Radeon ROCm 2.2 Released With Vega 20 Optimization, Caffe2 Multi-GPU Training

One month since the release of ROCm 2.1, the Radeon Open eCosystem stack has now been succeeded by ROCm 2.2.

The ROCm 2.2 release shipping today comes as a bit of a surprise. It's not the most feature-packed update but does have some nice additions while building on the already exciting ROCm 2.0 that shipped at the end of 2018.

ROCm 2.2 brings rocSPARSE optimizations for Vega 20 with cache usage improvements, improved DGEMM performance for reduced matrix sizes, and with Caffe2 there is now support for multi-GPU training.

This ROCm 2.2 release builds upon ROCm 2.1 from last month that had the RocTracer 1.0 preview release and ROCM-SMI tool enhancements while ROCm 2.0 was the big release with Vega 20 / Vega 7nm support, MIVisionX as their computer vision libraries, PyTorch and Tensorflow improvements, and full OpenCL 2.0 support.

Download links and more details on ROCm 2.2 are available from the RadeonOpenCompute GitHub.