Specifications
Manufacturer
Aetina Corporation
Series
AIB-AT78 / AIB-AT68
Module Compatibility
NVIDIA Jetson Thor T5000 (AIB-AT78) / T4000 (AIB-AT68)
GPU Architecture
NVIDIA Blackwell
GPU Cores
2560-core (T5000) / 1536-core (T4000)
Tensor Cores
96 fifth-generation (T5000) / 64 fifth-generation (T4000)
CPU
14-core Arm Neoverse V3AE (T5000) / 12-core Arm Neoverse V3AE (T4000)
AI Performance
2070 FP4 TFLOPS and 1035 FP8 TFLOPS (AIB-AT78); 1200 FP4 TFLOPS (AIB-AT68)
Memory
128GB 256-bit LPDDR5X (AIB-AT78) / 64GB 256-bit LPDDR5X (AIB-AT68)
Storage
1x M.2 M-Key 2280 with 1TB NVMe built in
High-Speed Interconnect
1x QSFP28 slot - 4x25GbE (T5000) or 3x25GbE (T4000), optimised for NVIDIA Holoscan
Networking
2x RJ45 10GbE + 1x RJ45 1GbE EtherCAT
USB
4x USB-A 3.2 Gen2 + 1x USB-C (OTG only)
Display
1x HDMI 2.0b + 1x DisplayPort 1.4a
Camera Interface
1x 120-pin MIPI supporting up to 8x GMSL2 cameras
Industrial I/O
4x isolated CAN FD (T5000), 4x isolated DI/DO at 3.3V, 2x RS232/RS485/RS422, 2x UART, 2x I2C, HD audio in/out
Expansion
M.2 M-Key 2280 (NVMe), M.2 E-Key 2230 (WiFi/BT), M.2 B-Key 3042/3052 (4G/5G), Micro SIM, out-of-band option
Power Input
DC 9V to 48V wide range
Power Envelope
Approximately 130W at full Jetson Thor performance
Operating Temperature
-25C to +70C
Dimensions
140 x 165 x 42.8 mm
Software Stack
NVIDIA JetPack 7, Isaac GR00T, Holoscan, Cosmos; custom BSP and sensor integration services
Ordering SKUs
AIB-AT78-1-A1 (T5000) and AIB-AT68-1-A1 (T4000)
Overview
The Aetina AIB-AT78/68 is an edge supercomputing platform built around the NVIDIA Jetson Thor family. The AIB-AT78 carries the Jetson T5000 module and the AIB-AT68 the T4000; both use the NVIDIA Blackwell GPU architecture and deliver 2070 FP4 TFLOPS (1035 FP8) and 1200 FP4 TFLOPS respectively. Against Jetson AGX Orin that is roughly 7.5x more AI compute at about 3.5x better energy efficiency, inside a package just 140 x 165 mm and 42.8 mm tall.
Memory capacity is the headline for generative robotics: 128GB of 256-bit LPDDR5X on the AIB-AT78 and 64GB on the AIB-AT68, enough to hold large perception and vision-language models plus their context without sharding. Aetina targets the deployments where that matters - humanoid and surgical robots, autonomous warehouse vehicles, agricultural drones and utility inspection robots - and reports the platform running full-body robot control with more than 41 degrees of freedom at low transmission latency.
Sensor integration is where the design diverges from a generic module carrier. A single QSFP28 port carries up to 4x 25GbE for low-latency streaming into NVIDIA Holoscan, a 120-pin MIPI interface supports up to eight GMSL2 automotive cameras for 360-degree perception, and a second 10GbE pair plus a dedicated 1GbE EtherCAT port separates high-bandwidth sensor traffic from deterministic motion control. Power is a wide 9V to 48V DC input, the operating range is -25C to +70C, and four isolated CAN FD channels, isolated DI/DO, dual serial and UART/I2C headers wire directly into robot actuator buses.
Key Benefits
2070 FP4 TFLOPS of Blackwell inference in a 140 x 165 mm chassis, 128GB of LPDDR5X so large perception and VLM models fit on-device, QSFP28 plus dual 10GbE and hardware EtherCAT for deterministic sensor and motion traffic, 120-pin MIPI for up to eight GMSL2 cameras, a -25C to +70C envelope with a 9-48V DC input for mobile platforms, and a JetPack 7 / Isaac GR00T / Holoscan software path with custom BSP support.
Applications
Humanoid and legged robots, autonomous mobile robots and warehouse AGVs, surgical and medical robotics, agricultural and inspection drones, utility and industrial inspection robots, and any physical-AI platform requiring on-robot LLM and multi-camera perception.
Request a Quote — AETINA AIB-AT78/68 — NVIDIA JETSON THOR EDGE SUPERCOMPUTING PLATFORM
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →