Questions from the Webinar: Introducing the Compute Express Link™ 2.0 Specification
Updated: Apr 9, 2021
By: Ahmad Danesh and Debendra Das Sharma
Last December, we presented a webinar titled “Introducing the Compute Express Link™ 2.0 Specification.” The presentation explored the new features and usage models of the CXL 2.0 specification including switching and pooling, hot-plug, fabric manager API, persistent memory, security and more. The webinar recording is available on BrightTALK and YouTube. Also, the presentation is available for download here.
We received great questions during the Q&A portion of the webinar and we couldn’t answer them all during the Webinar. Additional questions we received from attendees during the live webinar are answered below.
Q: What are your thoughts on extending CXL over a network (e.g. ethernet) especially for Storage Class Memory (SCM)? What are the major concerns?
A: If CXL is transported over a network, we would need to address a much larger scope for CXL such as provide the ability to perform recovery in the presence of node or link failures. We would also have to deal with the implications when increasing the latency. This was beyond the scope for the CXL 2.0 specification.
Q: For a CXL 2.0 switch fabric, what is the maximum number of hosts and devices?
A: There is no theoretical limit other than restricting a Multiple Logical Device (MLD) Type-3 device that can support a maximum of 16 hosts. The practical limits would be the number of Lanes and Links a switch can support.
Q: I understand that CXL 2.0 does not specify cascaded switch support, but does the specification inhibit such configuration?
A: We do not prohibit such a configuration as long as it can work within the specification.
Q: Are there any Physical Layer changes for CXL 2.0?
A: There are minor enhancements to the auto-negotiation flow for CXL 2.0 features and retimer detection to enable full backward compatibility.
Q: Does CXL 2.0 support degraded mode (8 GT/s) speed?
A: CXL 2.0, like CXL 1.0 and 1.1, supports an 8.0 GT/s data rate in degraded mode.
Q: For the hot-plug removal of the memory device, how is the data wiped out for other servers or users? Especially when Persistent Memory is used for Pooling?
A: Just like any hot-pluggable PCIe SSD, the device will have its protection mechanism to ensure that the persistent data cannot be accessed by unauthorized users such as Passphrase. While running the link, the LDID provides the host ID and it will use that information to ensure data protection from other hosts.
Q: Is the CXL switch the same as a PCIe 5 Switch? For Pooling, will that be an external PCIe 5 Switch?
A: The CXL switch is a superset of the PCIe switch. A CXL switch can act as a PCIe switch. However, a native PCIe switch cannot act as a CXL switch since the CXL.cache and CXL.memory protocols, as well as the CXL encoding mechanisms, would be missing.
About CXL Consortium
Q: Will the slides be available for download?
A: The presentation is available for download in CXL’s Resource Library.
Q: Where can I download the evaluation copy of the CXL 2.0 specification?
A: You can download the CXL 2.0 specification here.
Compute Express Link™ and CXL™ Consortium are trademarks of the Compute Express Link Consortium.