Further Exploring The Intel Tiger Lake Core i7-1165G7 Performance On Ubuntu Linux

So from this testing of the i7-1165G7 in five different Linux software configurations there are a few takeaways:

- Using anything newer than Mesa 20.0 as shipped by Ubuntu 20.04.1 LTS (and the stock Dell software image for the laptop) is very beneficial for the Gen12 Xe Graphics. Not unexpected at all given the new age of the hardware and why for the original tests Mesa 20.3-devel was used. Mesa 20.2 as tested on Ubuntu 20.10 was yielding similar graphics performance too. So any Tiger Lake users should consider this for much better graphics performance. With my tests I was using Oibaf PPA, which is easy for Ubuntu users to enable and upgrade.

- In single-threaded workloads and the same tests where the i7-1165G7 showing was poor in the original comparison to Ice Lake and the rest, the performance was much improved with Ubuntu 20.10. But why? Even when using Linux 5.9 or Linux 5.10 in its early development state wasn't able to match Ubuntu 20.10 with its Linux 5.8 stable kernel...

- But then in the heavy, multi-threaded workloads where Tiger Lake was providing uplift over Ice Lake in the prior articles, the Ubuntu 20.10 runs were now strangely much lower than the 20.04 configurations...

Seeing the change around Ubuntu 20.10 was unexpected. With my testing of Ubuntu 20.10 on other systems, I haven't seen such broad performance changes or anything else that would align with this behavior seen on Tiger Lake. Ubuntu 20.10 does introduce GCC 10 as the default compiler, Python 3.8, and numerous other package upgrades...

At this point it's half-way through the weekend and now at a point of seeing Ubuntu 20.10 performing better than Ubuntu 20.04 LTS even with Dell's stock software configuration shipped on this Tiger Lake laptop.

Thinking of the other software differences between 20.04 and 20.10, one of the early hopeful thoughts was that Intel's Thermald is to blame. Intel's Thermal Daemon for Linux is needed by newer mobile platforms that interfaces with RAPL power-capping, P-State, Intel PowerClamp, and INT340x kernel drivers for managing the thermal/power behavior of these platforms.

At first that looked hopeful after digging into package versions with Thermald in Ubuntu 20.04 LTS being at version 1.9 where as Intel added Tigerlake support with Thermald 2.0. But that thought was quickly foiled. Canonical ended up back-porting the Tigerlake PCI IDs to their Thermald package as an update in focal-updates, which was already running on the system throughout all of the 20.04 tests... But Ubuntu 20.10 does run Thermald 2.3 as the newest Intel Thermal Daemon release. So I was thinking that may have been the explanation for Ubuntu 20.10 offering better performance.

Fired up tests... Watched for a few hours, and sadly, when running Thermald on Ubuntu 20.04 or even using the Thermald package from Groovy back on Fossa, there wasn't any uplift in those single-threaded tests.

The next hypothesis was possibly some updated firmware on Ubuntu 20.10. The same CPU microcode for the i7-1165G7 was on both Ubuntu releases, but there's been some other Intel firmware updates in this time. Back in August was a big HuC update for the Tiger Lake graphics that only noted of a "major upgrade" but with no further details. There was also a DMC update around Tigerlake with the DC5/DC6/DC3C0 states also from August. Thinking perhaps those firmware updates on the graphics side might have something to do with the CPU power management, I ran some tests on Ubuntu 20.04 when making those firmware upgrades and deploying them.

It was hard to tell if that would have any impact since they are just binary blobs and not a fully documented public change-log. Sadly, those i915 firmware upgrades didn't do any good. So I then did a run with the entire linux-firmware.git tree deployed on the Dell XPS 9310 and repeated the tests...

Still no dice. Upgrading Thermald and linux-firmware.git on Ubuntu 20.04 LTS with Linux 5.9 didn't do anything to help improve the performance in the single-threaded conditions... Also did a run with using the Groovy kernel build rather than a Linux mainline kernel build on 20.04 and there wasn't any such change.