OAIBOX™ AI-RAN
with
DGX Spark™
AI-NATIVE 6G research platform in a box
WHY CHOOSE THE OAIBOX™ with DGX Spark™
GPU-accelerated open source 5G/6G platform (gNB, CN5G, UE) for AI-native 6G research on small form factor
Integrated with NVIDIA Sionna™ Ray Tracer (RT) for rapid AI/ML training and inference
Real-time GPU-accelerated digital twins
NVIDIA Sionna™ RT provides ray-tracing-based RF simulation for real-world propagation environments, including mobility and antenna radiation patterns
Support for customization to specific test scenarios using real 3D maps of target urban environments
Unified CPU/GPU architecture for efficient inline acceleration without the need of explicit memory transfers
Integrates with the OAIBOX™ Dashboard for E2E control and real-time network monitoring
Direct access to data pipelines in the 5G stack for AI/ML model and algorithm development
AI-RAN benchmarking platform for analyzing trade-offs between ML model complexity, performance, and latency constraints
Compatible with NVIDIA AI/ML tools (TensorFlow) and NVIDIA tutorials, including LDPC decoding and Neural Receiver implementations
Integrates with NVIDIA Aerial L1 with access to the CUDA open-source code
Affordable platform for prototyping AI-native 6G
TECHNICAL SPECIFICATIONS
Characteristics
OAIBOX™ with DGX Spark™
Open source 5G stack
OpenAirInterface (OAI) for both the gNB and 5G core network
Supports NVIDIA Aerial L1 in open source CUDA
3GPP release
Compliant with 3GPP Rel 16/17 and aligned with OAI roadmap
SDR card
NI USRP B206-mini-i, B200 and B210 (enabling MIMO 2x2)
UE
Quectel modem RM500Q with enclosure, one SIM card configured. OAIBOX UE
RF cable kit
RF cables and a circulator to connect the gNB to the UE, avoiding over-the-air transmission
Channel bandwidth
Up to 20 MHz, supports TDD and FDD
Max DL/UL throughput
70 Mbps (DL) and 40 Mbps (UL)
Tx power and range
10 dBm, range up to 100 m, designed for indoor operation
Frequency bands
Any 3GPP FR1 band (default bands n41, n77 and n78)
Dashboard
Web application for network control and visualisation
Digital Twin
Integrated with the NVIDIA Sionna™ RT
NVIDIA DGX Spark™ CPU
Grace Blackwell (GB10 Superchip) – Arm-based CPU with 20 cores
NVIDIA DGX Spark™ GPU
integrated Blackwell GPU with 6,144 CUDA cores, 5th-gen Tensor Cores, up to 1 PFLOP AI performance (FP4)
NVIDIA DGX Spark™ Memory
128 GB unified LPDDR5X shared memory with up to 273 GB/s bandwidth
Power supply
110 V and 220 V
OAIBOX™ Digital Twin
Virtual UE moving through the ray-traced scene
Validate & benchmark
AI-native Air Interface real-world test environment
