Microsoft's first generation memory pooling with CXL can significantly reduce the amount of DRAM needed for servers. Up to 25% of DRAM is stranded, and 50% of VMs never touch 50% of their memory. Reducing overprovisioning and underutilization with CXL pooled memory "represents hundreds of millions of dollars in cost savings for a large cloud provider."
Hmm, I have to think some more about how this dynamic will change competitive positioning.. But it's certainly going Jensen's way, with his "the data center as the unit of compute".
Thank you for the post. It seems that going forward public cloud is going to win on differentiated hardware compared to private cloud. Their scale not only brings basic cost advantage but also supports optimization for niche workloads.
Liked it enough, I linked it here:
https://www.libertyrpf.com/i/64621760/cxl-next-gen-interconnect-fabric-to-pool-huge-amounts-of-memory-in-data-centers-
Have a good weekend!
That's really cool tech, thanks for highlighting! 💚 🥃
The software around migrating workloads to near and far memory is going to be the big sticking point and advantage for big clouds versus smaller ones
Hmm, I have to think some more about how this dynamic will change competitive positioning.. But it's certainly going Jensen's way, with his "the data center as the unit of compute".
Question is if the big cloud guys are willing to give up control, if they just have customers demand and Nvidia section with their solutions, or what.
Thank you for the post. It seems that going forward public cloud is going to win on differentiated hardware compared to private cloud. Their scale not only brings basic cost advantage but also supports optimization for niche workloads.