NVIDIA IGX T5000

• 5th-Generation Tensor Cores: Advanced AI cores with hardware support for FP4, FP8, FP16, bfloat16, TF32, and INT8 formats.
• Sparse Computing Performance: With AI performance reaching up to 2070 TFLOPS FP4 and 1035 TFLOPS FP8, complex Large Language Models (LLMs) and deep learning inference workloads can be executed at the edge within milliseconds.
• Transformer Engine: Accelerates deep learning and Transformer models directly at the hardware level, maximizing power efficiency and inference performance.
• Multi-Instance GPU (MIG): Enables the GPU to be partitioned into physically isolated instances, allowing mission-critical workloads and general-purpose applications to run securely and independently on the same module.

• 14-Core Arm® Neoverse® V3AE: A 64-bit processor complex with a maximum operating frequency of up to 2.6 GHz, providing an excellent foundation for parallel data processing, multitasking, and complex system management.
• 128 GB LPDDR5X Memory: With a 256-bit memory bus and massive memory bandwidth of up to 273 GB/s, multiple high-resolution camera streams and large datasets can be processed smoothly without bottlenecks.

• Includes an integrated Functional Safety Island (FSI) to support compliance with industrial and medical safety standards.
• With 4 Cortex-R52 safety cores operating in Dual-Core Lock-Step (DCLS) mode and a Hardware Security Module (HSM), system monitoring and fault management can be performed with high reliability without requiring an external MCU

• Advanced Camera Support: Supports up to 32 camera data streams through a 16-lane MIPI CSI-2 interface (164 Gbps with C-PHY v2.1 / 40 Gbps with D-PHY v2.1) and Virtual Channels. It also enables connectivity for up to 20 cameras via the Holoscan Sensor Bridge (HSB).
• Advanced Hardware Accelerators:
  • 2x NVDEC / 2x NVENC: Real-time hardware-accelerated video decoding and encoding for 8K30 and multiple 4K60 streams.
  • PVA v3.0 (Programmable Vision Accelerator): A dedicated vector processor that handles computer vision and sensor data processing workloads, reducing the load on the GPU.
  • Optical Flow Accelerator (OFA): Hardware acceleration for depth perception (Stereo Disparity) and motion tracking.

• PCIe Gen 5: Supports up to 8 lanes of PCIe Gen 5, enabling ultra-high-speed data transfer with NVMe storage devices and external expansion cards.
• Ultra-High-Speed Networking: Integrated Multi-Gigabit Ethernet (MGBE) controllers provide 4x 25 GbE and 1x 1 GbE network interfaces.
• Rich Industrial Connectivity: 3x USB 3.2 Gen 2 (10 Gbps), 4x USB 2.0, 4x CAN bus, 4x UART, 13x I2C, 3x SPI, and HDMI 2.1 / DisplayPort 1.4a display outputs.

• 24/7 Continuous Operation: Designed for uninterrupted industrial operations with an operating lifetime of up to 10 years.
• Integrated TTP and Heatpipe: The integrated Thermal Transfer Plate (TTP) and heatpipe system enable easy integration of external active or passive cooling solutions.
• Temperature Durability: Wide operating temperature range from -40°C to 118°C (Tj).

Use Cases

• Medical Devices & Surgical Robotics: Real-time AI-powered imaging, ultrasound/endoscopy analysis, and ultra-low-latency intervention systems.
• Autonomous Mobile Robots (AMRs) & AGVs: Multi-camera/LiDAR data processing, hardware-accelerated depth mapping, and route planning in factories and logistics centers.
• Industrial Quality Control & Automation: Product defect detection and object classification using high-speed, multi-camera streams on production lines.
• Smart Cities & Security: Edge computing analytics systems capable of simultaneously processing 8K and multiple 4K video streams.