Coeus-3901T-64G-512G
The Coeus-3901T-64G-512G is an industrial AI Box PC built on the NVIDIA® Jetson AGX Orin™ 64GB module, pairing a 12-core Arm® Cortex-A78AE CPU with a 2,048-core NVIDIA Ampere™ GPU, dual NVDLA v2.0 deep-learning accelerators, a PVA v2.0 vision accelerator and 64 GB of LPDDR5 memory at 204.8 GB/s — delivering 275 TOPS of AI inference performance at the network edge. A pre-installed 512 GB NVMe SSD alongside 64 GB eMMC storage provides the capacity to maintain large inference model libraries and local data archives without additional hardware configuration.
Designed for engineers deploying AI-intensive workloads in industrial automation, autonomous mobile robotics and intelligent transport, the Coeus-3901T-64G-512G offers a connectivity profile matched to the demands of real-world systems integration: three 2.5 GbE LAN ports (Intel I226-IT), dual CAN DB9, dual RS-232/422/485 COM ports, 8 DI / 8 DO GPIO and MIPI CSI-2 camera interfaces. Optional NexRTOS firmware adds real-time EtherCAT master capability, and NVIDIA Isaac ROS (ROS 2 Humble) compatibility positions the platform as a compute node for next-generation autonomous mobile robot (AMR) fleets.
Key Features
- 275 TOPS — NVIDIA Jetson AGX Orin 64GB at Full Performance: 12-core Arm Cortex-A78AE CPU, 2,048-core Ampere GPU with 2× NVDLA v2.0 and 1× PVA v2.0 vision accelerator; 64 GB LPDDR5 at 204.8 GB/s sustains large transformer models, multi-modal perception networks and concurrent inference pipelines that exceed the capacity of lower-tier Jetson modules
- Triple 2.5 GbE LAN (Intel I226-IT): Three independent 2.5 Gigabit Ethernet ports — an unusually dense LAN configuration for an edge Box PC — allow simultaneous connection to operational technology networks, IT management networks and high-speed camera or sensor uplinks without a managed switch or additional NIC, simplifying network architecture at the deployment site
- EtherCAT Real-Time Control via NexRTOS + NexECM: Optional NexRTOS firmware and NexECM middleware transform the Coeus-3901T into a real-time EtherCAT master, enabling direct servo drive, I/O module and motion controller integration over standard Ethernet cabling — eliminating a dedicated motion controller from the control cabinet in AI-driven machine automation applications
- NVIDIA Isaac ROS Ready for AMR and Robotics: Full compatibility with NVIDIA Isaac ROS (ROS 2 Humble) and the NVIDIA Isaac ROS perception pipeline provides autonomous mobile robot developers with a validated, deployment-grade compute node for SLAM, object detection, path planning and fleet management workloads, with access to NVIDIA's hardware-accelerated ROS package library
- Rich Industrial I/O — CAN, Serial, GPIO and MIPI CSI-2: Dual CAN bus (DB9), dual RS-232/422/485 COM ports, 8 DI / 8 DO GPIO, 4× USB 3.2, 1× USB-C OTG and 2× MIPI CSI-2 camera interfaces cover the full range of industrial fieldbus, PLC communication, machine vision camera and discrete sensor integration requirements in a single unit
- 512 GB NVMe SSD + 64 GB eMMC Dual Storage: PCIe M.2 M-Key NVMe SSD provides fast read/write throughput for model loading, training data logging and inference result storage; 64 GB eMMC serves as the primary OS boot device, keeping the NVMe free for application workloads; further expansion via M.2 B-Key (5G/4G/Wi-Fi) and M.2 E-Key (Wi-Fi/PCIe) sockets
Coeus-3901T Series Variants
The Coeus-3901T is available in four configurations combining two Jetson AGX Orin module sizes with optional pre-installed NVMe SSD. All variants share the same chassis, I/O set and software compatibility.
| Model | Memory | GPU Cores | AI Performance | NVMe SSD |
|---|---|---|---|---|
| Coeus-3901T-32G | 32 GB LPDDR5 | 1,792-core Ampere | 200 TOPS | eMMC only |
| Coeus-3901T-32G-512G | 32 GB LPDDR5 | 1,792-core Ampere | 200 TOPS | 512 GB NVMe |
| Coeus-3901T-64G | 64 GB LPDDR5 | 2,048-core Ampere | 275 TOPS | eMMC only |
| Coeus-3901T-64G-512G | 64 GB LPDDR5 | 2,048-core Ampere | 275 TOPS | 512 GB NVMe |
The 64G-512G variant is recommended for workloads requiring large model storage alongside maximum inference throughput. Contact IPC2U for availability and lead times.
Application Use Cases
Autonomous Mobile Robots and AMR Fleet Management
The Coeus-3901T-64G-512G is a validated compute node for NVIDIA Isaac ROS (ROS 2 Humble), giving AMR developers access to hardware-accelerated SLAM, obstacle avoidance, object detection and fleet coordination pipelines running at 275 TOPS. The three 2.5 GbE ports handle separate network segments for robot-to-infrastructure communication, camera data ingestion and fleet management telemetry simultaneously, whilst the CAN bus interface connects directly to motor controllers and safety systems using standard automotive cabling.
Real-Time Industrial Automation with EtherCAT
When equipped with optional NexRTOS + NexECM firmware, the Coeus-3901T-64G-512G operates as a real-time EtherCAT master — enabling direct servo motor control, distributed I/O and precision synchronisation over standard Ethernet infrastructure. This allows machine builders to consolidate the AI inference engine and the motion controller into a single certified platform, reducing cabinet footprint, BOM cost and the integration complexity of separate motion and vision compute nodes on the same production machine.
Multi-Sensor Fusion and 3D Perception Systems
The combination of 275 TOPS compute, 204.8 GB/s memory bandwidth, dual NVDLA v2.0 accelerators and MIPI CSI-2 camera inputs addresses the data pipeline demands of multi-sensor AI systems fusing camera, LiDAR, radar and IMU streams. The 64 GB unified memory pool holds multiple large perception models simultaneously, eliminating the model-swap latency that limits smaller platforms in autonomous navigation and spatial reasoning applications requiring sub-100 ms decision cycles.
Intelligent Transport and Vehicle AI
Dual CAN DB9 interfaces, a DC 12–30 V wide-range input and MIPI CSI-2 camera connectivity position the Coeus-3901T-64G-512G for in-vehicle AI applications including fleet telematics, driver monitoring, vehicle-side inference for V2X communications and onboard predictive maintenance. The M.2 B-Key slot accommodates 5G modules for low-latency uplink, and the triple LAN configuration supports separate networks for camera ingestion, vehicle CAN gateway and fleet management connectivity without a router.
AI-Driven Smart Factory and Machine Vision
For factory automation, the Coeus-3901T-64G-512G runs multiple simultaneous inference models — surface defect detection, dimensional measurement and anomaly classification — at 275 TOPS while communicating with PLCs via RS-232/422/485 and issuing control signals through the 8 DI / 8 DO GPIO port. The 512 GB NVMe SSD supports local AI training data accumulation for continual learning deployments, and TensorFlow, Caffe, PaddlePaddle and TensorRT framework support covers the full range of industrial AI model formats in active use.
Why Choose the Coeus-3901T-64G-512G
275 TOPS with EtherCAT and Isaac ROS in a Single Platform
Most 275 TOPS Jetson AGX Orin platforms are positioned either as AI inference nodes or as industrial controllers, requiring system architects to deploy two separate units for combined AI and real-time motion control applications. The Coeus-3901T-64G-512G addresses this directly: the optional NexRTOS + NexECM stack adds a certified EtherCAT master to the same hardware that runs Isaac ROS, DeepStream and TensorRT workloads at full TOPS. For robotics and machine automation engineers, this consolidation reduces cabinet space, power consumption and inter-system integration risk in a single procurement decision.
Three 2.5 GbE Ports — Network Architecture Built In
The Intel I226-IT triple 2.5 GbE configuration is a deliberate engineering choice for applications where IT/OT network separation and high-bandwidth data ingestion must coexist at the edge. One port connects to the OT control network, one to the IT management and telemetry network, and one to a high-speed camera or sensor subnet — a topology that typically requires an unmanaged switch alongside a single-port AI box. The Coeus-3901T-64G-512G eliminates that extra switch, its power supply and its failure mode from the deployment BOM.
Full AI Software Stack Validated Out of the Box
Support for TensorFlow, Caffe, PaddlePaddle, TensorRT, DeepStream SDK and NVIDIA Isaac ROS means the Coeus-3901T-64G-512G arrives with a validated, production-ready software environment for the AI frameworks most commonly used in industrial and robotics programmes. Engineers do not need to port or revalidate existing model pipelines — the platform accepts standard ONNX, TensorRT engine and ROS 2 packages directly, shortening the integration phase from weeks to days for programmes that are already running these frameworks on development hardware.
Frequently Asked Questions
What processor and AI performance does the Coeus-3901T-64G-512G deliver?
The Coeus-3901T-64G-512G is powered by the NVIDIA Jetson AGX Orin 64GB module, featuring a 12-core Arm Cortex-A78AE CPU, a 2,048-core NVIDIA Ampere GPU, 2× NVDLA v2.0 deep-learning accelerators and 1× PVA v2.0 vision accelerator. Total AI inference performance is 275 TOPS with 64 GB of LPDDR5 at 204.8 GB/s memory bandwidth.
Does the Coeus-3901T support EtherCAT real-time control?
Yes. With the optional NexRTOS real-time operating system and NexECM EtherCAT master middleware from NEXCOM, the Coeus-3901T-64G-512G operates as a real-time EtherCAT master for direct servo drive, distributed I/O and precision motion control integration over standard Ethernet cabling.
Is the Coeus-3901T-64G-512G compatible with NVIDIA Isaac ROS?
Yes. The Coeus-3901T-64G-512G supports NVIDIA Isaac ROS (ROS 2 Humble), providing access to NVIDIA's hardware-accelerated ROS 2 packages for SLAM, perception, path planning and AMR fleet management on the Jetson AGX Orin 64GB platform.
What is the difference between the 32GB and 64GB variants?
The 64GB variant uses the Jetson AGX Orin 64GB module with a 12-core CPU, 2,048-core GPU and 275 TOPS. The 32GB variant uses the Jetson AGX Orin 32GB module with an 8-core CPU, 1,792-core GPU and 200 TOPS. Both share the same chassis, I/O and software support. The 64GB variant is recommended for large model inference, multi-sensor fusion and AMR applications requiring maximum compute and memory headroom.
What are the power and mounting options for the Coeus-3901T-64G-512G?
The Coeus-3901T-64G-512G accepts DC 12–30 V input, covering 12 V and 24 V industrial and vehicle power supplies. It supports wall mount and DIN rail installation. The unit measures 96.8 × 135 × 183 mm (226 mm with bracket) and weighs 1.95 kg net, with a DC 12 V dual ball-bearing fan for active thermal management.
The Coeus-3901T-64G-512G is an industrial AI Box PC built on the NVIDIA® Jetson AGX Orin™ 64GB module, pairing a 12-core Arm® Cortex-A78AE CPU with a 2,048-core NVIDIA Ampere™ GPU, dual NVDLA v2.0 deep-learning accelerators, a PVA v2.0 vision accelerator and 64 GB of LPDDR5 memory at 204.8 GB/s — delivering 275 TOPS of AI inference performance at the network edge. A pre-installed 512 GB NVMe SSD alongside 64 GB eMMC storage provides the capacity to maintain large inference model libraries and local data archives without additional hardware configuration.
Designed for engineers deploying AI-intensive workloads in industrial automation, autonomous mobile robotics and intelligent transport, the Coeus-3901T-64G-512G offers a connectivity profile matched to the demands of real-world systems integration: three 2.5 GbE LAN ports (Intel I226-IT), dual CAN DB9, dual RS-232/422/485 COM ports, 8 DI / 8 DO GPIO and MIPI CSI-2 camera interfaces. Optional NexRTOS firmware adds real-time EtherCAT master capability, and NVIDIA Isaac ROS (ROS 2 Humble) compatibility positions the platform as a compute node for next-generation autonomous mobile robot (AMR) fleets.
Key Features
- 275 TOPS — NVIDIA Jetson AGX Orin 64GB at Full Performance: 12-core Arm Cortex-A78AE CPU, 2,048-core Ampere GPU with 2× NVDLA v2.0 and 1× PVA v2.0 vision accelerator; 64 GB LPDDR5 at 204.8 GB/s sustains large transformer models, multi-modal perception networks and concurrent inference pipelines that exceed the capacity of lower-tier Jetson modules
- Triple 2.5 GbE LAN (Intel I226-IT): Three independent 2.5 Gigabit Ethernet ports — an unusually dense LAN configuration for an edge Box PC — allow simultaneous connection to operational technology networks, IT management networks and high-speed camera or sensor uplinks without a managed switch or additional NIC, simplifying network architecture at the deployment site
- EtherCAT Real-Time Control via NexRTOS + NexECM: Optional NexRTOS firmware and NexECM middleware transform the Coeus-3901T into a real-time EtherCAT master, enabling direct servo drive, I/O module and motion controller integration over standard Ethernet cabling — eliminating a dedicated motion controller from the control cabinet in AI-driven machine automation applications
- NVIDIA Isaac ROS Ready for AMR and Robotics: Full compatibility with NVIDIA Isaac ROS (ROS 2 Humble) and the NVIDIA Isaac ROS perception pipeline provides autonomous mobile robot developers with a validated, deployment-grade compute node for SLAM, object detection, path planning and fleet management workloads, with access to NVIDIA's hardware-accelerated ROS package library
- Rich Industrial I/O — CAN, Serial, GPIO and MIPI CSI-2: Dual CAN bus (DB9), dual RS-232/422/485 COM ports, 8 DI / 8 DO GPIO, 4× USB 3.2, 1× USB-C OTG and 2× MIPI CSI-2 camera interfaces cover the full range of industrial fieldbus, PLC communication, machine vision camera and discrete sensor integration requirements in a single unit
- 512 GB NVMe SSD + 64 GB eMMC Dual Storage: PCIe M.2 M-Key NVMe SSD provides fast read/write throughput for model loading, training data logging and inference result storage; 64 GB eMMC serves as the primary OS boot device, keeping the NVMe free for application workloads; further expansion via M.2 B-Key (5G/4G/Wi-Fi) and M.2 E-Key (Wi-Fi/PCIe) sockets
Coeus-3901T Series Variants
The Coeus-3901T is available in four configurations combining two Jetson AGX Orin module sizes with optional pre-installed NVMe SSD. All variants share the same chassis, I/O set and software compatibility.
| Model | Memory | GPU Cores | AI Performance | NVMe SSD |
|---|---|---|---|---|
| Coeus-3901T-32G | 32 GB LPDDR5 | 1,792-core Ampere | 200 TOPS | eMMC only |
| Coeus-3901T-32G-512G | 32 GB LPDDR5 | 1,792-core Ampere | 200 TOPS | 512 GB NVMe |
| Coeus-3901T-64G | 64 GB LPDDR5 | 2,048-core Ampere | 275 TOPS | eMMC only |
| Coeus-3901T-64G-512G | 64 GB LPDDR5 | 2,048-core Ampere | 275 TOPS | 512 GB NVMe |
The 64G-512G variant is recommended for workloads requiring large model storage alongside maximum inference throughput. Contact IPC2U for availability and lead times.
Application Use Cases
Autonomous Mobile Robots and AMR Fleet Management
The Coeus-3901T-64G-512G is a validated compute node for NVIDIA Isaac ROS (ROS 2 Humble), giving AMR developers access to hardware-accelerated SLAM, obstacle avoidance, object detection and fleet coordination pipelines running at 275 TOPS. The three 2.5 GbE ports handle separate network segments for robot-to-infrastructure communication, camera data ingestion and fleet management telemetry simultaneously, whilst the CAN bus interface connects directly to motor controllers and safety systems using standard automotive cabling.
Real-Time Industrial Automation with EtherCAT
When equipped with optional NexRTOS + NexECM firmware, the Coeus-3901T-64G-512G operates as a real-time EtherCAT master — enabling direct servo motor control, distributed I/O and precision synchronisation over standard Ethernet infrastructure. This allows machine builders to consolidate the AI inference engine and the motion controller into a single certified platform, reducing cabinet footprint, BOM cost and the integration complexity of separate motion and vision compute nodes on the same production machine.
Multi-Sensor Fusion and 3D Perception Systems
The combination of 275 TOPS compute, 204.8 GB/s memory bandwidth, dual NVDLA v2.0 accelerators and MIPI CSI-2 camera inputs addresses the data pipeline demands of multi-sensor AI systems fusing camera, LiDAR, radar and IMU streams. The 64 GB unified memory pool holds multiple large perception models simultaneously, eliminating the model-swap latency that limits smaller platforms in autonomous navigation and spatial reasoning applications requiring sub-100 ms decision cycles.
Intelligent Transport and Vehicle AI
Dual CAN DB9 interfaces, a DC 12–30 V wide-range input and MIPI CSI-2 camera connectivity position the Coeus-3901T-64G-512G for in-vehicle AI applications including fleet telematics, driver monitoring, vehicle-side inference for V2X communications and onboard predictive maintenance. The M.2 B-Key slot accommodates 5G modules for low-latency uplink, and the triple LAN configuration supports separate networks for camera ingestion, vehicle CAN gateway and fleet management connectivity without a router.
AI-Driven Smart Factory and Machine Vision
For factory automation, the Coeus-3901T-64G-512G runs multiple simultaneous inference models — surface defect detection, dimensional measurement and anomaly classification — at 275 TOPS while communicating with PLCs via RS-232/422/485 and issuing control signals through the 8 DI / 8 DO GPIO port. The 512 GB NVMe SSD supports local AI training data accumulation for continual learning deployments, and TensorFlow, Caffe, PaddlePaddle and TensorRT framework support covers the full range of industrial AI model formats in active use.
Why Choose the Coeus-3901T-64G-512G
275 TOPS with EtherCAT and Isaac ROS in a Single Platform
Most 275 TOPS Jetson AGX Orin platforms are positioned either as AI inference nodes or as industrial controllers, requiring system architects to deploy two separate units for combined AI and real-time motion control applications. The Coeus-3901T-64G-512G addresses this directly: the optional NexRTOS + NexECM stack adds a certified EtherCAT master to the same hardware that runs Isaac ROS, DeepStream and TensorRT workloads at full TOPS. For robotics and machine automation engineers, this consolidation reduces cabinet space, power consumption and inter-system integration risk in a single procurement decision.
Three 2.5 GbE Ports — Network Architecture Built In
The Intel I226-IT triple 2.5 GbE configuration is a deliberate engineering choice for applications where IT/OT network separation and high-bandwidth data ingestion must coexist at the edge. One port connects to the OT control network, one to the IT management and telemetry network, and one to a high-speed camera or sensor subnet — a topology that typically requires an unmanaged switch alongside a single-port AI box. The Coeus-3901T-64G-512G eliminates that extra switch, its power supply and its failure mode from the deployment BOM.
Full AI Software Stack Validated Out of the Box
Support for TensorFlow, Caffe, PaddlePaddle, TensorRT, DeepStream SDK and NVIDIA Isaac ROS means the Coeus-3901T-64G-512G arrives with a validated, production-ready software environment for the AI frameworks most commonly used in industrial and robotics programmes. Engineers do not need to port or revalidate existing model pipelines — the platform accepts standard ONNX, TensorRT engine and ROS 2 packages directly, shortening the integration phase from weeks to days for programmes that are already running these frameworks on development hardware.
Frequently Asked Questions
What processor and AI performance does the Coeus-3901T-64G-512G deliver?
The Coeus-3901T-64G-512G is powered by the NVIDIA Jetson AGX Orin 64GB module, featuring a 12-core Arm Cortex-A78AE CPU, a 2,048-core NVIDIA Ampere GPU, 2× NVDLA v2.0 deep-learning accelerators and 1× PVA v2.0 vision accelerator. Total AI inference performance is 275 TOPS with 64 GB of LPDDR5 at 204.8 GB/s memory bandwidth.
Does the Coeus-3901T support EtherCAT real-time control?
Yes. With the optional NexRTOS real-time operating system and NexECM EtherCAT master middleware from NEXCOM, the Coeus-3901T-64G-512G operates as a real-time EtherCAT master for direct servo drive, distributed I/O and precision motion control integration over standard Ethernet cabling.
Is the Coeus-3901T-64G-512G compatible with NVIDIA Isaac ROS?
Yes. The Coeus-3901T-64G-512G supports NVIDIA Isaac ROS (ROS 2 Humble), providing access to NVIDIA's hardware-accelerated ROS 2 packages for SLAM, perception, path planning and AMR fleet management on the Jetson AGX Orin 64GB platform.
What is the difference between the 32GB and 64GB variants?
The 64GB variant uses the Jetson AGX Orin 64GB module with a 12-core CPU, 2,048-core GPU and 275 TOPS. The 32GB variant uses the Jetson AGX Orin 32GB module with an 8-core CPU, 1,792-core GPU and 200 TOPS. Both share the same chassis, I/O and software support. The 64GB variant is recommended for large model inference, multi-sensor fusion and AMR applications requiring maximum compute and memory headroom.
What are the power and mounting options for the Coeus-3901T-64G-512G?
The Coeus-3901T-64G-512G accepts DC 12–30 V input, covering 12 V and 24 V industrial and vehicle power supplies. It supports wall mount and DIN rail installation. The unit measures 96.8 × 135 × 183 mm (226 mm with bracket) and weighs 1.95 kg net, with a DC 12 V dual ball-bearing fan for active thermal management.
