8GB VRAM
Entry tier for lightweight chat and coding models.
View recommended modelsHardware VRAM Tiers
12GB VRAM
Balanced tier for 7B/13B with practical local runs.
View recommended models16GB VRAM
Comfort zone for 13B/14B and selected 32B Q4 setups.
View recommended models24GB VRAM
High-value tier for large quantized models and long context tests.
View recommended modelsApple Silicon
Unified memory planning guide for local LLM inference on M-series devices.
Open Apple Silicon guideVerified on RTX 3090
Measured model tags and sustained-load benchmark snapshots on 3090.
Open verification page