ROCm Core SDK

Everything you need to develop GPU-accelerated applications on AMD hardware. The ROCm Core SDK includes the essential libraries, runtimes, compilers, and tools that form the foundation of the ROCm software stack.

Diagram of the ROCm Core SDK showing math, communication, media, storage, runtime, profiling, debugging, and monitoring libraries, including Composable Kernel, MIOpen, rocWMMA, hipBLAS, rocBLAS, RCCL, rocDecode, hipFile, HIP, LLVM, ROCm Compute Profiler, ROCgdb, AMD SMI, and RDC.

What’s New with ROCm 7.14

Production-ready TheRock for faster, modular ROCm development

Broad AI hardware support across AMD Instinct™ GPUs, Radeon™, and Ryzen™ AI

Optimized support for leading open-source AI frameworks and inference engines

Enterprise-ready AI deployment with cloud-native orchestration and developer tooling

Enhanced performance for distributed AI, profiling, and scalable deployment

AMD ROCm Lockup White

ROCm Evolution Over the Years

Leading enterprises and research institutes have been leveraging ROCm for nearly a decade. Explore the various milestones that are a part of the history of ROCm.

ROCm Core SDK Components

Explore the key component that power the ROCm Core SDK

AMD ROCm Lockup White

GPU-accelerated AI, math, and communications libraries for HPC. Native roc* libraries optimize performance on AMD GPUs, while hip* wrappers simplify CUDA portability. Scale distributed AI and HPC workloads with high-performance multi-GPU and multi-node communication libraries.

Build and run GPU applications efficiently with AMD compiler toolchains and runtimes. Measure performance, analyze GPU utilization, and identify bottlenecks with profiling and debugging tools to optimize AI, HPC, and accelerated computing applications.

Monitor and manage GPU hardware, including temperature, power, memory, and resource allocation, with runtime, driver, and operating system tools.

Monitor and manage GPU hardware state, including temperature, power, and resource allocation.

AI Inference

Production AI Inference with ROCm

Deploy leading open models on AMD GPUs using a fully open-source inference stack. Get up and running with prebuilt containers and upstream framework support.

Open, Out-of-the-Box Inference

From Fine-tuning to Production Serving—Without Changing Workflows

AMD ROCm delivers an open-source, end-to-end inference stack with upstream framework support (including PyTorch) and validated containers for vLLM and SGLang, so teams can move from fine‑tuning and benchmarking to production serving on AMD Instinct™ GPUs with a consistent workflow.

Efficient Inference, Optimized by Default

High Throughput Without Per-model Tuning Overhead

ROCm pairs quantized model packages with native low-precision execution and optimized transformer hot-path kernels, reducing tuning effort while maximizing throughput and efficiency for real-world serving.

Distributed Inference for Modern Serving

Built for Prefill, Decode, and Scale-out Workloads

ROCm supports distributed inference with prefill/decode disaggregation and optimized multi-GPU, multi-node communication—helping sustain responsiveness and utilization as inference scales. 

Production-Ready for Kubernetes with AI

Operate Inference Reliably at Fleet Scale

ROCm integrates with Kubernetes through the AMD GPU Operator, built-in telemetry, and a modular driver architecture—supporting automated rollout, observability, and predictable lifecycle management across secure and air-gapped environments. 

AI Training

End-to-End AI Training at Scale on AMD GPUs

Train and fine-tune AI models on AMD GPUs with an open, end-to-end software platform. Leverage optimized frameworks, efficient GPU acceleration, and seamless scaling from a single GPU to multi-node clusters.

Optimized AI Training Performance

Accelerate Model Training With Compute and Optimizations

ROCm leverages the parallel processing capabilities of AMD GPUs to reduce training time and improve overall performance for AI workloads. With optimized libraries, efficient resource management, and support for modern AI frameworks, developers can train larger models faster while maximizing GPU utilization.

Efficient Fine-Tuning for Specialized AI

Adapt Foundation Models for Domain-specific Tasks with ROCm

ROCm accelerates fine-tuning of large language models and other foundation models on AMD GPUs with optimized libraries, broad framework compatibility, and efficient resource management. Fine-tune models faster while integrating seamlessly with open-source frameworks, models, and tools.

Scalable Distributed Training

Run Multi-Node Training Reliably on AMD GPU Clusters

Primus helps teams scale training across AMD GPU clusters with distributed execution, topology-aware scheduling, and fault-tolerant operation. This supports more consistent training as workloads and clusters grow.

Enterprise-Ready Training Operations

Telemetry, Validation, and Faster Recovery for Long-running Jobs

Primus provides telemetry, dashboards, and node health benchmarking, giving teams visibility into training performance and cluster health. This helps organizations operate large training workloads with greater reliability and faster troubleshooting.

ROCm for HPC and Supercomputing

Build and Scale HPC Applications with ROCm

From optimized libraries to leading supercomputers - ROCm enables high-performance computing on AMD GPUs.

Accelerating HPC Innovation

Power Breakthroughs Across Science, Engineering, and Research

AMD ROCm™ enables high-performance computing (HPC) and supercomputing workloads across energy, molecular dynamics, physics, computational chemistry, climate science, and other research domains, providing an open software platform to help tackle some of the world's most complex computational challenges. 

Flexible HPC Programming

Develop With the Tools That Best Fit Your Application

ROCm supports multiple HPC programming models, including OpenMP®, HIP, OpenCL™, and Python™, giving developers the flexibility to build, optimize, and scale applications using the languages and frameworks that best suit their workloads.

Optimized HPC Libraries

Accelerate Applications With Optimized Libraries

ROCm includes a comprehensive set of math and communication libraries designed to simplify HPC development while improving application performance, scalability, and efficiency across AMD compute platforms.

Proven at Supercomputing

Scale Powering the World's Leading Supercomputers

AMD powers 177 of the Top500 supercomputers, including Frontier, LUMI, and the upcoming El Capitan system. ROCm provides the software foundation for running some of the world's largest and most demanding HPC applications at exascale and beyond.

Partner Case Studies

Partnerships and Proven Success

See how enterprises and leading research institutes have partnered with AMD.

Featured Blogs

Introducing ROCprofiler

Explore the new ROCprofiler-SDK and learn how its unified profiling infrastructure streamlines performance optimization for AI and HPC applications on AMD GPUs.

AMD Profiling Tools

Learn how AMD profiling tools provide deep performance insights to help developers benchmark, profile, and optimize heterogeneous applications running on CPUs and GPUs.

Supported Hardware

Developer Resources

ROCm Developer screen shot

ROCm Developer Hub

Start developing AMD GPU-accelerated applications. Visit the ROCm Developer Hub to get access to the latest user guides, containers, training videos, webinars, and more.

AI Developer Program

Access to free AMD Developer Cloud credits, exclusive training, monthly hardware sweepstakes, and community support designed to support your AI development work.

ROCm Newsletter

Receive the latest ROCm news.

Footnotes

©2024 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, AMD ROCm, AMD Instinct, EPYC, Radeon Instinct, and combinations thereof are trademarks of Advanced Micro Devices, Inc. PyTorch is a trademark or registered trademark of PyTorch. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.

  1. For a full list of Radeon parts supported by ROCm, go to https://rocm.docs.amd.com/en/latest/reference/gpu-arch-specs.html