As you know CXL is an interconnect system which maintains the memory coherency between the inner memory of attached devices, accelerators, and the cpu’s local memory. CXL 2.0 is in fact an expansion of the first version, indeed it is possible to have a single external device configured for multiple host accesses. This is in fact an evolution of pooling, where the device can now instead share the work with many hosts accelerating the whole cloud thing.
Links
https://www.rambus.com/blogs/compute-express-link/
