REDDIT · REDDIT// 15h agoINFRASTRUCTURE

Apple M5 Max doubles LLM prompt processing speeds

LocalLLaMA users evaluate upgrading from the M1 Max to the newly released M5 Max. The consensus reveals the biggest gains lie in massive unified memory capacity and faster prefill speeds rather than raw token generation.

// ANALYSIS

The M5 Max's architectural shift toward GPU-integrated Neural Accelerators makes it a compelling upgrade for heavy RAG workloads, though memory bandwidth remains the bottleneck for generation speed.

–Generation speed sees linear improvements (roughly 3x over M1 Max) due to memory bandwidth limits
–Prefill speeds double compared to the M4 Max, making long-context processing significantly faster
–The true value lies in supporting up to 192GB of unified memory, unlocking 70B+ parameter models locally

// TAGS

m5-maxapple-siliconllminferencegpu

DISCOVERED

15h ago

2026-04-11

PUBLISHED

19h ago

2026-04-11

RELEVANCE

8/ 10

AUTHOR

br_web