The FabreX™ AI Memory Fabric Platform by GigaIO is a next-generation, memory-centric fabric designed to revolutionize data center architectures in response to the exponential growth of data and the rapid adoption of advanced analytics and Artificial Intelligence (AI). By disaggregating traditional server components and enabling dynamic composition of resources, FabreX addresses the challenges posed by modern compute and storage clusters, offering unparalleled flexibility, performance, and efficiency. Key Features and Functionality: - Memory-Centric Fabric: FabreX connects memory, storage, and a wide array of accelerators—including GPUs, FPGAs, and custom ASICs—either directly or via configurations like NVMe-oF, delivering industry-leading low latency and high bandwidth. - High Performance: With latency from system memory of one server to another being less than 200 nanoseconds and bandwidth scaling up to 512 Gbits/sec in its Gen4 implementation, FabreX ensures true PCIe performance across entire clusters. - Unmatched Flexibility: The platform enables the composition of diverse resources, such as GPUs, DPUs, TPUs, FPGAs, SoCs, NVMe storage, and other I/O devices, across multiple servers and racks. It supports device-to-node, node-to-node, and device-to-device communication within the same high-performance PCIe memory fabric. - Open Standards Compliance: FabreX is 100% PCI-SIG compliant, ensuring seamless integration with heterogeneous computing, storage, and accelerator components into a unified system-area cluster fabric. Primary Value and User Solutions: FabreX addresses the critical need for scalable, flexible, and efficient data center architectures capable of handling the demands of AI, Machine Learning (ML), and Deep Learning (DL) applications. By disaggregating server components and enabling dynamic resource composition, it eliminates bottlenecks and configuration challenges inherent in traditional interconnect systems. This approach not only enhances performance but also optimizes resource utilization, reducing the total cost of ownership and allowing data centers to scale both up and out seamlessly.

The GigaIO™ Accelerator Pooling Appliance – MI300X is a high-performance PCIe accelerator appliance designed to enhance AI/ML training, high-performance computing (HPC), and data analytics applications. It fully supports PCIe Gen5, offering up to 2.048Tb/s total bandwidth for host server connections. Equipped with eight AMD Instinct MI300X 192GB 750W OAM GPUs, it provides a total of 1.54TB of high-bandwidth memory (HBM), enabling efficient processing of complex workloads. Key Features and Functionality: - High Capacity: Accommodates 8x AMD Instinct MI300X 750W accelerators, delivering substantial computational power. - Exceptional Performance: Offers ultra-low latency with 512Gb/s uplinks, ensuring rapid data transfer and processing. - Ample Memory: Provides a total of 1.54TB HBM (8x 192GB per MI300X), facilitating efficient handling of large datasets. - Simplified Deployment: Features RESTful APIs and a WebGUI for straightforward integration and management. Primary Value and User Solutions: The GigaIO Accelerator Pooling Appliance – MI300X addresses the need for scalable and efficient computational resources in demanding environments. By enabling dynamic provisioning and scaling of PCIe devices, it allows users to allocate GPU resources as needed, optimizing utilization and reducing idle hardware. Its centralized management and continuous monitoring capabilities enhance reliability and facilitate rapid problem resolution, making it an ideal solution for AI/ML training, HPC, and data analytics acceleration.

The GigaIO SuperNODE™ is a groundbreaking single-node supercomputer designed to meet the demands of next-generation AI and accelerated computing workloads. By integrating up to 32 AMD or NVIDIA GPUs into a single server, SuperNODE eliminates the complexities associated with multi-server configurations, offering a streamlined and efficient solution for intensive computational tasks. Key Features and Functionality: - High-Density GPU Integration: Supports up to 32 AMD Instinct™ MI210 GPUs or 24 NVIDIA A100 GPUs within a single node, providing exceptional computational power. - FabreX™ Memory Fabric: Utilizes GigaIO’s FabreX, a high-performance PCIe memory fabric, to seamlessly connect all accelerators, ensuring low-latency and high-bandwidth communication. - Energy Efficiency: Operates at approximately 7 kilowatts per 32-GPU deployment, reducing power consumption compared to traditional multi-server setups. - Space Optimization: Achieves a 30% reduction in rack space requirements, allowing for higher computational density within existing data center infrastructures. - Software Compatibility: Compatible with popular AI frameworks like PyTorch and TensorFlow, enabling users to run existing applications without modification. Primary Value and Problem Solved: SuperNODE addresses the challenges of deploying and managing large-scale AI and high-performance computing infrastructures by consolidating extensive GPU resources into a single, efficient node. This consolidation reduces network overhead, minimizes latency, and simplifies system administration. By eliminating the need for complex multi-server configurations and associated networking equipment, SuperNODE offers a cost-effective, energy-efficient, and high-performance solution for organizations aiming to accelerate their AI and computational workloads.

The GigaIO™ Fabric Card is a high-performance network adapter designed to facilitate non-blocking, low-latency composable fabric computing at rack scale. It enables users in AI/ML, HPC, and data analytics to construct tailored systems that optimize performance while reducing total cost of ownership. By supporting a high-speed, cabled interface to cluster subsystems across GigaIO's AI fabric network, the Fabric Card allows for the creation of shared pools of vendor-agnostic PCIe devices, including GPUs, FPGAs, storage, and memory. This flexibility ensures seamless integration and management of disaggregated resource pools. Key Features and Functionality: - High Performance: Delivers up to 512Gb/s speed and 128GB/s bandwidth, ensuring rapid data transfer and processing capabilities. - Low Latency: Achieves latency of less than 10 nanoseconds, facilitating real-time data access and communication. - Versatile Connectivity: Equipped with dual QSFP-DD connections, supporting both copper and optical cabling options for flexible deployment. - Compact Design: Features a low-profile form factor compatible with both full-height and half-height PCIe slots, allowing for easy integration into various server configurations. - Dual Operational Modes: Offers Host Mode for installation into host or head-node servers and Target Mode for integration into Accelerator Pooling Appliances or resource boxes, enhancing adaptability across different system architectures. Primary Value and User Solutions: The GigaIO Fabric Card addresses the growing need for scalable and flexible computing infrastructures by enabling the dynamic composition of hardware resources. It allows organizations to disaggregate and recompose their computing resources on demand, leading to improved resource utilization, enhanced system performance, and reduced operational costs. By supporting a wide range of PCIe-compliant devices, the Fabric Card empowers users to build customized, high-performance computing environments tailored to their specific workload requirements.