WT Microelectronics Singapore Pte. Ltd.

07/08/2026

07/08/2026

At FMS 2026, Marvell demonstrated the Marvell® Structera™ X 2504 memory expansion controller, which enables up to 8 TB of DDR5 memory across eight DIMMs. Built on the industry-standard Compute Express Link (CXL) protocol, it expands available memory capacity for memory-intensive applications such as AI, analytics, and in-memory databases.

This supports larger AI and data-intensive workloads by significantly increasing available memory capacity, and improves memory utilization and flexibility through CXL, allowing memory resources to be shared and allocated more efficiently. By decoupling memory from individual CPUs, data centers can scale infrastructure more effectively. This solution also future-proofs server deployments by enabling memory expansion without requiring a complete redesign of compute infrastructure and adapts to evolving workload demands while maintaining high performance for AI, cloud, and in-memory database applications.

Learn more: https://mrvl.co/4bxHfyH

07/08/2026

GPU utilization in AI inference is increasingly constrained not by compute capacity but by memory access. As KV caches grow into the hundreds of gigabytes and beyond, HBM is too expensive to scale proportionally and NVMe storage introduces too much latency. The memory hierarchy needs a new tier.

Marvell Photonic Fabric technology addresses this by enabling shared memory across multiple AI racks up to 30 meters apart, with approximately 350 nanoseconds of near-NUMA latency and 7.2 Tbps of optical fabric bandwidth per memory module. Memory capacity can scale independently of compute, keeping warm KV caches closer to accelerators and reducing the data movement that limits token generation performance.

Senior Director of Product Management Ravi Mahatme details the architecture, the platform components, and why optical connectivity is becoming a foundational requirement for next-generation AI inference infrastructure: https://mrvl.co/4c2mds1

07/08/2026

07/08/2026

CXL adoption is moving from evaluation into real-world deployment across hyperscale environments, and memory capacity and bandwidth have emerged as critical constraints on AI performance and scaling.

Marvell offers the industry's most comprehensive CXL portfolio spanning three complementary technologies. Structera X memory expansion, developed in close collaboration with leading hyperscalers, enables larger, more flexible memory pools and extends the value of existing DDR4 investments. Structera A near-memory acceleration brings compute directly to the data, with 16 Arm Neoverse V2 cores and 200 GB/s of memory bandwidth reducing data movement and freeing host cores for other tasks. Structera S CXL switching enables rack-scale memory pooling across up to 48 TB of shared memory, with benchmarks showing up to 4.8x higher inference throughput and an 82.7% reduction in time-to-first-token in GPU configurations.

Together, the three form a cohesive architecture for disaggregated, composable memory at hyperscale. Associate Vice President Khurram Malik details the full portfolio and the infrastructure trajectory it is designed to support: https://mrvl.co/4z7SrfE

06/08/2026

PCIe Gen 6 storage demands more than raw bandwidth. Intelligent flash management, enterprise-grade security, advanced error correction, and scalable architecture are equally critical as data centers push the limits of NVMe SSD performance.

The Marvell Bravera SC6 SSD controller addresses all of these requirements in a single platform. It combines a configurable PCIe Gen 6 host interface with backward compatibility across five prior generations, support for SLC, MLC, TLC, and QLC NAND flash, sixth-generation NANDEdge LDPC error correction, 15 embedded processor cores for parallel storage management, and hardware-accelerated AES, SHA, RSA, and ECC cryptography.

Product Management Director RC Camillo details the architecture and what it means for enterprise and hyperscale storage deployments in the AI era: https://mrvl.co/4xgfRgN

06/08/2026

At , Marvell is demonstrating a PCIe Gen 6 NVMe SSD controller purpose-built for enterprise and hyperscale data center storage workloads, delivering next-generation performance and scalability for modern infrastructure.

Optimized to support today's rapidly evolving AI and data-intensive workloads, the demo features a high-capacity SSD solution with the Marvell® Bravera™ SC6 SSD Controller, supporting TLC or QLC NAND flash, featuring a host-managed flash translation layer (FTL) to improve efficiency, scalability, and performance.

Learn more: https://mrvl.co/4xgfRgN

06/08/2026

Stop by the Marvell booth at FMS to see Marvell's Structera S 60260, the industry's highest-radix PCIe 6.0 switch, delivering 256 lanes of connectivity to enable large-scale AI and accelerated computing infrastructure.

Designed for PCIe-based scale-up architectures, it provides a high-performance alternative to proprietary interconnects. Enhanced security and manageability through integrated features including a dual-core Arm® Cortex-A55 processor, secure boot, and a Caliptra Root of Trust.

Learn more: https://mrvl.co/3Sb6MqO

05/08/2026