Governor for Phoronix benchmarks

[johnc]

I think nvidia binary and nouveau tests would also help, since their hardware is better than intel's.

I can tell you what a higher-end nvidia card + blob driver + ondemand governor + i7 CPU + Source engine looks like in practice: Really high frame rates with frequent (~every 5-10 secs) stutters. It makes the game unplayable. I'm not sure if a benchmark would necessarily pick that up. Forcing single-core rendering (which probably boosts the CPU frequency) or just switching to the performance governor resolves the issue.

[dietrdan]

I vote for [x] The kernel's default scheduler

To me it's looks like the default ondemand scheduler is broken (scales to conservatively / scales down to fast?), which needs fixing and maybe is now being addressed.
But it makes sense to benchmark with the default settings, which 95% of all users will find at their systems.

From time to time Michael can make benchmarks to show how big the difference between the ondemand & the performance governor still are.

[Temar]

I can tell you what a higher-end nvidia card + blob driver + ondemand governor + i7 CPU + Source engine looks like in practice: Really high frame rates with frequent (~every 5-10 secs) stutters.

Which game did you play? I have a GTX 580 with a i7 CPU and did not notice any such problems on Kubuntu 12.04 with the the default scheduler. I only tried Portal and L4D2 though.

[johnc]

Which game did you play? I have a GTX 580 with a i7 CPU and did not notice any such problems on Kubuntu 12.04 with the the default scheduler. I only tried Portal and L4D2 though.

L4D2, HL2... I think Portal as well.

[Gps4l]

That is a very good point about the low-end hardware being much less impacted. As I have stated in the past, there is not enough data to be pointing fingers at anyone or anything, and we need tests with Catalyst to know whether or not this is a driver issue or a governor issue. Monitoring of CPU frequency for these tests is also important.

I think nvidia binary and nouveau tests would also help, since their hardware is better than intel's.


Regardless of what's really going on, I'm still grateful for the radeon driver developers and what they have accomplished.

Serious sam 3 run like shit, allot of stutter, with the catalyst driver.

Changing the govenor to performance made things allot better.

Just two example from the steam forums.
Second thread is started by me.

http://steamcommunity.com/app/221410/discussions/1/882966056716541645/

http://steamcommunity.com/app/221410/discussions/0/864960353948684845/

This might have changed with the new kernel though.

[johnc]

I can tell you what a higher-end nvidia card + blob driver + ondemand governor + i7 CPU + Source engine looks like in practice: Really high frame rates with frequent (~every 5-10 secs) stutters.

Okay I take that back. I just tried L4D2 w/ the ondemand governor and it seems perfectly fine. So maybe a game or driver update fixed the stutter issues I was having before. I remember having them just a week or so ago.

[nachoig]

I vote for [x] The kernel's default scheduler

To me it's looks like the default ondemand scheduler is broken (scales to conservatively / scales down to fast?), which needs fixing and maybe is now being addressed.
But it makes sense to benchmark with the default settings, which 95% of all users will find at their systems.

Exactly. The end user doesn't care about configuring kernel's scheduler. The end user just plays with the system. If the default options has problems under games, the way to fix it is fixing these problems with patches, not with bullshit.

From time to time Michael can make benchmarks to show how big the difference between the ondemand & the performance governor still are.

This is a good idea.

[schmidtbag]

Serious sam 3 run like shit, allot of stutter, with the catalyst driver.

Changing the govenor to performance made things allot better.

Just two example from the steam forums.
Second thread is started by me.

http://steamcommunity.com/app/221410/discussions/1/882966056716541645/

http://steamcommunity.com/app/221410/discussions/0/864960353948684845/

This might have changed with the new kernel though.

Cool, thanks for the results. At this point this must mean the governor is the problem - due to Michael's nouveau tests suffering the same issue, this means that the radeon hardware isn't the problem either.

I find it ridiculous that this problem hasn't been caught earlier (not necessarily by Michael but by anyone). ondemand should not have this great of an impact on performance.

[curaga]

Cool, thanks for the results. At this point this must mean the governor is the problem - due to Michael's nouveau tests suffering the same issue, this means that the radeon hardware isn't the problem either.

I find it ridiculous that this problem hasn't been caught earlier (not necessarily by Michael but by anyone). ondemand should not have this great of an impact on performance.

It was, perhaps you missed the thread by Pontostroy?

[synaptix]

I can tell you what a higher-end nvidia card + blob driver + ondemand governor + i7 CPU + Source engine looks like in practice: Really high frame rates with frequent (~every 5-10 secs) stutters. It makes the game unplayable. I'm not sure if a benchmark would necessarily pick that up. Forcing single-core rendering (which probably boosts the CPU frequency) or just switching to the performance governor resolves the issue.

The same could be said for the Source engine on Windows as well. Disabling multicore rendering massively improves gameplay and makes it smoother. Issue with the Source engine itself, the "multicore" part of it is jacked up code by Valve.