HD4870 performance is truly awful - fglrx Hardy x86_64

[Spanner]

I'm curious...in the 3D > more settings section of the ATI control panel, does changing either of the two slider controls cause ANY change of behavior here?

No, not at all. On that scene in ONS-Crossfire (see my 1st post above), before the game has started, I average about 35 fps with "Catalyst AI" on or off, and with vsync "Always on" or "Always off". Makes no difference.

That's a hard scene, for sure - there's a lot of landscape detail. But I wasn't expecting the 4870 to choke on it! During the game I can often get up to 60fps average, but even so there's still a noticeable amount of stuttering.

[Spanner]

I guess you use the wrong 32 bit opengl lib.

I don't think so because UT2004 ships with a 64-bit binary! But if it's true, please can you tell me:

a) how would I tell?
b) what can I do about it?

Sorry if these are "noob" questions but I thought I was doing everything right. I've re-installed Catalyst since plugging in my HD4870 - are there parts of my X server I need to reinstall too?

It almost feels like I'm CPU bound but I've got a quad core Q9450 and 4Gb of RAM - surely not?

[RealNC]

That rig should give you a minimum of 200FPS on that game. So it's 100% a software problem. A friend of mine bought this card. It's the fastest one I ever personally tested; I guess it's a MS Windows-only card for now though.

[Kano]

@Spanner

Most easy way is to install a 32 bit distro. Then you have much less problems and you could use more easy a working Java plugin for Firefox. If you install the "server" kernel then you can access all 4 gb. If you use the generic one without PAE then not, but it is very unlikely that you can feel a diff between 3 and 4 gb.

[Spanner]

@Spanner
Most easy way is to install a 32 bit distro.

I'm not going to go 32-bit over a graphics card - that's really the tail wagging the dog. But I may install a 32-bit distro to test this issue - if I do, I'll be sure to post my results here.

Then you have much less problems and you could use more easy a working Java plugin for Firefox.

Couldn't get much easier than it is now - I don't even know if it's OpenJDK or nspluginwrapper, but my Java is fine in FF thankyou!

Thanks again for the suggestions.

[ivanovic]

So what exactly are you using the "benchmark" this? If you are using early versions PTS: Ignore the values. At the beginning it was really a CPU benchmark due to the high number of bots (18 bots being used). Bots in this game use *huge* amounts of CPU power. Have you checked what you get in a flyby benchmark or with a low number of bots?

Your symptoms do sound like a CPU-bound scenario since the FPS do not drop when upping the resolution. A reason for this can be that there are too many bots on the map. Or something in your system is just setup in a bad way...

To give you some usable numbers I ran PTS here on my HD3850 with my Q9300 on a 64bit gentoo. This is what I get in a test with latest PTS (basically the git version, should be almost identical to 1.2.0-beta1):

#################################################
Unreal Tournament 2004 Demo:
Map: DM-Rankin Botmatch - Resolution: 1920 x 1200

91.237648 Average FPS

Average: 91.23 Average FPS
#################################################

[Spanner]

So what exactly are you using the "benchmark" this?

If you are using early versions PTS: Ignore the values. At the beginning it was really a CPU benchmark due to the high number of bots (18 bots being used). Bots in this game use *huge* amounts of CPU power. Have you checked what you get in a flyby benchmark or with a low number of bots?

My figure of 30fps comes from flying around in "free camera" mode before the match starts (ie no bots at all) and zooming back so that a lot of the scenery is visible.

Your symptoms do sound like a CPU-bound scenario since the FPS do not drop when upping the resolution. A reason for this can be that there are too many bots on the map. Or something in your system is just setup in a bad way...

Something is wrong, for sure. 1 core is working at 100%, but it can't be bots so I'm guessing it's fglrx?

Physically this is a newly built machine, so maybe I need to check that PCIe is working at full speed, memory speed is OK etc... and the OS is an older install that's had NVIDIA drivers in previously, so I'll try a fresh install too.

Your PTS results are encouraging - I knew from the reviews on this site that I should be getting better results, I just wondered if there was anything obviously wrong in my log/config files before I dedicate many evenings of fiddling around.

Thanks for your feedback!

[ivanovic]

Something is wrong, for sure. 1 core is working at 100%, but it can't be bots so I'm guessing it's fglrx?

No, ut2004 is 100% single threaded. That is one core will be used completely. Try to benchmark at normal gameplay, do not trust into values from such extreme situations. Better get latest PTS, install the test ut2004-demo and check how it fares. Using PTS 1.0.5 should be fine, just use the map I used, too (the deathmatch map).

Like already said: I don't think it is anything directly related to your graphics card. So nothing about PCIe or anything like this should change anything. It is unlikely that it is your memory speed. Either it is due to the extreme situation you are using (zooming out a lot *is* problematic, too) or your driver is not setup correctly.

[legume]

Unreal Tournament 2004 Demo:
Map: DM-Rankin Botmatch - Resolution: 1920 x 1200

91.237648 Average FPS

I just tried alone on rankin and can get good fps with my 3850 and old [email protected].

On the outdoor maps - like primeval it's down to 30 depending where I look.

It is smooth though - but I've only just got the demo and haven't played properly yet.

Maybe the 4870s' drivers need a bit more tweaking.

I actually lost a few fps on etqw when I took out my X1600 and put in the 3850 - the game also stutters/jitters when it's showing plenty of fps and I have CPU spare.

[Spanner]

Better get latest PTS, install the test ut2004-demo and check how it fares. Using PTS 1.0.5 should be fine, just use the map I used, too (the deathmatch map).

OK I'll give it a shot, it'd be nice to be able to make a direct comparison. I'll be Tuesday before I can post back results though.

Either it is due to the extreme situation you are using (zooming out a lot *is* problematic, too) or your driver is not setup correctly.

Well I guess the PTS results will confirm or eliminate the "extreme situation" explanation, but at the moment I suspect the driver.

FWIW KDE 4.1 looks OK but menu drawing has huge delays - lots of crap appears instantly where the menu's about to appear, then the menu itself draws maybe 2 seconds later. I had a low-end Radeon X300SE in my machine before the HD4870, and its KDE4.1 performance was superior. So yes I guess it's either the drivers or RMA time.

I really appreciate your thoughts & will post back results when I have them. On another note I was happily playing Wesnoth all the time I suffered with that crappy X300, so thanks for your efforts there too!