Running Kimi K3 on my desk? It will require 4 x 512GB M3 Ultra Mac Studios (2TB
Source
alexocheema
Author
alexocheema
Date
Why it matters
If you are sizing hardware for local frontier-scale MoE models, this gives ballpark memory, bandwidth, and throughput figures plus the interconnect technique (RDMA over TB5) that makes multi-Mac inference viable.
Running Kimi K3 on my desk?
It will require 4 x 512GB M3 Ultra Mac Studios (2TB @ 3.2TB/s).
Waiting on active parameter count, but based on the rumours I expect we can run it at 30+ tok/sec with MTP + Tensor Parallelism using RDMA over Thunderbolt 5.
Prefill will be slow but https://t.co/JiTRtCMGhn
Transcript
Running Kimi K3 on my desk?
It will require 4 x 512GB M3 Ultra Mac Studios (2TB @ 3.2TB/s).
Waiting on active parameter count, but based on the rumours I expect we can run it at 30+ tok/sec with MTP + Tensor Parallelism using RDMA over Thunderbolt 5.
Prefill will be slow but https://t.co/JiTRtCMGhn