S922X Ugoos AM6B Device Tree - Performance/Efficiency - Testing Needed

EAS requires specific hardware properties and kernel features, including an asymmetric CPU topology, the presence of an Energy Model, and the Schedutil governor.

Ubuntu documentation calls out the need for schedutil if using energy model. We don’t have schedutil in our default kernel so does that mean we are not using the energy model?

Stress ng shows a difference with cache not in performance but in how the data is available in userspace for built-in prefetch.

Saw a weird behavior, the on-demand governor boots faster (consistently across 3 reboots).

Performance
systemd-analyze
Startup finished in 2.749s (kernel) + 6.095s (userspace) = 8.844s
kodi.target reached after 6.095s in userspace.

Ondemand
systemd-analyze
Startup finished in 2.734s (kernel) + 5.007s (userspace) = 7.741s
kodi.target reached after 5.007s in userspace.

Could be related to the IRQ rebalance in /proc/interrupts
IPI4: 268 243 279 22 169 13 IRQ work interrupts
The above numbers are right at boot.

Can someone else also try the on-demand governor and see how it goes?

Sure, more than 10% faster boot time but does it really matter?

1 Like

It added 2 fps in retreoarch (in performance cpu government). Yes it is matter when you have before 48 fps and after 50 fps on 50hz TV. And also after no tearing of GUI on ondemand cpu government.

3 Likes

is this enough to convince you that this is increasing performance and not just a placebo?
YadaYada?

It is measured and compared in same conditions !!!

1 Like

It would have to be repeated. Replace the old DTB test the framerate, add the new DTB reboot test the framerate. Repeat this so that you have 3-5 measurements with & without the changes.

2 frames is a 4% change, which could be real or variation. Retroarch is running on top of CE, there’s a lot of variables at play. Multiple measurements going back and forth between new and old DTBs would make it more convincing.

the cache enhancement can effect different parts with different levels, right.
the frame rate might not be the best part it’s doing the most on and frame rate increasing might not be linear, so to get an extra 2 frames might need more performance than just 2%?
either way if this verifiable and atirage21 seems like he’s been begrudging not having that extra 2 frames and noticing it for a while and now instantly see’s the improvement and also states nothing else changed except the DTB shows that this can’t be a placebo on this one test. so it opens the door to the viewpoint that it’s had benefits on other areas of the system.

The Odroid N2 got like 30% boost in performance (peak performance) to the N2+ but people said they don’t notice a difference. Don’t think anyone said they did. yet this DTB alterations has like 80%+ saying they can actually see the difference (because this change improves minimum performance, the maximum performance was already good that it’s not changed much, in smaller tasks)

you also have hdmkv who has enough hardware to see the differences and says it’s snappier. and other posters who say what took a while on the library is now instant.

I don’t think it’s a question of if it has an effect, more so of how much of an effect is it having. that’s if you really want to know those stats and willing to put in the effort to find some benchmarks. but it’s gone past the point , “is this a placebo”

We can speculate ad infinitum, but everyone’s time would be better served by just taking multiple measurements.

This is the minimum rigour you’d expect from any claim of a small change. Then ideally others would reproduce that difference.

The world (and AV community) is full of examples of people being fooled by placebo, so yes it’s always a possibility.

4 Likes

Also would be good implement to CE porting of 3D games from opensource world for better performane benchmarks without needing emulation. Good games are for example: tux racer, super tuxkart and supertux. They have fps measurement. But is too many work for CE team. But maybe emuelec would be use for testing of new dtb. But 64bit CE need more optimalization than 32bit due to lack of 50% performance against 32 bit on s922xj in Bettle Saturn kernel in retroarch (but maybe only issue of bettle saturn).

In retroarch in terms of coreelec is used openal in coreelec or only augesound? I ask therefore that, i found aeonwave which is 5 - 7 times effectiver as openal(-soft) for linux with EAX support according to author.

Game Consoles

Game consoles (including desktop and laptop PC’s) benefit from all features described under Immersive Simulation, Studio Effect Mixing and Embedded systems. But most of all game consoles benefit from the fact that AeonWave renders audio 60% to 400% faster than other libraries.

… in this time run process of fixing yet found bugs.

I answer alone. CE dont support Openal in retroarch now, although openal have better performance as alsathread according user of odroid:

https://forum.odroid.com/viewtopic.php?t=7580

How do i know if i have this new dtb file on latest nightly? Is there anyway to check?

Connect to the box over ssh: lscpu
you should see this
Caches (sum of all):
L1d: 192 KiB (6 instances)
L1i: 320 KiB (6 instances)
L2: 1.3 MiB (2 instances)

Thank you, it shows exactly that.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.