WOLF-2638
Overview
Dual NVIDIA Blackwell RTX 5000 HPEC with PCIe Gen5 Switch
The VPX6U-BW5000E-DUAL-VO module includes two NVIDIA RTX 5000 Blackwell embedded GPUs and a PCIe Gen5 switch in a rugged 6U VPX module. The NVIDIA RTX 5000 embedded GPU provides advanced processing capabilities for high performance embedded computing (HPEC) and artificial intelligence (AI) data processing.
The NVIDIA Blackwell architecture includes CUDA cores for HPC, and 5th generation Tensor cores for AI and data science computations. The Blackwell GPU has an improved architecture which provides increased efficiency. The module also supports 24GB of GDDR7 memory which provides over 50% higher bandwidth compared to the previous generation. The GPU supports PCIe Gen5, providing a fast data transfer path to/from the module.
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 path from the GPU die to the wedgelocks.
Block Diagram
The block diagram illustrates a 6U VPX dual-GPU architecture with two RTX 5000 Blackwell GPUs connected through a PCIe Gen5 switch. It highlights the PCIe fabric routes used for GPU-to-GPU/peer transfers and high-throughput uplinks to the VPX backplane, plus the display routing used for multi-output visualization configurations.
Features
WOLF-2638 is a rugged 6U VPX module that integrates two NVIDIA RTX 5000 Blackwell embedded GPUs with a configurable PCIe Gen5 switch for high-density embedded AI and HPEC acceleration. It includes 24 GB GDDR7 256-bit VRAM per GPU with ECC support and provides up to 8 DisplayPort outputs with options for HDMI by configuration. Module power is configurable from 190W to 320W, and the design supports liquid flow-through operation with OpenVPX and SOSA aligned profile support 10.6.4.
GPU Features
- Blackwell GPU parallel processing:
- CUDA Toolkit 12, CUDA Compute capability 10.0
- CUDA-X AI and CUDA-X HPEC libraries
- OpenCL 3.0, DirectX 12 Ultimate, OpenGL 4.6, OpenGL ES 3.2, Vulkan 1.2
- 320 Tensor Cores (4th Gen) per GPU
- 80 Ray Tracing cores (3rd Gen) per GPU
- NVENC (9th Gen) and NVDEC (6th Gen) with up to 8K video encoding and hardware decoding support
Connectivity / System Management
- IPMI system management
- NVIDIA GPUDirect RDMA support
- Configurable PCIe Gen5 switch
- Linux and Windows drivers
- NVIDIA driver support requires the following host CPU: Intel H/HX/P/PX/S Class, AMD H/HS Class
Mechanical / Open Systems Architecture
- High level of ruggedization:
- Rugged conduction cooled (CC), air cooled (AC), or liquid flow through (LFT)
- Operating temperature: CC: - 40°C to +70°C standard, operational to +85°C, AC: - 40°C to +60°C standard, operational to +71°C
- Vibration (sine wave): 10G peak, 5 - * Shock: 40G peak for conduction cooled, 30G peak for air cooled
- Dimensions: 160mm x 233mm x 25.4mm
- Weight (approximately): TBD
- ANSI/VITA 48, 65 (VPX- REDI, OpenVPX)
- OpenVPX and SOSA Aligned profile support 10.6.4
Specifications
Key configuration and performance parameters for WOLF-2638 are summarized below for quick comparison. Values shown reflect the standard module configuration; contact WOLF if you need a tailored build or a specific payload profile.
Configuration Guide
The following table defines a series of common order codes for the VPX6U-BW5000E-DUAL-VO module. The asterisks denote characters of the part number that are defined based on common configuration options.
Some common configuration options for this module include:
- Display interfaces
- Conformal coating type
- COTS, MCOTS, or locked
- PCIe switch configuration
- OpenVPX or SOSA aligned configuration
- Alternative cooling Configuration; AFT
6U VPX Blackwell RTX5000 sample configurations
Contact Sales for the latest Ordering Numbers and available options
Representative ordering numbers
| Ordering Number | Description |
|---|---|
| 263833-F***-***VPX6vA0 | 6U VPX, conduction cooled, 1", dual NVIDIA RTX5000 Blackwell, rear: 8x video outputs |
| 263833-F***-***VPX6vA0 | 6U VPX, conduction cooled, 1", single NVIDIA RTX5000 Blackwell, rear: 4x video outputs |
| 263823-F***-***VPX6vA0 | 6U VPX, Liquid Cooled, 1”, Dual NVIDIA RTX5000 Blackwell, Rear: 8x video outputs |
Related Products
- NVIDIA RTX 5000 Ada (AD103)
- 9728 CUDA Cores, 304 Tensor Cores
- 16 GB GDDR6 VRAM
- PCIe Gen 4
- SOSA Payload
- Module power 150–280 W
- NVIDIA RTX 5000 Turing (TU104)
- 3072 CUDA Cores, 384 Tensor Cores
- 16 GB GDDR6 VRAM
- PCIe Gen 3
- SOSA Payload
- Module power 100–300 W
FAQ
Choose WOLF-2638 when your application benefits from two GPUs in one slot—either for higher aggregate throughput, workload partitioning, or running multiple parallel pipelines—without consuming additional chassis slots.
This is most useful when:
- You need more compute density than a single GPU can provide
- You want to split workloads across GPUs (e.g., multi-sensor pipelines, concurrent AI + rendering)
- You want to scale performance while keeping the VPX slot count low
If your workload is single-stream and does not parallelize well, a single-GPU module can be a simpler, lower-power integration.
A PCIe Gen5 switch is primarily about topology flexibility and bandwidth headroom.
In practice it enables:
- Flexible PCIe routing to balance host traffic across both GPUs
- A cleaner path for peer/endpoint movement through the switch fabric
- Better alignment to platform PCIe constraints without redesigning the compute module
For many systems, the switch reduces integration risk by allowing the module to map cleanly into different backplane PCIe topologies. The performance benefit is most visible when your pipeline is sensitive to host-to-GPU transfer limits or you are feeding both GPUs heavily.
Confirm these early because they are the highest-risk integration items for a dual-GPU 6U module:
- Cooling method: the design supports conduction-cooled (CC), air-cooled (AC), or liquid flow-through (LFT). Your chassis must match the selected cooling style and its thermal headroom.
- Slot profile alignment: supports OpenVPX and SOSA aligned profile support 10.6.4—confirm your backplane/slot is designed for that profile family.
- Display routing plan: supports up to 8 DisplayPort outputs with HDMI option by configuration. Decide early whether outputs are front-access, rear-routed, or mixed (configuration dependent) so rear I/O and harnessing can be designed once.
Treat this module as both a compute and visualization resource; display routing and cooling choices frequently drive integration more than raw GPU specs.
Most ordering-time tradeoffs are system-integration decisions:
- Cooling variant (CC vs AC vs LFT): drives chassis compatibility, allowable power, and sustained performance.
- COTS/MCOTS/locked builds and conformal coating: impacts qualification, maintainability, and configuration control.
- PCIe switch configuration: determines how PCIe lanes are allocated and balanced across the two GPUs and uplinks.
- Display interface configuration: number of enabled outputs (up to 8 DP) and whether HDMI is required.
- Power threshold (190W-320W envelope): must be set to match chassis thermal capacity for deterministic sustained performance.
Contact Sales for the latest ordering numbers and available configuration options.