The rugged VPX3-730 high-performance embedded computing (HPEC) GPU card combines an NVIDIA® RTX PRO™ Blackwell-generation GPU and ConnectX-7® SmartNIC for 100 Gigabit Ethernet (GbE) and PCI Express® (PCIe) Gen 4/5 connectivity in a 3U OpenVPX module aligned to SOSA Payload profiles. Optional dual 8K video outputs are also offered in an OpenVPX compatible profile.

The VPX3-730 GPU card provides advanced processing capabilities and hardware acceleration for GPU-accelerated CUDA computing, graphics and artificial intelligence (AI) algorithms. Its on-board 100GbE Ethernet interface supports low-latency GPUDirect® Remote Direct Memory Access (RDMA) and RDMA over Converged Ethernet (RoCEv2) capabilities for fast direct data transfers to/from 24GB of local GDDR7 GPU memory, together with a 16-lane PCIe Gen 4/5 interface for connectivity to the VPX backplane.

The NVIDIA Blackwell architecture includes CUDA® cores for HPEC processing, Tensor cores for AI and data science computations, and Ray Tracing (RT) cores for visually accurate rendering. NVIDIA’s advanced Blackwell GPUs have more CUDA and Tensor cores to deliver significantly more performance per watt compared to previous Ada, Turing and Ampere generation products, and more than double the integer math throughput compared to prior Ada generation.

The VPX3-730 is aligned with The Open Group Sensor Open Systems Architecture® (SOSA) Technical Standard Payload profiles with 8-lane and 16-lane PCIe variants, as well as legacy OpenVPX profiles, with support for multiple DisplayPort++ interfaces to directly drive visual displays.

Optimized for high-speed backplane signal integrity, the VPX3-730 joins the Fabric100™ ecosystem of VPX system building blocks supporting 100 Gigabit Ethernet and next-gen PCIe connectivity with proven interoperability that reduces system integration risk.