NVIDIA Blackwell GPU architecture with 2,560 CUDA cores and 96 Tensor Cores for advanced AI inference
- Multi-Instance GPU (MIG) with 10 TPCs enables robust workload isolation for running multiple AI models concurrently
- Transformer Engine acceleration optimized for generative AI and large language model inference at the edge
- 14-core Arm Neoverse-V3AE CPU with 64-bit architecture for high-performance compute tasks
- 128GB LPDDR5X memory with full DRAM ECC support maintains performance without reducing usable capacity
- Functional Safety Island (FSI) on Thor SoC enables advanced safety functions for mission-critical applications
- NVIDIA ConnectX-7 SmartNIC with RDMA support for ultra-low latency sensor data ingestion
- PCIe Gen5 interface with integrated PCIe switch for high-speed peripheral connectivity
- Optional discrete GPU expansion slot supports NVIDIA RTX PRO Blackwell dGPUs
- Board Management Controller (BMC) provides remote system monitoring and management capabilities
- Programmable Vision Accelerator (PVA) v3 for efficient computer vision processing
- Dual video encoders and decoders with optical flow accelerator for real-time video analytics
- ISO 26262 and IEC 61508 functional safety compliance targeting ASIL D/SL3 certification
- Support for NVIDIA AI Enterprise software stack with optimized frameworks and containers
- Compatible with NVIDIA Isaac robotics platform, Metropolis visual AI, and Holoscan sensor processing
- MicroATX form factor enables integration into standard industrial enclosures and edge servers