AMD Launches ROCm 10 with New AI-Native Developer Tools

AMD ROCm 10 introduces ROCm.AI, Hyperloom, AMD Skills, and ROCm CLI to simplify AI development and optimization on AMD hardware.

AMD has officially released ROCm 10, marking a decade of development for its open software stack while introducing ROCm.AI, a new AI-native developer experience designed to help developers build, optimize, and deploy AI workloads on AMD hardware.

ROCm.AI is now generally available and brings together three main components: AMD Skills, ROCm CLI, and ROCm Hyperloom. Together, these tools are intended to simplify AI development while bringing AMD-specific expertise and agentic workflows directly into the development tools users already work with.

ROCm.AI Brings Agentic Workflows to AMD Development

One of the biggest additions in ROCm 10 is ROCm.AI, which AMD describes as an AI-native software experience focused on improving development and optimization across its platforms.

The platform combines AMD Skills, ROCm CLI, and ROCm Hyperloom to cover different parts of the AI development process, from workload optimization and diagnostics to deployment and system management.

AMD says a system configured with ROCm.AI achieved an average 3.3x improvement in inference performance and a 2.4x improvement in training performance compared with ROCm 7 on the same hardware. These figures come from AMD Performance Labs testing using AMD Instinct MI355X systems and selected AI models.

ROCm Hyperloom Automates Inference Optimization

ROCm Hyperloom is designed to handle end-to-end inference optimization across both host code and GPU kernels. Rather than focusing on a single portion of a workload, the autonomous system profiles an application, identifies bottlenecks, explores potential optimizations, implements changes, and then benchmarks and validates the results.

With ROCm 10, Hyperloom expands its support across AMD Instinct GPUs and adds support for vLLM and SGLang. Developers can target optimizations across HIP, Triton, and FlyDSL, with generated reports outlining proposed changes along with measured or expected performance improvements.

Hyperloom can be accessed through standalone workflows or through AMD Skills, giving developers different ways to incorporate automated optimization into their existing development processes.

AMD Skills Bring Platform Expertise to Coding Agents

AMD Skills is aimed at developers who want AMD-specific guidance without having to leave the tools they already use. The skills can be integrated with supported AI coding agents including Claude Code, Cursor, and Codex.

The catalog covers several areas, including client-native workflows for local AI and application integration, cross-stack workflows for diagnostics and optimization, and server-native workflows for AMD Instinct GPUs and AMD EPYC processors.

AMD says the skills previewed earlier this year are now available through the Claude Code, Codex, and Cursor marketplaces, as well as through an open catalog on GitHub. Each released skill undergoes structural and behavioral testing before being made available.

ROCm CLI Simplifies AI Workload Management

ROCm 10 also introduces the ROCm CLI as a Technology Preview. The command-line interface provides a unified way to set up, manage, and operate AI workloads on AMD hardware.

Developers can use the same interface manually, through an AI coding agent, or within continuous integration environments. The CLI can inspect systems, install and manage ROCm environments, serve models, perform diagnostics, update components, and control runtimes.

The ROCm CLI is available as a prebuilt binary for both Windows and Linux and does not require an existing ROCm installation. It also supports multiple ROCm runtimes, including runtime activation and rollback, as well as integrated model serving and engine management.

For monitoring, the included ROCm Console provides a real-time view of system status and workload activity. On supported AMD Instinct systems, developers can monitor information such as GPU utilization, benchmark telemetry, high-bandwidth memory usage, power consumption, and tokens per watt.

The CLI currently supports ROCm 7.13 and later releases, with official ROCm 10 support planned for a future update.

ROCm 10 Expands the Underlying Software Stack

These new developer tools build on the broader ROCm 10 software stack, which introduces a more modular ROCm Core SDK for developing and running AI workloads across AMD platforms.

AMD has also updated the stack's libraries, compilers, frameworks, tools, model support, and hardware compatibility. The goal is to provide developers with a more flexible foundation for deploying AI workloads across AMD's client and data center platforms.

AMD Continues to Push Open AI Development

With ROCm 10, AMD is positioning its software stack around more than simply providing an alternative platform for running AI models. The company is increasingly incorporating AI-assisted development, automated optimization, and simplified system management into ROCm itself.

For developers, the combination of ROCm.AI, AMD Skills, Hyperloom, and the ROCm CLI could make it easier to optimize workloads without manually working through every layer of the software and hardware stack.

ROCm 10 also marks 10 years of AMD's ROCm software stack, as the company continues expanding its AI ecosystem across AMD Instinct GPUs, EPYC processors, and client platforms.