API
< Back to all clusters
[TECHNOLOGY] · 2 sources

started · updated

AMD releases Loom 0.2 to enhance AI agent deployment

AMD has released version 0.2 of Loom, a software project within the open-source ROCm ecosystem designed to facilitate the deployment of AI agents. This update builds upon the initial 0.1 release and focuses on expanding C++ support, allowing developers to more easily link native applications and existing libraries to AI systems.

The release introduces several technical improvements, including new runtime managed objects, stateful finite services, and enhanced packaging capabilities for AMD GPUs, NPUs, and Vulkan execution. The shared target compiler has been strengthened to support multiple platforms such as x86, AIE2P, AMDGPU, SPIR-V, and WebAssembly.

To reduce software stack fragmentation, AMD has removed legacy ROCr-based drivers and independent runtime paths, moving instead to a unified hardware abstraction layer based on libamdf. The update also supports RDNA and CDNA architectures, enabling unified memory conventions between CPUs, GPUs, and NPUs. These improvements aim to reduce performance overhead in JIT compilation and GPU scheduling, helping to bridge the software gap with competitors.

Entities

AMD · Loom · ROCm