Latest Radeon Graphics Results On Ubuntu 14.10 Are A Bit Concerning

[dungeon]

I had to roll back Mesa to 10.4...

Hm, that then does not sound like what i pointed out . What kernel do you use?

I would like to bisect that, but can't critter work fine here so that does not apper here on radeonsi, so someone with r600 need to bisect that If you need guessing pointers, Dave Airlie from Red Hat played lately something with r600 dirver for Cayman so maybe his commits broke something

http://cgit.freedesktop.org/mesa/mesa/log/?qt=author&q=Dave+Airlie

[schmidtbag]

?? what you said just don't fit to what I wrote...
I never said that the radeon is faster than the catalyst.
Also I never said that the radeon does not need more speed.

so you disagree to the point that bugfixes,tearingfixes,stuttering fixes are good to have even if we lost some fps?

what is the alternative to this? a shit-load of crap with enormous high FPS rate?

I think you're interpreting my response as hostile, which I didn't intend.

What I'm saying is sacrificing FPS to fix stuttering or tearing isn't always the best option. So for example if you got 60FPS before but now this fix drops you down to 40FPS, your experience hasn't really improved much. Currently, this is really only a problem with GPUs that have been struggling to keep up to begin with. If you've got something high-end and a game it can handle then this regression (if that is what it is) is a non-issue.

But my main point is the open source drivers already performed worse than catalyst prior to this latest update, so having the performance drop even further is a bad sign. If the drivers performed on-par with catalyst WITH these fixes, then this wouldn't be an issue at all.


Remember, not everyone has high-end hardware, so there could be people now who were able to play a game with a smooth experience and now they won't.

[cocklover]

Well is funny because in other reviews of this site the performance of ATI have increased since 12.04 ...
anyways somebody knows why systemd is a dependancy of xf86-video-ati?

https://www.archlinux.org/packages/extra/x86_64/xf86-video-ati/

[Adriannho]

I'm so confused right now...wasn't it just 1 week ago that tests show a massive performance improvement for radeon git?

What am I to believe now?

[dungeon]

I'm so confused right now...wasn't it just 1 week ago that tests show a massive performance improvement for radeon git?

What am I to believe now?

This test is for r600 driver only (tested on HD 5830 card), i have APU using radeonsi driver and i don't see anything much changed ... as i point out up there there were some Cayman fixes/commits lately it seems (for HD 6950). If you ask me, Michel or someone with Cayman card should test what happened there for Cayman, there might be some improvments there or if not and if even Cayman regressed there then bisect and bug report about this r600 performance regression would be right thing to do .

[Luke]

Same results kernel 3.17 or 3.18

Hm, that then does not sound like what i pointed out . What kernel do you use?

I would like to bisect that, but can't critter work fine here so that does not apper here on radeonsi, so someone with r600 need to bisect that If you need guessing pointers, Dave Airlie from Red Hat played lately something with r600 dirver for Cayman so maybe his commits broke something

http://cgit.freedesktop.org/mesa/mes...&q=Dave+Airlie

Same results kernel 3.17 or 3.18, yes, this is r600. I run Mesa and X from a PPA, cannot bisect from source. If this commit was not in the very first version of Mesa 10.5, I could try updating from my cache of packages one day's worth at a time until it is found, but that's a lot of work. I've got one of the last 10.4 versions in now and it works well.

Now on to the flames this has generated: Mesa GIT code is ALPHA code, it is not expected to always work right! New code has to be tested so bugs like this can be found. Bringing in new features can create bugs or bottlenecks, this is how they are found and fixed. I will test new versions as I get them, roll back if still slow, and report my results here now that this is under discussion and likely to be seen by radeon devs. I still don't have the right place to post a bug report, and the only places I have accounts are here and Launchpad.

[necro-lover]

I think you're interpreting my response as hostile, which I didn't intend.

What I'm saying is sacrificing FPS to fix stuttering or tearing isn't always the best option. So for example if you got 60FPS before but now this fix drops you down to 40FPS, your experience hasn't really improved much. Currently, this is really only a problem with GPUs that have been struggling to keep up to begin with. If you've got something high-end and a game it can handle then this regression (if that is what it is) is a non-issue.

But my main point is the open source drivers already performed worse than catalyst prior to this latest update, so having the performance drop even further is a bad sign. If the drivers performed on-par with catalyst WITH these fixes, then this wouldn't be an issue at all.


Remember, not everyone has high-end hardware, so there could be people now who were able to play a game with a smooth experience and now they won't.

Dude no offence and i will make it short to not waste our both time.
You are wrong.

But why you are wrong?

60fps with stuttering is unplayable! (yes your magic 60fps)

But 30fps with perfect frame timing and no stuttering is perfectly
Playable and it just Doubles every frame so you get 60fps with perfect
Frame timing on very second frame.

You don't believe it? I really don't care!

In fact every anti stuttering patch is a gift and makes perfekt
Frame timing! And if its slower in benchmark bullshit fps meaning
It just don,t matters.

We play games and not fps bechmarks and stuttering is a bug to fix.

[necro-lover]

Dude no offence and i will make it short to not waste our both time.
You are wrong.

But why you are wrong?

60fps with stuttering is unplayable (...)
We play games and not fps bechmarks and stuttering is a bug to fix.

Simple example because why fps is worth nothing and frame
Timing is All:

Driver A makes 1 milllion fps and driver B makes 30fps

Driver A makes all frames in the First (1/60) second
The second driver makes a frame every (2/60) second.

The gaming experience is clear driver 1 is useless because
For the player it only makes 1 frame per second even if he can
Push 1 million frames per second.

The B driver makes 30 real frames and the player can see it.

So... FPS over all is bullshit.

[GreatEmerald]

This is certainly not a case where FPS was sacrificed for better latency. See the Urban Terror frame time. It shows that shortest frame times stayed the same, but the longest frame times became even longer; meaning that where it stuttered before, it now stutters even worse. Same with Xonotic results, max FPS is barely affected, it's the min FPS that took a dip.

[dungeon]

Ah, that might be not related to graphics driver at all... i missed that server CPU hardware is used there 2xOpteron with low single threaded ipc and ondemand cpufreq used

I don't believe by defaut that any gaming benchmark is done right with ondemand cpufreq used, no way