MSI has unveiled its new CXL-based server platform at the Flash Memory Summit 2024 event in Santa Clara, California. The showcase took place at the Samsung and MemVerge booths from August 6-8. The MSI CXL memory expansion server is designed to boost the performance of In-Memory Databases, Electronic Design Automation, and High-Performance Computing applications.
It integrates the latest 4th Gen AMD EPYC processors with CXL technology to expand memory capacity and bandwidth. Danny Hsu, General Manager of Enterprise Platform Solutions at MSI, said, “In collaboration with key players like AMD, Samsung, and MemVerge, we are driving CXL technology to meet the demands of high-performance data center computing.
CXL Memory Expansion is an advanced technology that enhances both the bandwidth and capacity of system memory. The CXL 2.0 standard introduces a high-performance memory access mechanism, allowing expansion devices to interface directly with the CPU’s memory.
The newly unveiled MSI S2301 server, equipped for select 5th Gen AMD EPYC CPUs, significantly boosts application performance through its support for CXL 2.0 Memory Expansion. It enables users to add CXL-compliant memory expansion devices seamlessly, increasing the total accessible system memory and enhancing memory bandwidth. Samsung’s introduction of the 256GB CXL memory modules based on CXL 2.0 is expected to accelerate the commercialization of next-generation memory solutions.
Jangseok Choi, Vice President of the Memory New Business Planning Team at Samsung Electronics, stated, “Samsung, MSI, and AMD look forward to creating a robust ecosystem for the CXL market.
Charles Fan, CEO and Co-founder of MemVerge, added, “The availability of servers with qualified CXL peripherals and software is a breakthrough for the industry. It marks the beginning of our ability to meet the pent-up demand from IT organizations ready to evaluate the added memory capacity and bandwidth expected from CXL.”
This joint demonstration underscores the potential of CXL technology in significantly expanding both the total system memory capacity and bandwidth. The advancements are anticipated to have a substantial impact on the next-generation computing market.
Marvell Technology has announced it will showcase its industry-leading cloud storage and CXL memory technologies for AI, ML, and other data-intensive workloads at the Flash Memory Summit 2024. The event will take place from August 6–8 at the Santa Clara Convention Center. Marvell received the “Most Innovative Sustainability Technology” award for its X CXL Memory-Expansion Controller at the FMS awards ceremony on August 6.
The Structera X expanders, designed by Marvell, enable terabytes of memory to be added to general-purpose servers, addressing high-capacity memory applications such as in-memory databases. Attendees are encouraged to visit Marvell’s booth to view cloud-optimized accelerated storage, CXL, and memory demonstrations, including Structera CXL near-memory compute and accelerated storage for RAG and AI inference with Flexible data placement in a QLC SSD. The Structera CXL product line is designed to help cloud data center operators overcome scaling and memory performance challenges in general-purpose servers.
Structera CXL innovations unveiled at summit
It includes the industry’s first PCIe 5.0 SSD controller, known for its unprecedented performance, best-in-class efficiency, and leading security features for scalable, containerized cloud storage infrastructure. Marvell will also host the “Pizza, Beer and Chat with the Experts” event on Wednesday, August 7, featuring discussions on topics such as PCI Express and next-generation controllers.
The company’s participation in various sessions includes topics such as Flash & Memory Controller Technologies for AI, SSD Technologies for Compute Use Cases, Technologies for Improving the Endurance and Reliability of SSDs, Performance Optimization and Modeling Techniques for SSDs, CXL AI Implications, Power Optimization and Telemetry Features for SSDs, and New Form Factors and Interfaces for SSDs. Marvell has unveiled a new addition to its CXL memory expansion family, the Structera series, which features advancements aimed at modern server architectures. The new Structera CXL family includes both CXL Type-3 Memory expansion devices and a near-memory accelerator complete with up to 16 Arm Neoverse V2 cores.
The Structera series is divided into two main product lines: Structera A and Structera X. The A series focuses on accelerators while the X series targets memory expansion. Both series are designed as CXL 2.0 and PCIe Gen5 devices built on TSMC’s 5nm technology.
The Structera A 2504 stands out as a CXL 2.0 memory expansion device that can support up to four memory channels and includes 16 Arm Neoverse V2 cores, the same cores used in NVIDIA’s Grace processors. This device also features built-in LZ4 compression and decompression to maximize the memory capacity, which can go up to DDR5-6400. The Structera X 2404 addresses a significant industry challenge with DDR4 memory expansion.
This device can support up to four channels and allows three DDR4 DIMMs per channel, enabling significant memory scaling. This is particularly useful for hyper-scalers looking to recycle DDR4 memory from decommissioned servers, offering a cost-effective solution for extending the lifecycle of existing hardware. The X 2504 model is designed for DDR5 and supports up to DDR5-6400.
This device is intended for performance-focused applications, offering enhanced memory capacity and bandwidth. Marvell’s Structera series presents an innovative approach to memory expansion and processing power in modern server environments. The combination of CXL 2.0 capabilities with Arm Neoverse V2 cores offers a versatile solution for enhancing both memory capacity and computational performance.
These devices promise to deliver substantial benefits for data-intensive applications, particularly in AI and big data analytics. The market can expect sampling to begin in Q4 2024, and it will be interesting to see how these products perform in various configurations and workloads.
