ROCm 7.0.0 vs. ROCm 7.2.3 Performance On The AMD Radeon AI PRO R9700

With the new System76 Thelio Major workstation review unit having arrived equipped with an AMD Radeon AI PRO R9700 graphics card, I took the opportunity of having the extra RDNA4 workstation GPU to satisfy a curiosity over whether there has been any meaningful performance gains from ROCm 7.0.0 released last year to now with the latest ROCm 7.2.3 stable release. Here are those benchmarks results if you are curious about the impact of just updating the user-space ROCm components from the end of last summer to the latest ROCm 7.2.3 milestone.

When loading Ubuntu 24.04 LTS onto the System76 Thelio Major for its official AMD ROCm support, I ran some benchmarks comparing the performance of the ROCm 7.0.0 packages to that of ROCm 7.2.3. Using the official AMD binaries each time with AMD thankfully still keeping older versions of their ROCm binaries online.

This ROCm 7.0.0 vs. 7.2.3 comparison was limited to just the user-space ROCm components with keeping to the stock Linux 6.17 AMDGPU/AMDKFD kernel driver components for this comparison and not opting for the AMDGPU/AMDKFD DKMS kernel modules to exclusively look at the ROCm bits.

From there a variety of different ROCm/HIP and ROCm-OpenCL benchmarks were conducted for seeing how ROCm 7.2.3 compares to ROCm 7.0.0 performance-wise that debuted last September. All tests were done on the AMD Radeon AI PRO R9700 within the Thelio Major workstation, kindly provided by System76 for testing at Phoronix.