Coeus-4002T
The NEXCOM Coeus-4002T is an edge AI system built on the NVIDIA Jetson Thor T5000 — the first Jetson module on the Blackwell GPU architecture — delivering 2070 FP4 TFLOPS from a 2560-core GPU and 14-core ARM CPU, with 128 GB LPDDR5 memory. That is roughly 8× the AI throughput of Jetson AGX Orin, sized for foundation-model inference, humanoid robotics and physical AI at the edge.
Networking spans five 2.5 GbE ports plus a QSFP28 port supporting up to 100 GbE — bandwidth for converging camera, LiDAR and tactile sensor streams on one controller. A 1 TB NVMe SSD stores model weights locally; the wall/DIN-rail chassis accepts 24–30 V DC and operates at −10 °C to +50 °C with airflow.
Key Features
- Jetson Thor T5000 — 2070 FP4 TFLOPS: 2560-core Blackwell GPU with 4th-gen Tensor Cores and Transformer Engine, plus 14-core ARM CPU — built for transformer-based vision, multimodal and language models.
- 128 GB LPDDR5: Accommodates foundation-model activation memory and multi-camera pipelines without bandwidth starvation of the GPU.
- 5× 2.5 GbE + QSFP28 (25/100 GbE): Per-sensor traffic isolation on the copper ports; optical high-speed backhaul to training clusters or multi-robot interconnects.
- 1 TB NVMe + 2× PCIe x4 slots: Local model and dataset storage with expansion for accelerators or additional NVMe.
- Developer I/O: 3× USB 3.0, USB-C OTG (device mode for commissioning), HDMI operator display.
- Wireless-ready: M.2 Key-B (4G/5G + SIM) for OTA model sync and fleet monitoring; M.2 Key-E (Wi-Fi 6/6E + BT).
- 24–30 V DC, −10 °C to +50 °C, wall/DIN-rail mount, CE/FCC.
Coeus-4002T vs MARS400-A10-04-C
| Attribute | Coeus-4002T | MARS400-A10-04-C |
|---|---|---|
| Jetson module | Thor T5000 (Blackwell) | AGX Orin (Ampere) |
| AI throughput | 2070 FP4 TFLOPS | 275 TOPS |
| Memory | 128 GB LPDDR5 | 64 GB LPDDR5 |
| High-speed networking | QSFP28 up to 100 GbE + 5× 2.5GbE | 5× GbE |
| EtherCAT master | No | Yes — NexECM, 250 μs |
| Best for | Foundation models, humanoid robotics, physical AI | EtherCAT motion + AI vision |
Application Use Cases
Humanoid Robot and Physical AI Controller
Full-body motion planning, scene understanding and multimodal instruction-following must run at control-cycle rates within an on-board power budget — the Thor T5000's 2070 TFLOPS and 128 GB memory make the Coeus-4002T one of the few edge platforms capable of it.
Multi-Robot Orchestration Node
The QSFP28 port links a robot-fleet switch at up to 100 GbE while the five 2.5 GbE ports carry per-robot channels; the Blackwell GPU runs fleet-level task optimisation and the 1 TB SSD archives telemetry.
Edge LLM Inference
Runs 7B–70B parameter language models locally for operator assistance, maintenance recommendation and document query — avoiding cloud latency, egress cost and data sovereignty risk for proprietary process data.
Advanced Autonomous Platform AI
Processes multiple high-resolution camera and LiDAR streams for Level 3+ industrial vehicles and inspection robots; QSFP28 uploads deployment data to remote training servers for continuous learning.
Why Choose the Coeus-4002T
Blackwell at the edge, first. Among the first commercial systems on Jetson Thor T5000, with the Transformer Engine gains that matter most on the transformer-based models dominating physical AI — and a longer time-to-obsolescence than Ampere-based alternatives.
100 GbE headroom. As sensor counts and resolutions climb, the QSFP28 port ensures the network interface — not the GPU — is never the bottleneck over the system's deployment life.
Production system, not a devkit. NEXCOM's industrial design (24–30 V DC, DIN-rail mounting, CE/FCC) plus IPC2U's European distribution and support turn the Thor module into a deployable product.
Frequently Asked Questions
What AI performance does the Coeus-4002T deliver?
2070 FP4 TFLOPS from the NVIDIA Jetson Thor T5000's 2560-core Blackwell GPU — approximately 8× the 275 TOPS of Jetson AGX Orin — backed by 128 GB LPDDR5.
What networking is provided?
Five 2.5 GbE RJ45 ports plus one QSFP28 port accepting 25 GbE or 100 GbE optical transceivers, with M.2 slots for 4G/5G (with SIM) and Wi-Fi 6/6E + Bluetooth.
Does the Coeus-4002T include EtherCAT motion control?
No. For combined EtherCAT servo control and AI inference, choose the NEXCOM MARS400-A10-04-C (Jetson AGX Orin with NexECM EtherCAT master). The Coeus-4002T targets maximum AI throughput.
What is the operating temperature?
−10 °C to +50 °C with adequate airflow; forced airflow is required at maximum AI load near the upper limit. Contact IPC2U for enclosure thermal guidance.
What power input does it use?
24–30 V DC, compatible with standard 24 V DIN-rail supplies; wall and DIN-rail mounting hardware is included.
The NEXCOM Coeus-4002T is an edge AI system built on the NVIDIA Jetson Thor T5000 — the first Jetson module on the Blackwell GPU architecture — delivering 2070 FP4 TFLOPS from a 2560-core GPU and 14-core ARM CPU, with 128 GB LPDDR5 memory. That is roughly 8× the AI throughput of Jetson AGX Orin, sized for foundation-model inference, humanoid robotics and physical AI at the edge.
Networking spans five 2.5 GbE ports plus a QSFP28 port supporting up to 100 GbE — bandwidth for converging camera, LiDAR and tactile sensor streams on one controller. A 1 TB NVMe SSD stores model weights locally; the wall/DIN-rail chassis accepts 24–30 V DC and operates at −10 °C to +50 °C with airflow.
Key Features
- Jetson Thor T5000 — 2070 FP4 TFLOPS: 2560-core Blackwell GPU with 4th-gen Tensor Cores and Transformer Engine, plus 14-core ARM CPU — built for transformer-based vision, multimodal and language models.
- 128 GB LPDDR5: Accommodates foundation-model activation memory and multi-camera pipelines without bandwidth starvation of the GPU.
- 5× 2.5 GbE + QSFP28 (25/100 GbE): Per-sensor traffic isolation on the copper ports; optical high-speed backhaul to training clusters or multi-robot interconnects.
- 1 TB NVMe + 2× PCIe x4 slots: Local model and dataset storage with expansion for accelerators or additional NVMe.
- Developer I/O: 3× USB 3.0, USB-C OTG (device mode for commissioning), HDMI operator display.
- Wireless-ready: M.2 Key-B (4G/5G + SIM) for OTA model sync and fleet monitoring; M.2 Key-E (Wi-Fi 6/6E + BT).
- 24–30 V DC, −10 °C to +50 °C, wall/DIN-rail mount, CE/FCC.
Coeus-4002T vs MARS400-A10-04-C
| Attribute | Coeus-4002T | MARS400-A10-04-C |
|---|---|---|
| Jetson module | Thor T5000 (Blackwell) | AGX Orin (Ampere) |
| AI throughput | 2070 FP4 TFLOPS | 275 TOPS |
| Memory | 128 GB LPDDR5 | 64 GB LPDDR5 |
| High-speed networking | QSFP28 up to 100 GbE + 5× 2.5GbE | 5× GbE |
| EtherCAT master | No | Yes — NexECM, 250 μs |
| Best for | Foundation models, humanoid robotics, physical AI | EtherCAT motion + AI vision |
Application Use Cases
Humanoid Robot and Physical AI Controller
Full-body motion planning, scene understanding and multimodal instruction-following must run at control-cycle rates within an on-board power budget — the Thor T5000's 2070 TFLOPS and 128 GB memory make the Coeus-4002T one of the few edge platforms capable of it.
Multi-Robot Orchestration Node
The QSFP28 port links a robot-fleet switch at up to 100 GbE while the five 2.5 GbE ports carry per-robot channels; the Blackwell GPU runs fleet-level task optimisation and the 1 TB SSD archives telemetry.
Edge LLM Inference
Runs 7B–70B parameter language models locally for operator assistance, maintenance recommendation and document query — avoiding cloud latency, egress cost and data sovereignty risk for proprietary process data.
Advanced Autonomous Platform AI
Processes multiple high-resolution camera and LiDAR streams for Level 3+ industrial vehicles and inspection robots; QSFP28 uploads deployment data to remote training servers for continuous learning.
Why Choose the Coeus-4002T
Blackwell at the edge, first. Among the first commercial systems on Jetson Thor T5000, with the Transformer Engine gains that matter most on the transformer-based models dominating physical AI — and a longer time-to-obsolescence than Ampere-based alternatives.
100 GbE headroom. As sensor counts and resolutions climb, the QSFP28 port ensures the network interface — not the GPU — is never the bottleneck over the system's deployment life.
Production system, not a devkit. NEXCOM's industrial design (24–30 V DC, DIN-rail mounting, CE/FCC) plus IPC2U's European distribution and support turn the Thor module into a deployable product.
Frequently Asked Questions
What AI performance does the Coeus-4002T deliver?
2070 FP4 TFLOPS from the NVIDIA Jetson Thor T5000's 2560-core Blackwell GPU — approximately 8× the 275 TOPS of Jetson AGX Orin — backed by 128 GB LPDDR5.
What networking is provided?
Five 2.5 GbE RJ45 ports plus one QSFP28 port accepting 25 GbE or 100 GbE optical transceivers, with M.2 slots for 4G/5G (with SIM) and Wi-Fi 6/6E + Bluetooth.
Does the Coeus-4002T include EtherCAT motion control?
No. For combined EtherCAT servo control and AI inference, choose the NEXCOM MARS400-A10-04-C (Jetson AGX Orin with NexECM EtherCAT master). The Coeus-4002T targets maximum AI throughput.
What is the operating temperature?
−10 °C to +50 °C with adequate airflow; forced airflow is required at maximum AI load near the upper limit. Contact IPC2U for enclosure thermal guidance.
What power input does it use?
24–30 V DC, compatible with standard 24 V DIN-rail supplies; wall and DIN-rail mounting hardware is included.
Get a Quote
