Blackwell
GPU: NVIDIA® Blackwell Architecture GPU
CPU: Arm 20-Core Processor (10× Cortex-X925 + 10× Cortex-A725)
AI Performance: Up to 1 PetaFLOP (FP4, Sparse)
Tensor Cores: 5th Generation
RT Cores: 4th Generation
Supported AI Data Formats: TF32, FP16, BF16, INT8, FP8, FP6, FP4
System Memory: 128 GB LPDDR5x Unified Memory
Usable Memory: ~100 GB for user workloads
Memory Interface: 256-bit
Memory Bandwidth: 273 GB/s
Storage: 4 TB NVMe M.2 SSD (Self-Encrypting Drive)
Networking:
NVIDIA ConnectX-7 Smart NIC
1 × 10 GbE RJ-45 LAN
QSFP 200 GbE (System-to-System Connectivity)
Wireless Connectivity: Wi-Fi 7, Bluetooth 5.4
USB Ports: 4 × USB 3.2 Type-C (up to 20 Gb/s)
Display Outputs:
1 × HDMI 2.1
4 × DisplayPort 1.4a (via USB-C)
Media Engines: NVENC ×1, NVDEC ×1
Audio: HDMI Multi-Channel Audio Output
Operating System: NVIDIA® DGX™ OS (NVIDIA AI Software Stack Included)
AI Model Capability:
Run models up to 200B parameters
Fine-tune models up to 70B parameters
Cluster up to 2 systems (supports models up to ~405B parameters)
Form Factor: Compact Desktop System
Dimensions: 151 × 151 × 52 mm (1.2 L)
Weight: 1.2 kg
Power: Standard Wall Outlet
Deployment Modes: AI Workstation / Personal AI Cloud
