AMD's ROCm 2.0 Radeon Compute Stack Being Prepared For Release

Last month AMD commented they would be releasing ROCm 2.0 prior to the end of 2018 and it looks like they will make good on their word. ROCm 2.0 is being prepared for release - source code is available albeit the reference Ubuntu/RHEL binaries are not yet out.

We've been looking forward to ROCm 2.0 for months as it's the release of the Radeon Open eCosystem stack delivering full OpenCL 2.0 support. But OpenCL 2.0 support is only part of what makes ROCm 2.0 exciting. ROCm 2.0 also includes:

- Vega 7nm (Vega 20) is officially supported.

- MIVisionX is now included as the AMD-optimized computer vision and machine intelligence libraries.

- Performance improvements, FP16 support in RNNs, and other improvements to MIOpen.

- TensorFlow 1.12 support along with FP16 support and multi-GPU support for Vega 7nm. PyTorch/Caffe2 also has Vega 7nm support, including FP16 support and performance improvements.

- 48-bit virtual addressing support for Vega 10 GPUs and newer, resulting in some performance improvements for workloads with heavy memory management and other areas.

- Support for Kubernetes.

- Official support for RHEL/CentOS 7.6 and Ubuntu 18.04.1 LTS.

The ROCm 2.0 sources are now available from the various trees like the kernel driver and runtime, among the many other repositories. However, the official binaries as of writing haven't yet been uploaded nor has the formal release announcement hit the wire, but these final release preparations are underway.

It's looking like ROCm 2.0.0 should officially be announced in time for some fun Christmas benchmarking.