ROCm 7.2.3 has been released, addressing idle gaps in vLLM profiling traces and optimizing MIGraphX for embedding-heavy models, significantly improving AI inference and profiling stability. The update includes enhancements to ONNX Runtime's multi-stream memory handling, but server administrators need to check firmware and driver compatibility, especially for MI325X KVM users who must avoid the 30.20.0 AMD GPU driver. Legacy profiling tools and ROCm SMI are being deprecated, urging teams to migrate to rocprofv3 and AMD SMI before the support windows close in mid-2026. Additional features include an early-access API for Accelerator-Initiated IO, while known issues, such as a minor performance drop for int8-quantized models, are being investigated for future fixes
ROCm 7.2.3 released
ROCm 7.2.3 finally patches those frustrating idle gaps in vLLM profiling traces that turn latency debugging into a guessing game. MIGraphX gets a solid performance bump for embedding-heavy models through fused gather operations, and ONNX Runtime gains better multi-stream memory handling to keep inference pipelines from choking on themselves. Server admins should double-check their firmware and driver matrices since MI325X KVM users absolutely must skip the 30.20.0 AMD GPU driver. The software stack is also actively retiring legacy profiling tools and ROCm SMI by mid-2026, so teams need to migrate to rocprofv3 and AMD SMI before those support windows close.
