Meta and AMD have partnered to introduce a form of memory that can seamlessly integrate into servers using Compute Express Link (CXL) technology. By leveraging this standard for high speed connections between processors and devices CXL memory offers the potential for more efficient memory utilization resulting in cost savings and improved performance. During the Open Compute Project (OCP) Global Summit 2023 a demo board was showcased featuring an AMD EPYC 9004 Genoa processor surrounded by four dual in line memory module (DIMM) slots, a heat sink and a fan. Notably the board included a PCIe x16 connector as reported by Serve the Home.
What makes this CXL 2.0 memory expansion demo unique is its foundation on an AMD chip of the traditional Intel Xeon chip based on Sapphire Rapids architecture. An important aspect of CXL memory is its ability to repurpose existing DDR4 or DDR5 RAM through memory controllers, which can lead to significant cost savings for hyperscalers who manage data centers during the transition from DDR4, to DDR5 RAM.The implementation of CXL memory provides an option for businesses seeking to repurpose and enhance older DDR4 RAM modules in combination with newer setups.
This approach effectively prolongs the lifespan of hardware offering a cost efficient alternative, to a complete scale upgrade. There are configurations available that can be explored for this purpose. Intels chip along with the Astera Labs CXL memory expansion card was showcased at SC22 as part of development, in CXL memory. FADU also demonstrated their Apollo CXL 20 switch, which aims to enhance server memory over CXL 2.0 by reducing latency and power consumption while increasing memory capacity.
