Radeon ROCm 5.3 Released With New APIs, Fine Grain Support
Officially released today is Radeon ROCm 5.3 as the newest version of AMD's open-source compute stack for Radeon and Instinct hardware.
Radeon ROCm 5.3 features a variety of new additions and improvements over the prior ROCm 5.2 series. Some of the ROCm 5.3 highlights include:
- Support for fine-grain region in memory allocation for supported AMD GPUs.
- The per-thread default stream is now supported in HIP.
- A variety of new HIP APIs have been added for management, device management, and more.
- Fixes for texture issues when using ROCm 5.3 on Linux with Blender's HIP back-end.
- Support for the OpenMP GPU Address Sanitizer (Lite) version along with other OpenMP offload improvements.
More details on ROCm 5.3 via GitHub and the AMD documentation portal.
Radeon ROCm 5.3 features a variety of new additions and improvements over the prior ROCm 5.2 series. Some of the ROCm 5.3 highlights include:
- Support for fine-grain region in memory allocation for supported AMD GPUs.
- The per-thread default stream is now supported in HIP.
- A variety of new HIP APIs have been added for management, device management, and more.
- Fixes for texture issues when using ROCm 5.3 on Linux with Blender's HIP back-end.
- Support for the OpenMP GPU Address Sanitizer (Lite) version along with other OpenMP offload improvements.
More details on ROCm 5.3 via GitHub and the AMD documentation portal.
4 Comments