AMD ROCm 7.14.0 Arrives with TheRock Modular Build, PyTorch 2.12, and New APU Support
AMD ROCm 7.14.0 launches with a major architectural shift to the modular TheRock build system, allowing developers to install only the components they need for AI, data science, or HPC workloads. The release updates core frameworks including PyTorch 2.12, JAX 0.10, and vLLM 0.23, while introducing batch memory management APIs and streaming performance monitors for improved profiling accuracy. AMD expands client-side support by adding seven new Ryzen AI APUs to the ROCm ecosystem and validates enterprise OS versions like Ubuntu 26.04, RHEL 10.2, and Debian 13 for Instinct and Radeon PRO GPUs. Cloud providers can now leverage multi-VF partitioning on MI355X and MI350X hardware for finer-grained resource sharing, though users should note that the new Streaming Performance Monitors are currently in beta and not recommended for production use.
AMD ROCm 7.14.0 Arrives with TheRock Modular Build, PyTorch 2.12, and New APU Support
AMD ROCm 7.14.0 has been released, featuring a modular architecture called TheRock that allows developers to install only the components needed for AI, data science, or HPC workloads. The update includes major framework upgrades like PyTorch 2.12, JAX 0.10, and vLLM 0.23, along with new batch memory management APIs and streaming performance monitors for enhanced profiling accuracy. Additionally, seven new Ryzen AI APUs have been added to expand client-side support, and cloud providers can now utilize multi-VF partitioning on MI355X and MI350X hardware for better resource management. However, users should take note that the new Streaming Performance Monitors are currently in beta and not recommended for production environments
