ROCm 7.13: Expanding Hardware, Tools, and Reach — ROCm Blogs

Skip to main content

Back to top

Ctrl+K

ROCm blogs

ROCm 7.13: Expanding Hardware, Tools, and Reach

Contents

ROCm 7.13: Expanding Hardware, Tools, and Reach#

May 20, 2026 by Liam Berry, Layla Frischman, Anshul Gupta, Mohammed Faraaz Mustafa, Saad Rahim.

9 min read. | 2176 total words.

Ecosystems and Partners

Hardware, HPC, Installation, Optimization, Performance, AI/ML, Profiling, Fine-Tuning

Developers

Liam Berry, Layla Frischman, Anshul Gupta, Mohammed Faraaz Mustafa, Saad Rahim

English

-->

AMD released ROCm Core 7.13, the AMD GPU Driver 31.30, and AMD GPU Virtualization 9.0. With these releases, ROCm software expands hardware support across enterprise datacenters. The platform introduces AMD’s latest Instinct accelerators, enables GPU virtualization on VMware ESXi and KVM, and delivers optimized performance for generative AI and large language models. The developer experience has been refined with streamlined profiling tools and open-source visibility into low-level performance analysis. As a result, AI development has become more practical across a broader range of hardware, spanning the latest AMD Instinct hardware and newly supported virtualized GPU partitioning modes.

In this blog, you will explore the key highlights of these releases. ROCm 7.13 includes powerful new profiling and visualization capabilities, and a new modular packaging and installation model. AMD GPU Driver 31.30 enables the new MI350P GPU, expanding our hardware ecosystem. AMD GPU Virtualization 9.0 brings new capabilities to GPU partitioning support to KVM and support for newer ESXi versions.

Finally, ROCm introduces new packaging and installation distributions tailored to specific workloads, along with ROCm-Extras packages, starting with ROCm Validation Suite (RVS), all supported as optional packages on top of the ROCm Core SDK. Whether you’re prototyping at home or deploying production infrastructure at scale, the platform provides the capabilities to match your environment.

TheRock Powers the ROCm Core SDK#

ROCm 7.13 continues the evolution of TheRock, AMD’s automated, open-source build and release system for the ROCm software stack. Building on a foundation laid by preview releases from ROCm 7.9 to 7.12, TheRock streamlines ROCm by packaging the necessary foundational components for running high-performance workloads on AMD GPUs.

TheRock introduces the ROCm Core, a base installation containing the essential components that most users need, with optional expansion SDKs available for specialized domains, including HPC, computer vision (ROCm-CV), data science (ROCm-DS), and life sciences (ROCm-LS). Built on a unified, pure CMake build system, TheRock provides stable nightly builds with support for Linux and Windows distributions.

The open-source nature of ROCm accelerates adoption through public PRs, transparent CI pipelines, and code developed directly in public GitHub repositories. This approach results in a faster, continuous release cycle with improved quality and stability, reliable validation, and a smoother out-of-the-box experience.

Expanded Hardware & Platform Enablement#

ROCm 7.13 brings support for AMD Instinct MI350-series datacenter accelerators, adding bare-metal and Kubernetes support for the MI350P and extending GPU partitioning to the MI350X and MI355X.

ROCm 7.13 adds bare-metal support for the MI350P in SPX and CPX modes with NPS1 partitioning, giving you two ways to configure the accelerator depending on your workload. SPX mode dedicates all four XCDs to a single compute partition for maximum throughput. In contrast, CPX mode splits the GPU into four independent compute partitions with one XCD each, allowing multiple workloads to run on a single accelerator. ROCm validates MI350P on Ubuntu 24.04.4, Ubuntu 26.04, and RHEL 9.6.

Beyond bare-metal deployments, the AMD GPU Operator now supports MI350P on Vanilla Kubernetes v1.31 and Red Hat OpenShift v4.21, allowing you to orchestrate MI350P accelerators like any other cloud resource. The operator handles driver installation, GPU health monitoring, metrics export, and partition management, making MI350P practical for containerized AI workloads in production Kubernetes environments.

The MI350X and MI355X gain bare-metal support for QPX with NPS2 partitioning, offering flexible compute configurations for different workload requirements.

GPU Virtualization Across KVM and VMware ESXi#

Running AI workloads in virtual environments means you can share GPU resources across teams, isolate workloads for security, and manage infrastructure with the same tools you already use for the rest of your datacenter. ROCm 7.13 expands GPU virtualization support across the AMD Instinct lineup on both KVM and VMware ESXi, giving you flexible options for allocating GPU resources to virtual machines.

GPU passthrough dedicates an entire GPU to a single virtual machine, delivering bare-metal performance...