Ollama 0.30 发布：NVIDIA 性能提升 20%，默认启用 Vulkan 并增强 GGUF 支持
Improved performance and model support with GGUF · Ollama Blog
Ollama 0.30 发布，通过 llama.cpp 增强 GGUF 模型兼容性，并补充 Apple silicon 上的 MLX 引擎。NVIDIA 硬件性能最高提升 20%，实测使用 Gemma 4 26B 模型在 NVIDIA RTX 5090 上以 Q4_K_M 量化运行。Vulkan 现默认启用，将 GPU 加速扩展至 AMD 和 Intel 设备；同时新增对 LFM、Prism 及 Unsloth 微调模型的支持，并可直接运行 Hugging Face 上的 GGUF 模型。
正文
当前显示原文，中文全文译文尚未就绪。
June 5, 2026
Ollama 0.30 is now available with improved performance and GGUF model compatibility through llama.cpp. This augments Ollama’s MLX engine on Apple silicon, bringing support to more models on a wider range of hardware.
Performance across more GPUs
Faster throughput on NVIDIA hardware
With Ollama 0.30, performance on NVIDIA hardware is now up to 20% faster, leveraging optimizations contributed by the NVIDIA and llama.cpp teams.
Tested with the Gemma 4 26B model running on an NVIDIA RTX 5090 using the Q4_K_M quantization.
Wider hardware support with Vulkan
Vulkan is now enabled by default, extending Ollama’s GPU acceleration to a wider range of hardware, including AMD and Intel devices. More users can now run models on the GPU out of the box, without installing vendor-specific libraries.
Support for more models
Ollama 0.30 expands compatibility with the GGUF ecosystem, so more models run out of the box—including model families such as LFM and Prism, as well as fine-tuned models published by Unsloth.
Run GGUF models from Hugging Face
To use a model, first download the GGUF file or a directory containing GGUF files. Next, create a Modelfile with the FROM command pointing to the path of the GGUF file (or directory):
Then create and run the model:
Coding agents and assistants
If a model supports tool calling, that capability carries over to Ollama. You can use these models with your favorite coding agents and personal assistants in a single command.
To verify that a GGUF file supports tool calling, look for the tools capability with ollama show:
Acknowledgements
We’d like to acknowledge the work done by Georgi Gerganov and the llama.cpp maintainer teams, as well as hardware partners including NVIDIA, AMD, Qualcomm, and Intel, who have worked hard to optimize performance with the GGML ecosystem on their respective platforms.
If you have any feedback, join Ollama’s Discord or reach out at hello@ollama.com.