Windows 10 vs. Linux - Intel UHD Graphics OpenGL Performance

[LinuxID10T]

I doubt that. It's not the classical version of powersafe, where the cpu was just throttled to minimum speed. It's the intel_pstate governor. Sure there's some overhead of the powersaving mode, but it shouldn't make a huge difference, as we can see from the xonotic benchmarks which probably are more cpu-limited as the unigine benchmarks.

BTW. funny use of language "Actually, everyone else is wrong."

I believe the performance governor also sets the range of speeds for the integrated graphics as well. What would be interesting is to see the frequency the CPU and GPU are running at while the test is running. On the hardware I've tested, it makes a pretty big difference (at least on my i3 6100 and Core M 5Y10c.)

[LinuxID10T]

Doesn't modesetting force 2D rendering to be ran through 3D?

It means it is going to use a generic code path, which yes, in this case I believe runs it through the 3D hardware. This is similar to how it works on the AMD side. It seems like a lot of Linux distributions are defaulting to this now since the Intel SNA driver is kinda buggy. There is also a video at some Linux conference about the decision (maybe XDC?)

[ext73]

Is that true? 'Cause when I mentioned 4.14 on OMG! and said there were serious Intel regressions, I only got a couple of replies stating that nothing was wrong at most people's machines and that my hardware was to blame, etc.

Unfortunately, it's true - I'm building, among others optimized kernels for Ubuntu and derivatives - I tested this regression from the kernel version 4.14.0 to 4.14.4 - make many changes in the configuration and still the same bug - applies to machines from Broadwell up, for example Haswell is approx.

[edddeduck_feral]

The p-state governor can make a huge impact on results.

https://www.gamingonlinux.com/articl...ux.9369/page=6

this is why we recommend users change their p-state to performance when playing as the differences can be dramatic, it’s more noticeable on Vulcan games than OpenGL however depending on your setup you can easily remove frame drops and increase overall performance by 10% plus.

http://support.feralinteractive.com/...i_cpu_governor

[Michael]

The p-state governor can make a huge impact on results.


this is why we recommend users change their p-state to performance when playing as the differences can be dramatic, it’s more noticeable on Vulcan games than OpenGL however depending on your setup you can easily remove frame drops and increase overall performance by 10% plus.

Right but for the purposes of this comparison though was just an out-of-the-box OS comparison.

[edddeduck_feral]

Right but for the purposes of this comparison though was just an out-of-the-box OS comparison.

Of course this was not anything about your all default out the box testing method. I have found a lot of people think the intel_pstate will have little effect where in reality with games it can be a massive factor. I was just providing a little extra background and examples to how p-state does effect game performance.

This was the post in question.

I doubt that. It's not the classical version of powersafe, where the cpu was just throttled to minimum speed. It's the intel_pstate governor. Sure there's some overhead of the powersaving mode, but it shouldn't make a huge difference, as we can see from the xonotic benchmarks which probably are more cpu-limited as the unigine benchmarks.

[Michael]

Of course this was not anything about your all default out the box testing method. I have found a lot of people think the intel_pstate will have little effect where in reality with games it can be a massive factor. I was just providing a little extra background and examples to how p-state does effect game performance.

Ah okay, understood. I will probably run some fresh benchmarks in the days ahead on Linux 4.15 with Ryzen + Intel comparing the different governor options again these days and their impact on performance with the latest games for helping people understand the difference. It's been a few kernels since I last ran such comparison so should be interesting to see with the latest CPUfreq and P-State happenings.

[GreenReaper]

The people saying it makes a difference is right. It can also impact server latency, so be sure to test CPU performance bias on your web servers as well. Of course it makes a difference to wattage, too, and that may be a big consideration if you're paying for it, or having to cool it (as in Michael's basement den).

In many cases that's someone else's job, though, and part of what you pay for - so why pay the same for less?

If people would stop turning C-states off in the BIOS (they're a crucial performance tool for enabling turbo-boost, as well as saving some power) and turn performance bias on, they'd have the best of both worlds. The impact can clearly be seen on a moderately-used server.

[bwat47]

Michael do You know that kernels 4.14 and 4.15 have a mega bug [regression] in Intel i915 drivers ?!?! Somehow, for now, the Intel developers 'are not eager' to fix it
- for example:

https://github.com/NixOS/nixpkgs/issues/31999

https://forum.manjaro.org/t/poor-ope...nux-4-14/35453

https://bbs.archlinux.org/viewtopic.php?id=232354

Regards

GLX gears runs butter smooth here on very similar hardware (xps 13 kaby lake). Running kernel 4.14.4 on solus: https://i.imgur.com/fqq5Kci.png

[ext73]

GLX gears runs butter smooth here on very similar hardware (xps 13 kaby lake). Running kernel 4.14.4 on solus: https://i.imgur.com/fqq5Kci.png

That's how I wrote about it in this place

https://forum.manjaro.org/t/poor-opengl-performance-on-linux-4-14/35453/106