Running The SteamOS Kernel On Ubuntu Linux

[ninez]

Yes, this would be very useful. Input latency is very important factor in gaming. I will happily sacrifice some FPS for lower input latency. Most likely Valve have used real time patches to lower audio and input latency.

probably...and i am sure they want to be able to have granular control of the rtprios of IRQs, kthreads, etc.

It's also no secret to anyone who is familiar with patching nvidia for -rt, that Nvidia (finally circa 325xx+ driver) included the spinlock code into their driver (from old nvidia-rt/compat patches) AND added the IGNORE_PREEMPT_RT_PRESENCE=1 flag into their driver ~ to allow PREEMPT_RT_FULL kernels to compile/use nvidia, where previously it would fail. ~ I had previously thought they were just improving nvidia for -rt (because it is used professionally in some areas, like flight simulation, for example).. But now, it's very obvious (to me anyway) that this work has been done for Valve/SteamOS.

So this is about more than just input/audio latency - this is probably about getting the best deterministic behavior and performance out of the nvidia driver, itself - ie: they probably don't want nvidia being interrupted by certain OS level operations and/or user space stuff. (same goes for input handling, audio device's IRQ, etc)...

PS: they are also packages jack2 (1.9.8) in their repos, which i thought was interesting.

[hjhamala]

I agree. This should be tested in some way, but I suposse it's quite more difficult and maybe it requires special hardware.

What about the same tests with the same kernel and P-state enables? And 3.13 with equivalent patches and P-state enabled too?

Eurogamers Digital Foundry did some tests few years ago:
http://www.eurogamer.net/articles/di...factor-article

It is definitely doable but automation would be pretty hard.

[asdfblah]

This is likely due to the P-State driver not being used on the SteamOS kernel

You could have proved it by running both tests under the "performance" governor. In fact, you should add that to your testing setup: a way to test things under different CPU governors, although it obviously would make tests longer.

[Michael]

You could have proved it by running both tests under the "performance" governor. In fact, you should add that to your testing setup: a way to test things under different CPU governors, although it obviously would make tests longer.

PTS already detects and report governor information and you can run whatever test you want in any situation.

[serus]

I would like to test the current SteamOS Kernel in Ubuntu. Could someone please explain me the process or hint me to resources that help understand what I have to do?

Also, is someone actually using this combination and what are more recent impressions?