Ollama Releases Engine Improving Local Inference

Ollama released version 0.17, introducing a new inference engine, broader hardware support, and multiple under-the-hood improvements. The update claims up to 40% faster prompt processing and up to 18% faster token generation on NVIDIA GPUs, adds AMD RDNA4 and improved Intel GPU compatibility, enhances multi-GPU tensor parallelism, and brings 8-bit KV cache quantization—boosting local LLM responsiveness and enterprise practicality.