SBC Includes Xavier with Volta GPU & Carmel CPU, WOLF FGX
- NVIDIA Volta GPU with 512 CUDA cores & 64 Tensor cores, 1.4 TFLOPS FP32, 22.6 DL TOPS 8-bit
- NVIDIA Carmel ARM64 CPU with 8 cores, 2.26 GHz
- 32GB or 16 GB LPDDR4 256-bit memory with up to 137 GB/s
- WOLF FGX for support of additional video formats such as SDI and CVBS
- Module power: 30-50W
The VPX3U-XAVIER-SBC module includes both an NVIDIA Jetson AGX Xavier APU and a WOLF FGX in a rugged 3U VPX module. The Tegra Xavier SoC includes an advanced Volta GPU, Carmel ARM64 CPU, and additional accelerator circuits for deep learning inference, machine vision, audio processing and encoding/decoding. The Xavier also achieves a huge upgrade in power efficiency thanks to a 12nm fabrication process. The WOLF FGX provides support for video formats which are not native to the GPU.
The Volta GPU with its Tensor cores provides this module with the underlying architecture required for an efficient AI inference engine. The module can achieve up to 32 TeraOPS (TOPS) of deep learning inference computing using the GPU and Deep Learning Accelerator (DLA) engines. Intended to work in conjunction with TensorRT, CUDA and CuDNN, the Tensor Cores supports Int8 and experimental Int4 / 1 bit matrix operations.
Unlocking the best performance requires the best cooling capability. WOLF’s advanced cooling technology is designed to move heat using a low weight, high efficiency pipeline from the SoC die to the wedgelocks.
WOLF-12TP 3U VPX Module front panel for Air Cooled model
This product is currently in production as a COTS product with configurable options.
Contact WOLF to discuss your customization or MCOTS requirements.
Video Overview of 3U VPX Xavier SBC
- 3x DisplayPort 1.2a up to 4K-UHD, or HDMI / DVI
- 2x 3G/HD-SDI outputs, 2x CVBS outputs
- Two Deep Learning Accelerator (DLA) engines for inference operations, 11.4 TOPS Int8
- Two Vision Accelerator engines: 7-way VLIW Vision Processor, 1.7 TOPS Int8
- Dedicated programmable audio processor
- Two NVENC/NVDEC for HEVC (H.265) and AVC (H.264), encoding up to 4K-UHD resolution
- 2x 1000BASE-T Ethernet
- CUDA® 10, OpenGL® 4.6, OpenGL ES 3.2, Vulkan™ 1.0
- Flash Storage: 32 GB 8-bit eMMC 5.1 with support for ECC
- Optional NVME SSD, 256GB to 1TB
- Optional SATA connection through P2
- USB ports with support for USB 3.1 Gen 1 (5 Gb/s) and USB 2.0 (up to 480 Mb/s)
- PCI Express x8, x4
- Daisy Chain and NTB Options Available
- Standalone operation or connect to Windows, Linux or ARM root complex
- High level of ruggedization:
- Rugged Air-cooled or Conduction-cooled
- Operating temperature: -40° to +85°C
- Vibration (sine wave): 10G peak, 5 - 2000Hz
- Shock: 30G peak for air-cooled, 40G peak for conduction-cooled
- Dimensions: 160mm x 100mm x 25.4mm
- Weight: To be Announced
- +12V or +5V Only Power Source Options
- ANSI/VITA 48, 65 (VPX REDI, OpenVPX)
Open the product datasheet - includes block diagram, order codes, etc.