llama.cpp b10430 adds virtual iGPU support for local LLM runs

llama.cpp, the widely-adopted C/C++ library for efficient local LLM inference, has released a new version b10430 from ggml-org on August 14. The headline change is a pull request (#26953) that allows the engine to recognize and use virtual iGPU devices. Virtual iGPUs are software-defined GPU resources commonly found in virtualized environments or modern systems where the integrated GPU is shared or partitioned. By explicitly supporting these devices, llama.cpp now handles hardware configurations that previously may have been ignored or misdetected, making it easier to deploy local models on laptops, VMs, and edge devices with non-standard GPU setups.

The release also includes a follow-up commit that improves code comments around the virtual iGPU logic, signaling a focus on maintainability. Alongside this functional update, b10430 provides a fresh set of prebuilt binaries across multiple platforms and backends: macOS (Apple Silicon and Intel), iOS, Linux (CPU, Vulkan, ROCm, OpenVINO, SYCL), Android (ARM64), and Windows (CPU, CUDA 12/13, Vulkan, OpenVINO, SYCL, ROCm). This broad distribution means developers can quickly test the virtual iGPU support without compiling from source. For users running local AI workloads, this change reduces friction when targeting virtualized GPUs, potentially unlocking better performance and broader hardware compatibility for on-device inference.