ROCm 7.0.1 has been released to fix an issue concerning incorrect memory page thresholds on AMD Instinct Series GPUs, specifically relating to the AMD GPU Driver (amdgpu) version 30.10.1. This update addresses the failure to declare Out-Of-Band Common Platform Error Records (CPERs) when thresholds are exceeded on the AMD Instinct MI300 and MI350 Series GPUs.
To ensure the effective operation of ROCm 7.0.1, users must align various components, including the GPU, baseboard firmware, AMD GPU drivers, and user space software. Compatibility requirements vary by GPU model, necessitating specific firmware and driver versions. For instance, the Instinct MI300X Series requires firmware version 01.25.03.12 or later, AMD GPU driver version 30.10.1, and a virtualization driver (GIM) version of 8.4.0.K. Users can select ROCm user space software from versions 6.4.z (where z is 0-3), 6.3.y (where y is 1-3), or 6.2.x (where x is 1-4). Similarly, the MI350X Series mandates firmware version 01.25.13.09 or later along with the same AMD GPU driver version.
It's important to note that ROCm 7.0.1 does not introduce significant new features or enhancements; it primarily focuses on resolving existing issues. Users are advised to consult the detailed release notes for ROCm 7.0 and earlier versions to grasp the updates made in these releases.
For those utilizing AMD Radeon PRO or Radeon GPUs in workstations connected to displays, confirming compatibility and system requirements before upgrading to ROCm 7.0.1 is crucial. Additional information regarding native Linux compatibility can be found in the documentation.
In conclusion, ROCm 7.0.1 is a critical release aimed at resolving memory management issues for AMD Instinct GPUs. Users are reminded to maintain synchronization across software components and to refer to comprehensive release documentation to fully understand the changes and improvements made. This ensures an optimized experience for users relying on ROCm for their high-performance computing needs
To ensure the effective operation of ROCm 7.0.1, users must align various components, including the GPU, baseboard firmware, AMD GPU drivers, and user space software. Compatibility requirements vary by GPU model, necessitating specific firmware and driver versions. For instance, the Instinct MI300X Series requires firmware version 01.25.03.12 or later, AMD GPU driver version 30.10.1, and a virtualization driver (GIM) version of 8.4.0.K. Users can select ROCm user space software from versions 6.4.z (where z is 0-3), 6.3.y (where y is 1-3), or 6.2.x (where x is 1-4). Similarly, the MI350X Series mandates firmware version 01.25.13.09 or later along with the same AMD GPU driver version.
It's important to note that ROCm 7.0.1 does not introduce significant new features or enhancements; it primarily focuses on resolving existing issues. Users are advised to consult the detailed release notes for ROCm 7.0 and earlier versions to grasp the updates made in these releases.
For those utilizing AMD Radeon PRO or Radeon GPUs in workstations connected to displays, confirming compatibility and system requirements before upgrading to ROCm 7.0.1 is crucial. Additional information regarding native Linux compatibility can be found in the documentation.
In conclusion, ROCm 7.0.1 is a critical release aimed at resolving memory management issues for AMD Instinct GPUs. Users are reminded to maintain synchronization across software components and to refer to comprehensive release documentation to fully understand the changes and improvements made. This ensures an optimized experience for users relying on ROCm for their high-performance computing needs
ROCm 7.0.1 released
ROCm 7.0.1 has been released, addressing a specific issue related to incorrect memory page thresholds on AMD Instinct Series GPUs. To ensure compatibility, users must use specific versions of the ROCm components along with their corresponding hardware and software configurations. The required version combinations vary depending on the GPU model but generally include firmware and driver updates to maintain synchronization between the GPU, baseboard firmware, and user space software.
