MSI has launched a compute express link (CXL) memory expansion 2U server platform optimized for data-intensive workloads and in-memory databases that can support up to a staggering 18TB DDR5 RAM.
The S2302 has a dual-socket 4th-Gen Xeon Scalable processor and 32 DDR5 DIMM slots alongside five PCIe 5.0 x16 slots. With eight E3.S CXL drive bays in addition to two ES3.S PCIe 5.0 SSD slots, the system can boost its memory capacity from a default of 6TB by an additional 12TB.
MSI launched this alongside a less powerful S1301 server, which can offer up to 11TB of extended memory capacity and features dual AMD EPYC 9004 Series CPUs. This iteration also features 24 DDR5 DIMM slots and two PCIe 5.0 x16 slots.
The power of CXL
Revealed at SC23, MSI unveiled several systems including GPU alongside CXL memory expansion servers, with the former featuring liquid cooling.
The G4101 4U server is designed for AI inference computer workloads, for example, and features one AMD EPYC 9004 CPU alongside a liquid cooling module. The G3101, meanwhile, is a 3U single-socket platform that can fit up to four GPUs, and is powered with an AMD EPYC 7003 processor, also featuring a liquid cooling module.
Demand for products like the S2302 and S1301, meanwhile, has grown in recent years, with the CXL standard proving a solid high-speed and low-latency bridge between CPUs and various expansion systems. Through CXL, servers can benefit from memory expansion and the addition of components like accelerators to seriously augment performance.
Systems like S2302, in particular, may be particularly useful for organizations in need of additional memory to support their systems that are running AI inference and machine learning workloads. It can also be used for high-performance computing (HPC0, virtualization and CXL memory expansion and pooling.
There are other finished products and prototypes on the market, including this remarkable CXL 2.0 switch concept developed by FADU and demonstrated at FMS 2023. The device can be plugged into even a laptop to give it a massive memory boost, but it's likely to be most useful by servers that are charged with handling next-gen workloads like AI.
More from TechRadar Pro
- We've rounded up the best DDR5 RAM units
- DDR5 vs DDR4 RAM: The differences explained
- What is non-binary memory: the new RAM configurations explained
