Further Exploring The Intel Tiger Lake Core i7-1165G7 Performance On Ubuntu Linux
Last week I published initial benchmarks of the Intel Core i7 1165G7 "Tiger Lake" performance on Linux with the Dell XPS 13 9310 Developer Edition laptop. Of most surprise from those preliminary Linux figures were finding that for some single-threaded workloads the performance was actually worse than the previous generation Ice Lake. Since then I've been running more tests around the clock with some interesting discoveries to note today. It is possible to enhance the single-threaded performance so it's performing better than Ice Lake as would be expected, but comes with lowering the multi-threaded performance compared to the results shared last week.
Since the prior the article, the focus of my testing with the Dell XPS 9310 Developer Edition Tiger Lake laptop has been exploring why for those lightly loaded, single-threaded workloads from MP3 encoding to web browser performance that in some cases the performance ended up lower than the previous-generation Dell XPS with Core i7 1065G7 Ice Lake on the same Ubuntu software stack. As shown in the previous article, the CPU peak frequencies were often well below the maximum turbo frequencies advertised by the i7-1165G7, but no clear reason why considering even with Ubuntu 20.04.1 LTS as shipped by Dell experienced the same behavior and also moving to the very latest Linux kernel. So in the days since it's been relentlessly testing various configurations to see what's going on.
So initially I looked at the difference of Dell's stock configuration for the XPS 9310 Developer Edition (Linux 5.6 OEM kernel with Mesa 20.0), Ubuntu 20.04 with Linux 5.9 and Mesa 20.3-dev (this is the configuration I used for the laptop/CPU comparison in the original article in always preferring the very latest kernel and Mesa for new GPU/CPU launches), Ubuntu 20.04.1 LTS then moved to Linux Git currently in the pre-RC1 state half-way through the merge window, Ubuntu 20.10 in its near final development state, and then a run of Ubuntu 20.10 when using Mesa 20.3-devel rather than Mesa 20.2 as is set to ship for next week's 20.10 release.NCNN NCNN
One of the first findings was not unexpected at all: using Mesa 20.3 (or even 20.2) yields much better graphics performance for the Gen12 / Xe Graphics on the i7-1165G7 compared to Mesa 20.0 as shipped by Ubuntu 20.04.1 LTS and also what Dell is shipping on the Developer Edition laptop. Keep in mind with last week's article I was using Mesa 20.3-devel (and Linux 5.9) already, so the "Dell Stock" simply shows the worse, out-of-the-box state for the Tiger Lake graphics if not moving past Mesa 20.0.... But as you can see, for OpenGL and Vulkan the performance gains are very worthwhile in the newer state of Intel's Iris Gallium3D and ANV Vulkan drivers. It's too bad Dell didn't opt for shipping a newer stable Mesa release with their laptops for a much better initial experience. But anyways, this isn't entirely surprising and why I was using Mesa 20.3-devel from the start, but let's keep going in trying to find something new:
My curiosity was piqued when hitting some of the other games tested... In some cases, using Ubuntu 20.10 is yielding much more uplift than Ubuntu 20.04 -- even when modifying Ubuntu 20.04 with the newer Mesa 20.3-devel and Linux 5.9/5.10! Hmmmm, these open-source games are also lightly loaded like the CPU tests in the prior article where Tiger Lake was coming up short of Ice Lake... But why Ubuntu 20.10 when the newer Mesa and Linux kernel on 20.04 didn't yield similar results?