Specifications
Board Family
Abaco SBC3901 (2024) and SBC3902 (2023) — 3U VPX single board computers, SOSA aligned
AI Compute Module
NVIDIA Jetson AGX Orin (Ampere architecture GPU + Arm Cortex-A78 CPU)
CUDA / Tensor Cores
2048 NVIDIA CUDA cores, 64 Tensor Cores, 12-core Arm Cortex-A78
AI Performance
Up to 248 TOPS (SBC3901) / up to 275 TOPS (SBC3902) INT8
Memory
64 GB DRAM on-board, 205 GB/s memory bandwidth
On-board Storage
64 GB eMMC internal storage; optional 2 TB NVMe (SBC3901)
Deep Learning Accelerators
2x NVIDIA NVDLA v2 deep learning accelerators + PVA v2 programmable vision accelerators
Video Encode / Decode
Dedicated NVENC hardware encoder and NVDEC hardware decoder
Data Plane
1x 100GbE + 1x 25GbE high-bandwidth data plane (SBC3901)
Control Plane
1x 10GbE control plane connectivity
Expansion
PCIe Gen 4 data and expansion lanes
I/O Interfaces
USB 3.2, DisplayPort and RS-232 serial (SBC3902 IO-intensive card)
SOSA Slot Profile
SOSA slot profile 14.2.16 for SBC3902; compute-intensive slot profile for SBC3901
Cooling / Ruggedization
Conduction-cooled 3U VPX; built to meet MIL-STD-810 environmental standards
Form Factor
3U VPX conduction-cooled module for rugged ATR and SOSA chassis
Software Stack
NVIDIA JetPack, CUDA and full NVIDIA AI frameworks compatibility
Target Workloads
Sensor fusion, real-time GPU processing, deep learning inference, tracking
Vendor
AMETEK Abaco Systems — Huntsville, Alabama; 30+ years rugged embedded electronics
Overview
The AMETEK Abaco Systems SBC3901 and SBC3902 are rugged 3U VPX single board computers built around the NVIDIA Jetson AGX Orin module, bringing 2048 CUDA cores, 64 Tensor Cores and up to 275 TOPS of AI performance into a conduction-cooled, SOSA-aligned form factor. They are designed for autonomous and embedded edge computing where commercial single board computers cannot survive shock, vibration and extended temperature extremes.
The SBC3901 pairs the Jetson AGX Orin with a 100GbE plus 25GbE high-bandwidth data plane and a 10GbE control plane, 64 GB of on-board DRAM, 64 GB eMMC and an optional 2 TB NVMe module. Two NVDLA v2 deep learning accelerators, PVA v2 vision accelerators and dedicated NVENC/NVDEC video engines let a single 3U slot run multiple concurrent AI and video pipelines.
The SBC3902 is the IO-intensive variant aligned to SOSA slot profile 14.2.16, exposing USB 3.2, DisplayPort and RS-232 alongside PCIe Gen 4 expansion and 10GbE, with 205 GB/s of memory bandwidth feeding real-time inference in deployed platforms.
Key Benefits
Rugged by design: conduction-cooled 3U VPX construction engineered to MIL-STD-810 and SOSA standards, so Jetson-class AI performance can be deployed in airborne, naval and land platforms instead of a lab. High-bandwidth data paths: a 100GbE + 25GbE data plane and 10GbE control plane move sensor data and inference results without external switches. Full NVIDIA software stack: CUDA, TensorRT and the JetPack ecosystem carry over unchanged, shortening AI application porting from datacenter prototypes to deployed hardware.
Applications
Sensor fusion for autonomous ground and airborne platforms, radar and EO/IR processing, target tracking and classification, unmanned systems, robotics and industrial machine vision in defense, aerospace and heavy industrial environments.
Request a Quote — ABACO SBC3901/SBC3902 3U VPX AI SINGLE BOARD COMPUTERS — NVIDIA JETSON AGX ORIN, 248-275 TOPS, SOSA ALIGNED
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →