NVIDIA® High-Performance TESLA® CUDA compute engine with 8GB GDDR5 and PCIe x16 Gen 3

NORTH AMERICA – WOLF Advanced Technology’s VPX3U-TESLA-M6 board incorporates a high performance GPGPU processor, 8GB GDDR5 memory, GPUDirect™ DMA, and a high speed PCIe x16 Gen 3 interface. This board provides advanced parallel processing and high data throughput along with the highest performance per watt CUDA compute engine solution currently available in a single 3U VPX slot.

WOLF's VPX3U-TESLA-M6 board uses NVIDIA’s latest Maxwell-2 28nm GPU technology, with 2857 GFLOPs single precision (32-bit), 8GB GDDR5 video memory with a theoretical maximum memory bandwidth of 147.2 GB/s,  and 1536 NVIDIA CUDA cores. The VPX3U-TESLA-M6 can perform computationally intensive operations, such as image processing, video stabilization, filtering, terrain analytics, 3D visualization of geospatial data, object recognition and tracking. Using the TESLA M6 GPGPU the board can also be used to process non-graphical data that would normally be processed by a CPU, taking advantage of the GPU’s high number of cores to significantly speed up computational problems which can take advantage of multiple threads.

The VPX3U-TESLA-M6 board can operate in a high performance compute mode (HPC mode) when used with server class CPUs. When the board is in this mode the entire 8GB of Tesla memory is exposed to the entire PCIe BAR. The board supports the GPUDirect API to write into the Tesla’s memory, reducing latency, system power and bottlenecks from the system.

Greg Maynard, WOLF’s CTO, says “allowing devices to DMA directly into the GPU's memory is a huge advantage, especially in a VPX architecture where there are often multiple SBCs and GPUs in a system which can all benefit from device to GPU, second system domain to GPU, or GPU to GPU direct connections.”

The NVIDIA’s TESLA M6 is compatible with NVIDIA GRID 2.0 vGPUTM, an advanced technology for sharing true virtual GPU hardware acceleration between multiple users. With this technology up to 16 users can simultaneously access a powerful graphics processor, and GPU resources can be assigned using the GRID vGPU Manager based on project and application needs.

The  VPX3U-TESLA-M6 is an ancillary processor board designed to be installed into a VPX system with a VPX-REDI (ANSI/VITA 48.x) in a VPX (ANSI/VITA 46.x) or OpenVPX (ANSI/VITA 65) configuration. The board implements PCIe x16 and can support specialized 3U VPX systems designed to provide the full x16 connection. It also implements PCIe Gen 3, which for a x16 has a data throughput of 15.76 GB/s per direction, almost twice the 8 GB/s data throughput of PCIe 2.0.

“It is unusual to find a 3U VPX board that has a PCIe x16 Gen 3 implementation” says Greg Maynard. “It can support specialized systems designed to provide the full x16 connection, at the faster data rates allowed by Gen 3.”

This board has been designed to operate in rugged environments and has been manufactured to MIL-STD-810 standards including 20g 11ms shock, 0.1g²/Hz @ 5 - 2000Hz vibration,  -40 to +71°C, and can be conduction or air-cooled. Due to the critical importance of size, weight, and power (SWaP) in aerospace and defense applications, each compact, ruggedized VPX3U-TESLA-M6 graphics processing board includes a utility and API allowing the maximum power to be controlled in order to maximize GPGPU capability while minimizing power usage.

FOR MORE INFO: http://wolfadvancedtechnology.com/products/vpx3u-tesla-m6/

WOLF Advanced Technology designs, develops and manufactures ruggedized boards for video capture, process, encode and display. All WOLF solutions are designed to operate flawlessly in even the harshest aerospace and defense environments without sacrificing any of the processing power available from the latest generation high-speed NVIDIA, AMD and Xilinx GPUs, APUs and FPGAs. WOLF’s designs include rapidly available COTS solutions, feature-tailored MCOTS solutions, and full custom designs. WOLF supports a variety of board architectures, including 3U and 6U VPX, XMC, PMC and MXC/MXM.