AMD has released ROCm 7.14, a major update to its open-source AI software platform, featuring TheRock, its first production-ready automated build and release system. TheRock modernizes how ROCm is built, packaged, validated, and delivered, marking a significant milestone in AMD's AI software ecosystem. The release includes expanded hardware support, enhanced AI frameworks, and improved developer tooling for a broader range of applications. TheRock introduces ROCm Core SDK, a base installation containing essential components for most users, with optional expansion SDKs for specialized domains. Built on a unified, pure-CMake build system, TheRock provides stable nightly releases with support for multiple Linux distributions and Windows. The open-source nature of TheRock accelerates adoption through public PRs, transparent CI pipelines, and code developed directly in public GitHub repositories. This results in a faster, continuous release cycle with improved quality and stability, reliable validation, and a smooth out-of-box experience. Source: amd
ROCm 7.14 continues AMD's investment in expanding support across its growing AI portfolio, enabling production support of AMD Instinct™ MI350 PCIe® GPUs. OS coverage broadens to match MI350X and MI355X, with new support added for RHEL 9.8, SLES 15 SP7, SLES 16, and Debian 13 across both bare-metal and passthrough virtualization environments. MI350P is also now enabled for Vanilla Kubernetes on Ubuntu and Red Hat OpenShift v4.21, providing a validated path for deploying MI350P in containerized and orchestrated production environments. This provides a consistent development experience across the AMD Instinct accelerator family, allowing developers to use the same open-source ROCm frameworks, libraries, compilers, and profiling tools to build, optimize, and deploy AI inference applications on PCIe-based enterprise servers. The release brings full ROCm enablement to the AMD Ryzen AI MAX+ PRO 495 and Ryzen AI MAX PRO 490 and 485 processors, enabling developers to leverage the complete ROCm software stack for AI, HPC, and content creation workloads. Source: amd
The AMD Enterprise AI Reference Stack continues to expand its support for production AI deployments on AMD platforms. AMD Solution Blueprints will also expand with industry-focused solutions for healthcare, telecommunications, and financial services, providing validated starting points for deploying enterprise AI workloads tailored to vertical-specific use cases. AMD Enterprise AI Reference Stack now includes support for AMD Radeon™ GPUs as well as additional capabilities for AMD Inference Microservices (AIMs) and AMD AI Workbench. RDNA 4 (Radeon AI PRO R9600 & R9700) and RDNA 3 (Radeon PRO W7900 & W7800) GPUs are now supported. AIMs now expose an OpenAI-compatible API for LLM serving, with prebuilt inference containers that automatically configure based on the detected hardware profile, built-in model download and caching, and native integration with Kubernetes and Red Hat OpenShift for enterprise deployment. AMD AI Workbench complements this with prebuilt training and fine-tuning pipelines, automatically configured developer workspaces, GPU-as-a-Service support, and built-in ROCm profiling integration. Source: amd
