Home Update Samsung Unveils CXL Memory Module Box: Up to 16 TB at 60…

Samsung Unveils CXL Memory Module Box: Up to 16 TB at 60…

94
Samsung Unveils CXL Memory Module Box: Up to 16 TB at 60...


Composable disaggregated knowledge heart infrastructure guarantees to vary the best way knowledge facilities for contemporary workloads are constructed. However, to totally notice the potential of recent applied sciences, similar to CXL, the business wants brand-new {hardware}. Recently, Samsung launched its CXL Memory Module Box (CMM-B), a tool that may home as much as eight CXL Memory Module – DRAM (CMM-D) units and add loads of reminiscence linked utilizing a PCIe/CXL interface.

Samsung’s CXL Memory Module Box (CMM-B) is the primary machine of this sort to accommodate as much as eight 2 TB E3.S CMM-D reminiscence modules and add as much as 16 TB of reminiscence to as much as three trendy servers with applicable connectors. As far as efficiency is anxious, the field can provide as much as 60 GB/s of bandwidth (which aligns with what a PCIe 5.zero x16 interface provides) and 596 ns latency. 

From a pure efficiency standpoint, one CXL Memory Module—Box is slower than a dual-channel DDR5-4800 reminiscence subsystem. Yet, the unit remains to be significantly quicker than even superior SSDs. At the identical time, it gives very respectable capability, which is usually simply what the physician ordered for a lot of functions.

The Samsung CMM-B is suitable with the CXL 1.1 and CXL 2.zero protocols. It consists of a rack-scale reminiscence financial institution (CMM-B), a number of utility hosts, Samsung Cognos administration console software program, and a top-of-rack (ToR) swap. The machine was developed in shut collaboration with Supermicro, so anticipate this server maker to supply the product first.

Samsung’s CXL Memory Module – Box is designed for functions that want a whole lot of reminiscence, similar to AI, knowledge analytics, and in-memory databases, albeit not always. CMM-B permits the dynamic allocation of needed reminiscence to a system when it wants this reminiscence after which makes use of DRAM with different machines. As a consequence, operators of datacenters can spend cash on procuring costly reminiscence (16 TB of reminiscence prices rather a lot), scale back energy consumption, and add flexibility to their setups.

Source: Samsung



Source

LEAVE A REPLY

Please enter your comment!
Please enter your name here