Skip to content
AIPC.dealsAI PCs for real workloads.

Apple Mac Studio (M5 Ultra)

Apple M5 Ultra · Agent Computer

The Apple Mac Studio (M5 Ultra) is Apple's desktop built on the Apple M5 Ultra, and it gives a local model about 72GB of usable AI memory at 1200 GB/s. No verified price is currently on record, so the price block below shows its status rather than a number. It scores 43.9 overall on the AIPC Score.

Price and memory shown are for the 96GB / 1TB configuration.

AIPC Score: 43.9

  • AI memory52.8
  • AI compute26.2
  • Memory bandwidth66.7
  • Model compatibility69
  • GPU capability26.2
  • NPU capability32
  • Software support80
  • Power efficiency26.7
  • Expandability15
  • Price/value16.4
How this score is calculated →

Specifications

AI Memory
72GB
CPU
Apple M5 Ultra (36-core)
GPU
Apple M5 Ultra 80-core GPU
NPU
Neural Engine (32 TOPS)
Memory bandwidth
1200 GB/s
Storage
1TB
Configuration
96GB / 1TB
Type
Desktop

How we calculate AI memory: About 75% of unified memory, the macOS default wired limit. Raisable with iogpu.wired_limit_mb. Source →

Configurations

ConfigurationAI MemoryMemory bandwidthPriceRetailer
96GB / 1TB72GB1200 GB/sPrice awaiting re-verificationApple
256GB / 2TB192GB1200 GB/sNo verified price—
512GB / 4TB384GB1200 GB/sNo verified price—

What can this PC run?

ModelQuantizationFitsModeStateSpeed
DeepSeek-R1-Distill-Qwen-7BQ4YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Qwen-7BQ8YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Qwen-14BQ4YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Qwen-14BQ8YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Qwen-32BQ4YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Qwen-32BQ8YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Llama-70BQ4YesFull memoryCalculatedPerformance not yet benchmarked
DeepSeek-R1-Distill-Llama-70BQ8LimitedOffload requiredCalculatedPerformance not yet benchmarked
Gemma 3 4B ITQ4YesFull memoryCalculatedPerformance not yet benchmarked
Gemma 3 4B ITQ8YesFull memoryCalculatedPerformance not yet benchmarked
Gemma 3 12B ITQ4YesFull memoryCalculatedPerformance not yet benchmarked
Gemma 3 12B ITQ8YesFull memoryCalculatedPerformance not yet benchmarked
Gemma 3 27B ITQ4YesFull memoryCalculatedPerformance not yet benchmarked
Gemma 3 27B ITQ8YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.2 3B InstructQ4YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.2 3B InstructQ8YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.1 8B InstructQ4YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.1 8B InstructQ8YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.1 70B InstructQ4YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.1 70B InstructQ8LimitedOffload requiredCalculatedPerformance not yet benchmarked
Llama 3.3 70B InstructQ4YesFull memoryCalculatedPerformance not yet benchmarked
Llama 3.3 70B InstructQ8LimitedOffload requiredCalculatedPerformance not yet benchmarked
Llama 4 Scout 17B-16E InstructQ4LimitedOffload requiredCalculatedPerformance not yet benchmarked
Mistral Small 3 (Mistral-Small-24B-Instruct-2501)Q4YesFull memoryCalculatedPerformance not yet benchmarked
Mistral Small 3 (Mistral-Small-24B-Instruct-2501)Q8YesFull memoryCalculatedPerformance not yet benchmarked
Devstral Small 1.1 (Devstral-Small-2507)Q4YesFull memoryCalculatedPerformance not yet benchmarked
Devstral Small 1.1 (Devstral-Small-2507)Q8YesFull memoryCalculatedPerformance not yet benchmarked
Phi-4Q4YesFull memoryCalculatedPerformance not yet benchmarked
Phi-4Q8YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-4BQ4YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-4BQ8YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-8BQ4YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-8BQ8YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-14BQ4YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-14BQ8YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-30B-A3BQ4YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-30B-A3BQ8YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-32BQ4YesFull memoryCalculatedPerformance not yet benchmarked
Qwen3-32BQ8YesFull memoryCalculatedPerformance not yet benchmarked
gpt-oss-20bQ4YesFull memoryCalculatedPerformance not yet benchmarked
gpt-oss-20bQ8YesFull memoryCalculatedPerformance not yet benchmarked
gpt-oss-120bQ4LimitedOffload requiredCalculatedPerformance not yet benchmarked
Whisper large-v3PublishedYesFull memoryCalculatedPerformance not yet benchmarked
FLUX.1 schnellPublishedYesFull memoryCalculatedPerformance not yet benchmarked
FLUX.1 devPublishedYesFull memoryCalculatedPerformance not yet benchmarked
Stable Diffusion XL Base 1.0PublishedYesFull memoryCalculatedPerformance not yet benchmarked
Stable Diffusion 3.5 LargePublishedYesFull memoryCalculatedPerformance not yet benchmarked
HunyuanVideoPublishedYesFull memoryCalculatedPerformance not yet benchmarked
Wan2.1 T2V 14BPublishedLimitedOffload requiredCalculatedPerformance not yet benchmarked
Wan2.2 T2V A14BPublishedLimitedOffload requiredCalculatedPerformance not yet benchmarked
Does not fit (16)

Alternatives