Can 96GB Mac Studio run Qwen3.8? Analyzing SSD offload feasibility
Mxmtm · reddit · 2026-08-27
A user plans to buy a 96GB Mac Studio to run the 125B Qwen3.8-Flash-Next, relying heavily on llama.cpp's PLE/n-gram SSD offload feature. The post details memory calculations (Model 103.8 GiB + KV Cache 8 GiB), concluding that without offload, 128GB RAM is required, but with it, the 96GB model (84 GiB usable) might suffice. The user discusses the current implementation status (uncertain Metal I/O support), limitations of MLX, and refutes pessimistic claims of 1GB/token traffic, estimating actual load at 256KB/token.
More from Infra
- ASML CFO says US Match Act restricting DUV sales to China would backfire — pstAsiatech · 2026-10-15
- Bloomberg breakdown: a 1GW Texas AI data center may need nearly $3B for overruns alone — rohanpaul_ai · 2026-10-15
- 7 System Design Algorithms to Know: Merkle Trees to CAP/PACELC — goyalshaliniuk · 2026-10-15
- MiniCPM-V 4.7 runs a 35B MoE multimodal model on phones with just 3GB memory — nevrekaraishwa2 · 2026-10-15
- Sipeed MaixStation product page goes live for edge AI dev — xlltt · 2026-10-15
- Nvidia Is Now Worth More Than the Entire Indian Stock Market — theteknosaur · 2026-10-15