EIZO Rugged Solutions has developed a new, cutting-edge embedded 3U VPX single-board computer featuring NVIDIA® Jetson Thor. The Condor™ T5-IOX is EIZO’s next-generation SOSA® aligned OpenVPX embedded SBC powered by the industrial-grade NVIDIA Jetson Thor SoM. With a compact, rugged design optimized for SWaP-constrained platforms, the Condor™ Thor 3U VPX modules bring server-class compute directly into embedded ISR systems, reducing latency and reliance on centralized processing.
The Condor T5-IOX is engineered to ingest and process multiple ultra-high-definition video streams simultaneously, enabling multi-sensor fusion across ISR, radar, and EO/IR systems. The single board computer feature high-speed I/O options such as 12G-SDI, USB 3.2, and DisplayPort™, along with external storage support. The series supports 25GbE Gigabit Ethernet and NVIDIA GPUDirect RDMA over Converged Ethernet (RoCE), enabling rapid communication between sensors, RF transceivers, and platforms.
Accelerated by an NVIDIA Blackwell GPU and a 14-core Arm® Neoverse®-V3AE CPU, the Jetson Thor module delivers exceptional edge-compute performance with up to 128 GB of high-bandwidth LPDDR5X memory. Its Blackwell architecture, featuring 2,560 CUDA cores, provides up to 2,070 FP4 TFLOPS of AI performance, enabling real-time deep learning and multi-sensor fusion for AI-driven autonomy. The dual-compute architecture maximizes performance density by combining an embedded GPU and CPU in a single 3U VPX slot, reducing overall system size, weight, and power (SWaP).
Designed for flexible deployment, Jetson Thor operates within a configurable power range of 40 W to 130 W, offering 7.5x higher AI compute and 3.5x greater energy efficiency than NVIDIA AGX Orin™.
Like NVIDIA IGX Orin, the new Thor provides a transparent, unified memory model between the CPU and the GPU, allowing for reduced latency and improved throughput when transferring memory between the two. This, combined with the new features of Blackwell, possess incredible benefits for SWaP-C.
NVIDIA Thor positions itself as a high-performance, high-power alternative to the AGX Orin module. When compared to the Orin, Thor provides up to 7.5x higher AI compute and 3.5x better energy efficiency. The Jetson Thor T5000’s 14-core CPU provides 2.5x the performance of the Jetson Orin’s CPU. The Thor T5000’s Blackwell GPU offers 25% more CUDA cores compared to the Jetson AGX Orin 64GB and 50% more Tensor cores. It also has two NVENC and two NVDEC cores, allowing for twice as much video encoding and decoding as the Orin. Additionally, PCIe bandwidth has been doubled due to the inclusion of Gen 5 PCIe.
The Jetson T5000 module offers up to 273GB/s memory bandwidth with its LPDDR5 memory. This improved bandwidth unlocks the power of large models that can now be inferred upon at the edge. This improved bandwidth ensures sensor-rich ISR workloads can be processed with minimal latency and maximum throughput—both critical for real-time decision-making at the tactical edge. With bandwidth comparable to modern data center accelerators, Jetson Thor enables the execution of complex neural networks for sensor fusion, signal processing, and multi-domain intelligence applications directly at the edge, eliminating reliance on remote servers and ensuring timely, mission-critical insights.
Jetson Thor modules contain up to two NVENC encoder cores, two NVDEC decoder cores, and a JPEG processing block (NVJPEG). NVENC enables full hardware acceleration for various encoding standards including H.265 and H.264. NVDEC enables full hardware acceleration for various decoding standards including H.265, H.264, AV1, VP9.
The dual NVENC/NVDEC video encoder and decoder engines in NVIDIA Jetson Thor enable simultaneous support for a high number of video streams—up to 6x 4Kp60 in H.265 to 48x 1080p30 in H.264.
NVIDIA has continued refining the Tensor Cores within its GPUs to improve the performance of AI models. The 5th Generation of their Tensor cores that ship in the Thor modules improve the performance of sparse AI models by using structured sparsity and data compression. These improvements add up to a 2x performance improvement for sparse model inference on Tensor Cores.
Additionally, the Thor is the first of the line to introduce Multi-Instance GPU, or MIG. MIG allows a GPU to be partitioned into multiple instances – each running with its own memory, cache, and CUDA cores. While this is currently limited in that only one instance can process graphics, it offers a promising option for containerized GPGPU applications in particular – such as those using NVIDIA Holoscan – and improves safety and security by allowing applications to run sandboxed within a particular region of the GPU.
The Ethernet capability of the Thor has been improved as well. The T5000 module supports four 25-Gigabit Ethernet interfaces for a combined 100 gigabits per second of network bandwidth. This is a vast improvement over the Orin, which has only 1/10th of that capacity. Via these interfaces, the Thor also integrates support for NVIDIA’s new Camera over Ethernet interface, which provides zero-copy, low-latency camera feeds by leveraging Holoscan Sensor Bridge and RDMA over Converged Ethernet (RoCE) technologies on supported cameras.
For system integrators looking to streamline AI/ML deployment in embedded VPX environments, the Condor Thor series is the perfect match. By combining NVIDIA’s most advanced AI platform with our rugged video and graphics expertise, we are providing the integrators with the hardware to rapidly deploy next-generation multi-sensor applications and autonomous systems that demand both performance and reliability.
Explore the industries we proudly serve!
EIZO Rugged Solutions is a pioneer in providing video, graphics, encoding, and recording solutions for over three decades.
