Inference
-
Ollama on Apple Silicon: Unified Memory Sizing
Ollama on Apple Silicon: how much unified memory a model actually gets, what each M-series chip's bandwidth means for speed, and which Mac to size for.
-
Ollama Tokens per Second: What Sets Your Speed
Ollama tokens per second explained: how memory bandwidth, quantisation, and context length set generation rate, plus how to measure your own numbers.
-
Ollama Without a GPU: CPU-Only Speed and RAM
Ollama without a GPU: what CPU-only inference costs in tokens per second, how much system RAM each model class needs, and when it is still worth running.
-
Ollama Not Using GPU: Fix 100% CPU and Partial Offload
Ollama not using GPU? If ollama ps shows 100% CPU or a CPU/GPU split, check the 550+ driver, Docker GPU access and iGPU flag, then fix VRAM or context.
-
How Much VRAM Do You Need to Run Local LLMs with Ollama?
VRAM sizing for local LLMs: how quantisation, parameter count, and context length set your GPU memory bill, and which model fits 8, 12, 16, or 24 GB.