Introduction

AMD ROCm (Radeon Open Compute Platform) is a renowned industry-leading open-source artificial intelligence software platform. It contributes to fostering the AI industry ecosystem by supporting heterogeneous hardware, optimizing platform tools, enhancing developer-friendliness, expanding partnerships.

Optimized GPU Software Stack and Comprehensive Developer Tool Set

AMD ROCm is an open software stack including drivers, development tools, and APIs that enable GPU programming from low-level kernel to end-user applications. ROCm is optimized for Generative AI and HPC applications, and it is easy to migrate existing code into ROCm software.

ROCm includes a comprehensive set of tools to provide developers and researchers with flexible management, quality control, and monitoring capabilities.

ROCm optimizes AI workloads and supports the broader AI software ecosystem, including open frameworks, models, and tools

ROCm accelerates AI deployment, allowing mainstream models such as Llama and DeepSeek to run on the day of release. Through close collaboration with the open-source community, over 2.1 million Hugging Face models are now supported on ROCm software. The newly released ROCm 7 preview offers an average 3.5x improvement in inference performance and an average 3x improvement in training performance compared to ROCm 6, enabling efficient inference for models like DeepSeek R1.

ROCm 7 enhances capabilities integrating distributed inference and enterprise-grade optimization. The new features include distributed inference with open-source framework support, enterprise-ready AI tools with orchestration and endpoint integration, support for large-scale models with new data types FP6 and FP4 for enhanced flexibility and performance, enhanced code portability with HIP 7.0, etc.

ROCm provides developers with resources through its developer hub and industry-academia-research collaboration

AMD collaborates with global tech and large model companies such as Meta, Microsoft, DeepSeek, and Alibaba Cloud to build an open and mutually beneficial AI ecosystem. At the same time, AMD is deepening industry-academia-research cooperation by engaging with universities both domestically and internationally to promote the ROCm ecosystem. We have visited top institutions such as Tsinghua University, Peking University, Stanford University, and MIT to conduct technical sharing sessions, explain the principles behind running large models on the ROCm software stack, and share performance optimization strategies-helping to drive ROCm adoption in academic research. In July this year, China's first ROCm software lab was inaugurated in Nanjing Software Valley. Looking ahead, AMD will uphold the philosophy of “Together We Advance," working together with customers and partners to implement the AI action plan and empower industries across the board.