Search
Close this search box.

“Introducing the CXL 3.1 Specification” Webinar Q&A Recap

1 min read

The CXL® Consortium recently hosted a webinar covering the CXL 3.1 specification, which builds on previous iterations to deliver improved fabric manageability, optimized resource utilization, and extended memory sharing and pooling. During the webinar, Mahesh Wagh, CXL Consortium Technical Task Force Co-Chair, and Rob Blankenship, Protocol Working Group Co-Chair, introduced the new features and use cases enabled by CXL 3.1 and answered audience questions about the usage models it will enable

 

If you were not able to attend the live webinar, the recording is available via YouTube and the webinar presentation slides are available for download on the CXL Consortium Website. We received great questions from the audience during the live Q&A but were not able to address them all during the webinar. Below, we’ve included answers to the questions we didn’t get to during the live webinar.

 

Q: What do you mean by CXL type 1 device supporting CXL.mem protocol (P2P)? Will a CXL type 1 device in this case have CXL memory inside it?

A CXL Type-1 device in this context can generate Direct P2P.mem accesses targeting a Type-3 Memory expander. As background, CXL traditionally defines CXL Type devices as:

  • Type-1 is a CXL.cache source and CXL.io
  • Type-2 is a CXL.cache source and CXL.mem target and CXL.io
  • Type-3 is a CXL.mem target and CXL.io.

 

Additionally, in this context, the new “Direct P2P CXL.mem” can be sourced from an accelerator device that is implemented as any of the types described as the traditional types don’t define CXL.mem as a source.

 

Q: What is the impact of the metadata on performance with CXL 3.1?

Messages carrying Extended Metadata will see a reduced efficiency. The CXL link layer allows for up to 3 Extended Metadata to be packed together in a 16B slot. In the case of CXL.mem 100% DRS (read response), we would see a data efficiency reduction of 83% to 77% (5-7% reduction) when all DRS carry the additional information. Note that the host can use extended metadata selectively, so efficiency loss only applies when the requester needs the additional information.

Facebook
Twitter
LinkedIn