DESIGN TOOLS
Memory

Powering the next generation of Supermicro Petascale platforms with Micron CZ120 memory expansion using CXL™

Luis Ancajas | May 2024

Today marks an expansion of our leading-edge product line-up for artificial intelligence (AI) and high-performance computing (HPC), as Micron announces the shipment of our CZ120 memory expansion module using Compute Express LinkTM (CXLTM) with the Supermicro X13 and H13 Petascale servers. For the past year, Supermicro and Micron have been collaborating on the next generation of Petascale platforms. By leveraging Micron’s CZ120 memory expansion module for memory capacity and bandwidth expansion, we are fueling mission-critical workloads, with faster and more efficient data access. This milestone positions both companies at the forefront of the industry, accelerating the adoption of CXL solutions by data center operators to tackle their growing memory challenges stemming from existing data-intensive workloads and newer AI and machine learning (ML) workloads.

Micron CZ120 memory module using CXL 

CXL: Breaking the memory wall for tomorrow’s data-intensive workloads of AI, high-performance compute, in-memory databases, and general-purpose compute
 

With existing server infrastructure struggling to meet the needs of data-intensive workloads, CXL offers a new architecture to break through legacy data bottlenecks. The Petascale platform from Supermicro and CZ120 memory expansion modules using CXL from Micron provide the needed building blocks to solve this challenge: 

  • Memory capacity expansion: With 4 CZ120 256GB modules, up to 1TB capacity can be added to a Petascale server. 
  • Bandwidth expansion: Each CZ120 module can add up to 36GB/s data transfer bandwidth, equivalent to the bandwidth of a DDR5 RDIMM operating at a speed of 4800MT/s. Server performance improvements are obtained by adding memory capacity and/or bandwidth. An analysis of the combination of the Petascale platform and CZ120 CXL memory modules is provided across multiple workloads in this Micron white paper: CXL Memory Expansion: A Closer Look on Actual Platform (micron.com).

By testing the combined products for interoperability, compatibility and across operating systems, the Supermicro Petascale server with Micron CZ120 memory expansion modules are being integrated by customers into their data centers. The Petascale and CZ120 combination is shipping now to multiple customers in medium to large data centers and high-performance computing centers.  Initial use-cases include: 

  • Storing databases entirely in memory for increased throughput (such as SQL databases  and vector databases) 
  • Storing the key-value cache used in transformer models for accelerated inference 
  • Accelerating hypervisor-centric workloads 
  • Using CPU-only inferencing schemes with the latest AI instruction sets 

With Micron and Supermicro taking the lead to enable a joint solution, data center operators now have architectural options for memory-intensive workloads. Contact our teams to explore how Petascale servers populated with Micron CZ120 can help address your data center challenges. For more info, visit CZ120 CXL TEP or the Supermicro Petascale website.

References:

Supermicro 1U Petascale Storage Server with 8 E3.S NVMe PCIe Gen 5 SSD bays and 4 CXL E3.S bays, AMD EPYC™ 9004 Processor.  (Model ASG-1115S-NE3X12R).

Supermicro 1U Petascale Storage Server with 8 E3.S NVMe PCIe Gen 5 SSD bays and 4 CXL E3.S bays, dual 5th Gen Intel® Xeon® Scalable Processors. (Model SSG-121E-NE3X12R). 

Director of CXL Product Management

Luis Ancajas

He is responsible for the partnerships with the hypervisor and independent software vendors partnerships and contributes to the CXL go to market strategy and execution, as well as to the overall CXL product strategy.

Luis has held executive roles in sales, business development, strategic marketing as well as having an engineering background in SoC and CPU architecture and cybersecurity. He has an MSEE degree in Computer Architecture from Stanford University, and BS degree in EECS from University of California, Berkeley.