Improved performance and model support with GGUF

Enhanced Performance on NVIDIA Hardware

Ollama 0.30 introduces performance improvements on NVIDIA hardware, with throughput increases of up to 20% as reported in the release notes. These optimizations are attributed to collaborative efforts between the NVIDIA and llama.cpp teams. The improvements are particularly notable when running large models such as Gemma 4 26B on high-end GPUs like the RTX 5090, using quantization methods such as Q4_K_M. This suggests that the platform is becoming more efficient in handling complex workloads on NVIDIA's latest hardware.

Wider Hardware Support with Vulkan

The release also marks the default activation of Vulkan support, which extends GPU acceleration capabilities to a broader range of hardware, including AMD and Intel devices. This change reduces the need for users to install vendor-specific libraries, making it easier for a wider audience to leverage GPU acceleration without additional configuration. This is a significant step toward cross-platform compatibility and accessibility.

Expanded Model Compatibility with GGUF

Ollama 0.30 expands its compatibility with the GGUF model format, allowing more models to run out of the box. This includes support for model families such as LFM and Prism, as well as fine-tuned models from Unsloth. The inclusion of GGUF support enables users to run models directly from Hugging Face by downloading the GGUF file and using a Modelfile to define the model configuration. This simplifies the process of deploying models on the Ollama platform, reducing the barrier to entry for developers and researchers.

Integration with Coding Agents and Assistants

Models that support tool calling can now be used with various coding agents and personal assistants within Ollama. This integration allows users to leverage models like Claude Code, Hermes Agent, and OpenClaw with a single command. The ability to verify tool calling support via the ollama show command ensures that users can confirm model capabilities before deployment.

Conclusion

Ollama 0.30 represents a significant step forward in performance, compatibility, and usability. The improvements on NVIDIA hardware, the inclusion of Vulkan support, and the expansion of GGUF model compatibility all contribute to a more versatile and accessible platform. These changes are likely to benefit a wide range of users, from developers to researchers, by enabling more efficient and flexible model deployment across diverse hardware environments.

Source: Ollama. Analysis based on publicly available information; no hands-on testing is claimed.