If you have an AMD Ryzen laptop/desktop and looking to leverage AI/LLM capabilities with the integrated graphics, Linux 7.4 is going to be a real treat especially for lower-end hardware. AMD's new PerfOpt feature being introduced in Linux 7.4 is set to provide some solid gains as shown in these benchmarks today across different AMD Ryzen/Radeon hardware.

Ahead of the Linux 7.4 merge window opening in late October, queued up recently into the IOMMU subsystem's Git tree is a patch enabling a new AMD PerfOpt feature for AMD integrated graphics systems.

AMD PerfOpt is the AMD IOMMU Performance Optimization as part of the AMD I/O Virtualization Technology (IOMMU) specification. With PerfOpt you can bypass the IOMMU when directly accessing system memory to avoid that overhead. The AMDGPU driver will enable AMD PerfOpt by default when using integrated graphics. With the patches going into Linux 7.4, PerfOpt will be automatically enabled on AMD iGPU setups while it can be forced off if wanting to compare the impact via the amdgpu.iommu_perfopt=0 kernel module option.

Curious about the impact of PerfOpt, I built a test kernel from the IOMMU Git tree of that code coming for Linux 7.4. I then tested PerfOpt enabled (default) compared to being disabled on the same kernel. I did this on an AMD Strix Halo desktop given my lack of a Strix Halo laptop for continued hands-on testing plus Strix Point and Krackan laptops for a diversified look at the impact to performance.

For this initial round of testing I used the Lemonade local AI server with Llama.cpp back-end for local AI/LLM inferencing on these AMD Ryzen AI systems. I'll have more AMD PerfOpt benchmarks when looking at the overall Linux 7.4 kernel performance in the weeks ahead. The Linux 7.4 cycle kicks off in the back half of October while the stable kernel will be out around the end of the calendar year. It also stands good chances of being this year's LTS kernel version.