Mehmet Özgen ÖZDOĞAN·2d agoBuilding an exo-Style Distributed LLM Cluster on Ubuntu Server VMs With Mixed GPUs: Architecture…You have a few Ubuntu Server VMs. Each one has a different GPU attached through passthrough: maybe a 24 GB card in one, a 16 GB card in…
Mehmet Özgen ÖZDOĞAN·Oct 5What’s Actually New in Local AI Right Now: Distributed Clusters, Smarter Serving, and Router-Based…I try to check in on the local AI space every couple of weeks because honestly it moves fast enough that a month of not paying attention…
Nexus Studio·Aug 30128 GB + 48 GB ≠ one 176 GB Mac: My EXO Cluster ExperimentBenchmarks on an M4 Max Mac Studio and M4 Pro Mac mini over Thunderbolt 5 RDMA show why aggregate unified memory is not pooled memory.
