The Obox is a demonstrator compute platform for a multi-domain controller to support SAE autonomy levels 3 through 5.
The platform provides a scalable, flexible, silicon-agnostic architecture enabling multiple combinations of CPU, GPU and FPGA compute elements along with a GbE/10GbE network switch and various additional general I/O options.
The flexible architecture supports up to two
Intel® CPUs of either Xeon® D or Xeon E3 class providing high performance general purpose compute capability.
In addition, it supports up to two ‘accelerator processing’ nodes which can be either FPGAs or NVIDIA® GPUs, depending on processing requirements. These enable the platform to run high intensity sensor processing algorithms
using artificial intelligence or conventional techniques.
A central, fully managed network switch
enables a high level of connectivity
to external sensors and other
subsystems as well as
enabling domain partitioning
for safety and security.
* CANbus is a future option, please contact Abaco for more details.
Up to 14 TFLOPs compute power
Up to two Intel Xeon D 12-core processor with 32GB SDRAM
Up two Intel Xeon E-1505 4-core processor with 12GB SDRAM
Up to two NVIDIA GP104 GPUs
Up to two Xilinx Ultrascale+ FPGAs.
One 1GbE or 10GbE fully managed network switch providing 8+ external ports.
Additional I/O: DVI , RS422/485 ports, USB2.0, USB3.0, CANbus*, GPIO.
operating temperature: -40ºC to +71ºC
Internal SSD up to 4 TB
Ubuntu® 16.04 currently supported
Capable of QNX, Green Hills INTEGRITY™, VxWorks®, Windows®.
The two ‘accelerator’ processors support the ability to run high performance neural network-based inference engines to perform tasks such as object detection and image segmentation of multiple video image streams and Lidar point clouds. Developers can utilize GPUs via NVIDIA’s Deep Learning SDK, or FPGAs via Xilinx’s ML suite.
Abaco has successfully integrated the Obox – running ROS - onto its robotic vehicle demo platform. The vehicle uses multiple cameras, Lidar and radar sensors, and enables Abaco to understand the challenges of autonomous driving and to develop solutions.”
The high-performance Intel Xeon D processor, with up to 16 cores, is a dense, low power system-on-chip solution, ideal for High Performance Embedded Computing (HPEC) applications where high core count and substantial memory capacity are of paramount importance.
The high-performance Intel Xeon E3 processor, with four cores plus GPU system-on-chip solution, is ideal for general purpose applications where high performance and significant I/O management are key requirements.
NVIDIA’s GP104 GPU is fully supported by the CUDA™ and Deep Learning SDKs, enabling rapid deployment of both traditional and AI-based perception processing. With 2,048 processing cores and single- and double-precision floating point units, the GP104 GPU is capable of CUDA Compute Capability v3.0.
Xilinx® Ultrascale+™ FPGAs provide powerful, flexible custom processing capability while minimizing power consumption and can also support AI base solutions.
Transfers between the CPU and acceleration processors are provided by x8 PCIe™ Gen3 lanes to ensure maximum data rates between
AXIS ImageFlex provides optimized visualization capabilities to combine video, Lidar, radar and other sensor information into one unified viewer application. It provides high performance algorithms for image transformation to correct for optical distortions in real-time and also to project the image for different viewpoints such as ‘bird’s eye. It also provides the ability to stitch multiple video streams for 360° surround imagery via a ‘Skybox’.
CPU options (up to 2)
• Xeon D high density, low power SoC
• 12-core CPU standard
• 32 GB DDR4 SDRAM with ECC
• 32 GB onboard SSD (on SBC)
• 2 x8 PCIe Gen 3 to GPU
• Xeon E3, low power SoC
• 4-core CPU, integrated GPU
• 16 GB DDR4 SDRAM with ECC
• 32 GB onboard SSD (on SBC)
• x8 PCIe Gen 3 to GPU
Acceleration processing options (up to 2)
• Up to NVIDIA GP104 GPU
• 2048 Cores
• Up to 16 GB GDDR5
• Virtex Ultrascale+ XCKU115
• Zync Ultrascale
• 10 GB DDR4 SDRAM
• FMC site for I/O
• 12x 1000BASE-T and 2x 10GBASE-KX4
• Wire speed Layer 2/3 switching
• Multicast support
• IEE 802.3-2005; 1D, 3ad, 1Q, 1Q-2005, 1X
• PTP 1588v2 support
• 2x DVI
• 8x 1000BASE-T Ethernet
• 4x USB 2.0
• 2x USB 3.0
• 2x RS232/422/485
Removable or Fixed SSD
• 128 GB - 4 TB
• Ubuntu 16.04 supported
• QNX, GHS Integrity, VxWorks capable
• Windows 10 capable
• NVIDIA CUDA & Deep Leaning SDKs
• Abaco AXIS ImageFlex +
• Convection cooled
• Operating temperature: -40°C to + 71°C
• Air-cooled variants available