The WOLF-1636, also known as the VPX3U-BW5000E-VO-HPC, combines NVIDIA Blackwell GPU processing with a rugged 3U VPX architecture.
The module includes 10,496 CUDA Cores for parallel processing and 320 fifth-generation Tensor Cores for artificial intelligence and high-performance embedded computing. Furthermore, its 24 GB of GDDR7 256-bit memory supports ECC and provides 896 GB/s memory bandwidth.
As a result, defence hardware, software and systems integration engineers can incorporate GPU processing and video output within an open-architecture VPX platform.
NVIDIA Blackwell GPU Processing
The NVIDIA RTX 5000 embedded GPU supports CUDA Toolkit 13 and Compute Capability 12.0. In addition, developers can use CUDA-X AI and CUDA-X HPEC libraries for supported AI and embedded computing applications.
GPU capabilities include:
- 10,496 CUDA Cores
- 320 fifth-generation Tensor Cores
- 80 RT Cores
- 24 GB GDDR7 256-bit VRAM
- ECC memory support
- 896 GB/s memory bandwidth
- CUDA Toolkit 13
- CUDA Compute Capability 12.0
- CUDA-X AI and CUDA-X HPEC libraries
- NVIDIA GPUDirect RDMA
- OpenCL 3.0
- DirectX 12 Ultimate
- OpenGL 4.6
- OpenGL ES 3.2
- Vulkan 1.2
The fifth-generation Tensor Cores also support additional data precisions, including FP4, FP6 and second-generation FP8.
Configurable Power and Performance
The WOLF-1636 has a configurable module power range from 90 W to 150 W. However, the selected power setting and available cooling will affect the GPU clock speed and processing performance.
At the default 100 W setting, the GPU provides up to 24.2 TFLOPS at its base clock and up to 38.5 TFLOPS at its boost clock. Where sufficient cooling is available, the module can be configured for up to 150 W and a maximum performance of up to 50 TFLOPS.
Therefore, the required performance setting should be considered alongside the chassis power and thermal design.
Hardware-Accelerated Video Processing
The Blackwell GPU includes ninth-generation NVENC video encoding and sixth-generation NVDEC video decoding hardware. Consequently, the module supports hardware-accelerated video encoding and decoding at resolutions up to 8K without placing the full processing load on the host CPU.
Supported capabilities include AV1 hardware encoding and decoding through the NVIDIA Video Codec SDK.
Front and Rear DisplayPort Outputs
Firstly, two DisplayPort outputs are accessible beneath a removable front panel. These outputs are available with the supported SOSA-aligned and OpenVPX slot profiles.
In addition, an OpenVPX 14.2.7 configuration can route two DisplayPort outputs through the rear P2B connector. An HDMI option is also available for rear outputs.
However, the P2B DisplayPort pin assignment is not pin-compatible with the previous-generation WOLF-1538, WOLF-1448 or WOLF-1348 modules. Therefore, the backplane and video-routing requirements should be checked when selecting or replacing a module.
SOSA-Aligned and OpenVPX Integration
Depending on the system architecture, the WOLF-1636 can be configured for:
- SOSA-aligned 14.6.11-0 payload slot profile with P2 depopulated and PCIe x8
- SOSA-aligned 14.6.13-0 payload slot profile with P2B depopulated and PCIe x16
- OpenVPX 14.2.7 slot profile
The module supports PCIe Gen4 connectivity, IPMI system management and NVIDIA GPUDirect RDMA. Moreover, driver support is available for Linux and Windows.
Configuration options include the slot profile, display interfaces, power threshold, variant locking and conformal coating.
Rugged Conduction-Cooled Construction
The WOLF-1636 uses a rugged, conduction-cooled design. Because GPU performance depends partly on temperature, WOLF’s thermal design transfers heat from the GPU die to the module wedgelocks.
Mechanical and environmental specifications include:
- Standard operating temperature from -40°C to +70°C
- Operational temperature up to +85°C
- VITA 47.1 Class V3 random vibration from 5 to 2000 Hz
- Sine vibration of 10g peak from 5 to 2000 Hz
- Shock rating of 40G to MIL-STD-810H, Method 516.8
- Dimensions of 160 mm × 100 mm × 25.4 mm
- Approximate weight of 1.3 kg
- ANSI/VITA 48 and VITA 65 support
WOLF designs and manufactures the module in North America with full component traceability. In addition, WOLF’s quality management systems are certified to AS9100D and ISO 9001:2015.
Key Features
- NVIDIA RTX 5000 Blackwell embedded GPU
- 10,496 CUDA Cores and 320 Tensor Cores
- 24 GB GDDR7 memory with ECC support
- Rugged conduction-cooled 3U VPX construction
- Configurable PCIe Gen4 x8 or x16 connectivity
- Configurable module power from 90 W to 150 W
- Two front-panel DisplayPort outputs
- Two rear DisplayPort outputs with OpenVPX 14.2.7
- HDMI rear-output option
- Hardware-accelerated video encoding and decoding up to 8K
- NVIDIA GPUDirect RDMA
- IPMI system management
- SOSA-aligned and OpenVPX profile options
- Linux and Windows driver support
Typical Applications
- Defence and aerospace embedded computing
- High-performance embedded computing
- Artificial intelligence processing
- Data science computations
- GPU parallel processing
- Hardware-accelerated video encoding and decoding
- Embedded graphics processing
- Mission visualisation
Frequently Asked Questions
What GPU does the WOLF-1636 use?
The WOLF-1636 integrates an NVIDIA RTX 5000 embedded GPU based on the Blackwell GB203 architecture. It includes 10,496 CUDA Cores, 320 Tensor Cores, 80 RT Cores and 24 GB of GDDR7 memory with ECC support.
Is the WOLF-1636 SOSA aligned?
Yes. The module supports SOSA-aligned 14.6.11-0 and 14.6.13-0 payload slot profiles. In addition, an OpenVPX 14.2.7 configuration is available.
How are the DisplayPort outputs accessed?
Two DisplayPort outputs are located beneath a removable front panel. Alternatively, the OpenVPX 14.2.7 configuration can route two DisplayPort outputs through the rear P2B connector. Therefore, the required display routing should be confirmed when selecting the module configuration.
Local WOLF Product Support
Metromatics supplies and supports WOLF rugged embedded computing technology throughout Australia and New Zealand.
Our team can assist defence hardware, software and systems integration engineers with product selection, slot profiles, video interfaces, power options and technical enquiries. Therefore, customers can discuss their VPX integration requirements with a local supplier before selecting the module configuration.
Discuss Your VPX GPU Requirements
Looking for an NVIDIA Blackwell 3U VPX GPU module for embedded AI, HPEC, graphics or video processing?
Contact Metromatics to discuss the WOLF-1636, required slot profile, video interfaces, power configuration and VPX integration requirements.
To view other 3U Wolf Versions
