Skip to content
Run LLM Locally

Mac · Apple Silicon

M2 Ultra with 128 GB: which local LLMs run?

With 128 GB of unified memory, the M2 Ultra gives the GPU about 96 GB by default (800 GB/s): enough for models up to about gpt-oss-120b.

  • Unified memory128 GB
  • GPU usable~96 GB
  • Bandwidth800 GB/s
  • GPU cores76
ShareFacebookXWhatsAppLinkedInEmail

Unified memory is not VRAM: macOS reserves part of it. The limit can be raised with sudo sysctl iogpu.wired_limit_mb=…, at the cost of memory for apps. There is no offload to system RAM — it is the same memory.

CPU cores
24
GPU cores
76
Bandwidth
800 GB/s
Memory configurations
64 GB · 128 GB · 192 GB
Sold in
Mac Studio, Mac Pro
Released
June 2023

Official specData updated 24 days agoSource: Apple tech specs

What the M2 Ultra with 128 GB runs

M2 Ultra · 128 GB · 8K
gpt-oss-120bMXFP4 · 8K~63 GB · 68–83 tok/s measuredExcellent
Mixtral 8x7B InstructQ8_0 · 8K~50 GB · 26–39 tok/s estimateExcellent
Qwen3 32BQ8_0 · 8K~36.8 GB · 14–21 tok/s estimateExcellent
DeepSeek R1 Distill Qwen 32BQ8_0 · 8K~36.8 GB · 14–21 tok/s estimateExcellent
Qwen2.5 32B InstructQ8_0 · 8K~36.8 GB · 14–21 tok/s estimateExcellent
Qwen2.5 Coder 32B InstructQ8_0 · 8K~36.8 GB · 14–21 tok/s estimateExcellent
QwQ 32BQ8_0 · 8K~36.8 GB · 14–21 tok/s estimateExcellent
Qwen3 30B-A3BQ8_0 · 8K~33.2 GB · 50–88 tok/s estimateExcellent
Qwen3 Coder 30B-A3B InstructQ8_0 · 8K~33.2 GB · 50–88 tok/s estimateExcellent
Gemma 3 27B ITQ8_0 · 8K~30 GB · 17–25 tok/s estimateExcellent
Gemma 2 27B ITQ8_0 · 8K~31.4 GB · 17–25 tok/s estimateExcellent
Mistral Small 3.2 24B InstructQ8_0 · 8K~26.4 GB · 19–29 tok/s estimateExcellent
gpt-oss-20bMXFP4 · 8K~12.8 GB · 100–122 tok/s measuredExcellent
Qwen3 14BQ8_0 · 8K~17.3 GB · 29–45 tok/s estimateExcellent
DeepSeek R1 Distill Qwen 14BQ8_0 · 8K~17.6 GB · 28–45 tok/s estimateExcellent
Qwen2.5 14B InstructQ8_0 · 8K~17.6 GB · 28–45 tok/s estimateExcellent
Phi-4 14BQ8_0 · 8K~17.4 GB · 29–45 tok/s estimateExcellent
Mistral Nemo 12B InstructQ8_0 · 8K~14.7 GB · 34–54 tok/s estimateExcellent
Gemma 3 12B ITQ8_0 · 8K~13.9 GB · 34–57 tok/s estimateExcellent
Gemma 2 9B ITQ8_0 · 8K~12.5 GB · 41–68 tok/s estimateExcellent
Qwen3 8BQ8_0 · 8K~10.3 GB · 47–79 tok/s estimateExcellent
Llama 3.1 8B InstructQ8_0 · 8K~10 GB · 49–81 tok/s estimateExcellent
DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10 GB · 49–81 tok/s estimateExcellent
Qwen2.5 7B InstructQ8_0 · 8K~9 GB · 54–89 tok/s estimateExcellent
Mistral 7B Instruct v0.3Q8_0 · 8K~9 GB · 54–90 tok/s estimateExcellent
Gemma 3 4B ITQ8_0 · 8K~5.2 GB · 87–172 tok/s estimateExcellent
Qwen3 4BQ8_0 · 8K~6 GB · 78–151 tok/s estimateExcellent
Phi-4 Mini Instruct 3.8BQ8_0 · 8K~5.8 GB · 84–160 tok/s estimateExcellent
Llama 3.2 3B InstructQ8_0 · 8K~4.9 GB · 99–190 tok/s estimateExcellent
Qwen3 1.7BQ8_0 · 8K~3.5 GB · 140–321 tok/s estimateExcellent
Llama 3.2 1B InstructQ8_0 · 8K~2.3 GB · 231–508 tok/s estimateExcellent
Qwen3 0.6BQ8_0 · 8K~2.3 GB · 205–666 tok/s estimateExcellent
Llama 4 Scout 17B-16E InstructQ6_K · 8K~88.8 GB · 24–36 tok/s estimateCompatible
Qwen2.5 72B InstructQ8_0 · 8K~79.1 GB · 6.5–9.5 tok/s estimateCompatible
Llama 3.3 70B InstructQ8_0 · 8K~76.6 GB · 6.5–9.5 tok/s estimateCompatible
DeepSeek R1 Distill Llama 70BQ8_0 · 8K~76.6 GB · 6.5–9.5 tok/s estimateCompatible
Qwen3 235B-A22BQ3_K_M · 8K~112 GB · No dataCompatible with limits
DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No dataIncompatible

For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.

EstimateData updated 24 days agoSource: Run LLM Locally engine