Guide
Mac vs PC for a local LLM
A Mac spends one pool of memory on the model, the KV cache and macOS. A PC GPU has a separate VRAM pool, and anything that spills goes through system RAM. The catalog can say which pool is large enough. It cannot say which box feels faster, because we did not measure that.
Two different reserves
On a discrete GPU the formula subtracts 1.2 GB per card. On a Mac it subtracts 6 GB at 16 GB, 8 GB at 32 GB, 8 GB at 36 GB, 10 GB at 64 GB and 12 GB above that. A PC can also offload into system RAM. A Mac cannot: unified memory is the whole pool.
Apple figures that have a URL
- Mac Studio M4 Max 36GB lists 36 GB and 410 GB/s, with bandwidthSourceUrl https://support.apple.com/en-us/122211. Mac Studio M4 Max 36GB: $2,116 estimate, €1,890 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked used. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac Studio M4 Max 128GB lists 128 GB and 546 GB/s, with bandwidthSourceUrl https://support.apple.com/en-us/122211. Mac Studio M4 Max 128GB: $3,683 estimate, €3,290 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked used. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac mini M6 16GB lists 16 GB and 153 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-mini/specs/. Mac mini M6 16GB: $894 estimate, €799 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac mini M6 32GB lists 32 GB and 170 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-mini/specs/. Mac mini M6 32GB: $1,398 estimate, €1,249 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac mini M5 Pro 64GB lists 64 GB and 307 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-mini/specs/. Mac mini M5 Pro 64GB: $3,123 estimate, €2,790 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac Studio M5 Max 36GB lists 36 GB and 460 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-studio/specs/. Mac Studio M5 Max 36GB: $2,900 estimate, €2,590 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac Studio M5 Max 128GB lists 128 GB and 614 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-studio/specs/. Mac Studio M5 Max 128GB: $5,810 estimate, €5,190 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac Studio M5 Ultra 96GB lists 96 GB and 1200 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-studio/specs/. Mac Studio M5 Ultra 96GB: $7,266 estimate, €6,490 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
- Mac Studio M5 Ultra 256GB lists 256 GB and 1200 GB/s, with bandwidthSourceUrl https://www.apple.com/mac-studio/specs/. Mac Studio M5 Ultra 256GB: $11,744 estimate, €10,490 incl. VAT. Global prices (USD), EUR available. Dated 2026-10-14. Marked new. Global prices (USD), EUR available. EUR includes VAT. USD is an FX estimate. No shop listing was checked. sourceUrl is null.
Mac mini M4 16 GB and Mac mini M4 Pro 64 GB are still in the catalog for the used market. Their bandwidth fields are marked estimated and have no source URL. Do not quote those GB/s figures as Apple specs.
The same 70B question on both sides
- Mac mini M5 Pro 64GB: Llama 3.3 70B Q4 is a full fit, Qwen2.5 32B Q4 is a full fit. Mac mini M5 Pro 64GB has 54.0 GB usable after the reserve in this formula (catalog memory 64 GB, unified).
- Mac Studio M4 Max 128GB: Llama 3.3 70B Q4 is a full fit, Qwen2.5 32B Q4 is a full fit. Mac Studio M4 Max 128GB has 116.0 GB usable after the reserve in this formula (catalog memory 128 GB, unified).
- PC with RTX 4090 24GB: Llama 3.3 70B Q4 is offload only, not a full fit, Qwen2.5 32B Q4 is a full fit. PC with RTX 4090 24GB has 22.8 GB usable after the reserve in this formula (catalog memory 24 GB, vram).
- Dual RTX 3090, 48 GB: Llama 3.3 70B Q4 is a full fit, Qwen2.5 32B Q4 is a full fit. Dual RTX 3090, 48 GB has 45.6 GB usable after the reserve in this formula (catalog memory 48 GB, vram).
Dual RTX 3090, 48 GB lists 1872 GB/s as two catalog card figures added together. bandwidthSourceUrl is null. The speed estimate then multiplies by 0.8. Mac mini M5 Pro 64GB points at https://www.apple.com/mac-mini/specs/ for 307 GB/s. A lower sourced bandwidth and a higher unsourced one are not a bake-off.
GPU builds in the catalog carry a TDP field (the dual 3090 build lists 700 W). Mac rows leave TDP null. This page does not invent a watt or decibel number for a Mac.
Tokens per second on the calculator are estimates from the methodology page. benchmarkSourceUrl stays null. This guide does not add a measured llama-bench result.
Compare them in the calculator: 70B Q4, Apple only, 70B Q4, NVIDIA only, best Mac for a local LLM, methodology.
Related: How much VRAM does a 70B model need?, Best GPU for local LLMs in 2026, Running DeepSeek and Qwen locally.