Configure a rig →
AppleAppleMac Studiounified memory

Mac Studio M2 Ultra (192GB) for local AI

Mac Studio M2 Ultra (192GB) provides 192 GB of VRAM for local AI. In the LocalIA catalog, 233 out of 242 models run comfortably on a single card.

View all compatible models →Rig around the (192GB) ↗
VRAM
192GB
Category
Apple
Series
Mac Studio
Vendor
Apple

Models that run comfortably

These models fit in 192 GB with room for context and stable inference.

DeepSeek V2deepseek148.4 GBcomfortableQ4 · / 192 GB
DeepSeek Coder V2deepseek148.4 GBcomfortableQ4 · / 192 GB
★Qwen 3 235B A22Bqwen147.7 GBcomfortableQ4 · / 192 GB
★Qwen3 235B A22Bqwen147.7 GBcomfortableQ4 · / 192 GB
Falcon 180Bfalcon138.3 GBcomfortableQ5 · / 192 GB
Mixtral 8x22Bmistral157.6 GBcomfortableQ8 · / 192 GB
★Mistral Large 123Bmistral137.5 GBcomfortableQ8 · / 192 GB
★NVIDIA Nemotron 3 Super 120B A12B BF16nemotron134.1 GBcomfortableQ8 · / 192 GB
★Llama 4 Scout 17Bx16llama121.8 GBcomfortableQ8 · / 192 GB
★Command R+ 104Bcommand116.2 GBcomfortableQ8 · / 192 GB
★Qwen3 Next 80B A3B Instructqwen89.4 GBcomfortableQ8 · / 192 GB
★Qwen 2.5 72Bqwen160.9 GBcomfortableFP16 · / 192 GB
Qwen 2.5 VL 72Bqwen160.9 GBcomfortableFP16 · / 192 GB
★Qwen2.5 72B Instructqwen160.9 GBcomfortableFP16 · / 192 GB
Llama 2 70Bllama156.5 GBcomfortableFP16 · / 192 GB
Llama 3 70Bllama156.5 GBcomfortableFP16 · / 192 GB
Llama 3.1 70Bllama156.5 GBcomfortableFP16 · / 192 GB
★Llama 3.3 70Bllama156.5 GBcomfortableFP16 · / 192 GB
CodeLlama 70Bcodellama156.5 GBcomfortableFP16 · / 192 GB
★DeepSeek R1 Distill 70Bdeepseek156.5 GBcomfortableFP16 · / 192 GB
Hermes 3 70Bhermes156.5 GBcomfortableFP16 · / 192 GB
★Llama 3.1 Nemotron 70Bnemotron156.5 GBcomfortableFP16 · / 192 GB
Athene 70Bathene156.5 GBcomfortableFP16 · / 192 GB
★Llama 3.3 70B Instructllama156.5 GBcomfortableFP16 · / 192 GB
★Llama 3.1 70B Instructllama156.5 GBcomfortableFP16 · / 192 GB
★Mixtral 8x7Bmistral105.1 GBcomfortableFP16 · / 192 GB
Falcon 40Bfalcon89.4 GBcomfortableFP16 · / 192 GB
Command R 35Bcommand78.2 GBcomfortableFP16 · / 192 GB
Aya 23 35Baya78.2 GBcomfortableFP16 · / 192 GB
CodeLlama 34Bcodellama76.0 GBcomfortableFP16 · / 192 GB

Unlocked in a 2x rig

With two cards in parallel (384 GB total), larger models become reachable.

★Llama 3.1 405Bllama311.2 GBcomfortableQ5 · / 384 GB
Hermes 3 405Bhermes311.2 GBcomfortableQ5 · / 384 GB
★Llama 4 Maverick 17Bx128llama307.3 GBcomfortableQ5 · / 384 GB
Nemotron 340Bnemotron261.2 GBcomfortableQ5 · / 384 GB

Unlocked in a 4x rig

Server-style configuration (768 GB total) for the largest open-weight models.

★DeepSeek V3.2deepseek526.3 GBcomfortableQ5 · / 768 GB
★DeepSeek V4 Prodeepseek526.3 GBcomfortableQ5 · / 768 GB
★DeepSeek R1deepseek515.6 GBcomfortableQ5 · / 768 GB
★DeepSeek V3deepseek515.6 GBcomfortableQ5 · / 768 GB
★DeepSeek R1 (0528 snapshot)deepseek515.6 GBcomfortableQ5 · / 768 GB

Similar GPUs

VRAM estimates updated 2026-05-12. Apple Silicon: part of unified memory remains reserved for the system.