ATI 3870X2 CrossfireX Review - Part II: The DX9 Games

[Lupine]

As promised in the closing comments of part I, Rage3D is back with even more testing of the new ATi 3870X2, this time abandoning the new-age glitter of DX10 titles in order to explore the well established maturity of DX9 ones. Stick with us to see what happens when a pair of new GPUs deals with an old challenge.

Check out Rage3D's ATI 3870X2 CrossfireX Review - Part II: The DX9 Games

[jam2k]

Great !

[friskyolive]

In a couple of areas that required ATI confirmation, and "looking into" sure sounds encouraging. Along with that, a very simple to understand, and great review on assigned DX9 games!!

[narciso]

The stuttering in the witcher game is very noticeable. But I also had it when I had a 8800GTX, so it must be the engine itself. But ATI driver support for this game is not gonna hurt it

[TheNGTV]

Err... in STALKER, shouldn't the "16 RBEs" be 16 "TMUs"?


Because RBEs definitely don't do AF.

[-n7-]

Great review, though you really need some 2560x1600 action...

But now, sorry, but i have to do some bitching here due to a pet peeve.

It seems nearly every damn review i see does silly UT3 fly-byes
And then they explain that z0mg, the game is CPU limited.
That's such BS though.

While it is indeed true that UT3 loves a strong CPU (including preferring quads over duals), it's a hell of a lot more GPU intensive when doing botmatches, which are a FAR more realistic representation of gameplay.

UT3 can actually get quite low with fps when testing in some maps during botmatches...i would know.

I've not only played well over a hundred hours worth of UT3, i've also benched it using the exact same tool you guys did for my review in the General Graphics section.

Sure, there are wider variations in fps during a botmatch, but that's what multiple runs averaged are for

[Monrad]

Originally Posted by -n7-

Great review, though you really need some 2560x1600 action...

But now, sorry, but i have to do some bitching here due to a pet peeve.

It seems nearly every damn review i see does silly UT3 fly-byes
And then they explain that z0mg, the game is CPU limited.
That's such BS though.

While it is indeed true that UT3 loves a strong CPU (including preferring quads over duals), it's a hell of a lot more GPU intensive when doing botmatches, which are a FAR more realistic representation of gameplay.

UT3 can actually get quite low with fps when testing in some maps during botmatches...i would know.

I've not only played well over a hundred hours worth of UT3, i've also benched it using the exact same tool you guys did for my review in the General Graphics section.

Sure, there are wider variations in fps during a botmatch, but that's what multiple runs averaged are for

I don't know if what you're saying about botmatches in UT3 makes sense. If you play a multiplayer game with 8 enemies and the game doesn't slow down, why would it slow down when you have 8 bots? I guess it's because CPU power is not enough to calculate everything + AI.

Now... if the game is also slowing down when playing against real people it's probably GPU limited as you said.

[icecold1983]

Originally Posted by -n7-

Great review, though you really need some 2560x1600 action...

But now, sorry, but i have to do some bitching here due to a pet peeve.

It seems nearly every damn review i see does silly UT3 fly-byes
And then they explain that z0mg, the game is CPU limited.
That's such BS though.

While it is indeed true that UT3 loves a strong CPU (including preferring quads over duals), it's a hell of a lot more GPU intensive when doing botmatches, which are a FAR more realistic representation of gameplay.

UT3 can actually get quite low with fps when testing in some maps during botmatches...i would know.

I've not only played well over a hundred hours worth of UT3, i've also benched it using the exact same tool you guys did for my review in the General Graphics section.

Sure, there are wider variations in fps during a botmatch, but that's what multiple runs averaged are for

its impossible for me to get a single gpu limited scene in ut3 for me at 1920x1200 without aa. the engine just chokes on itself constantly.

[-n7-]

I don't want to take this thread off topic.

But here's just a simple example in DM-ShangriLa.
1920x1200 maxed flybye
53.29 FPS average, 99.57 percent of time spent > 30 FPS

1920x1200 maxed botmatch
45.63 FPS average, 95.58 percent of time spent > 30 FPS

Botmatch is roughly 85% of the performance of the flybye.

And minimum fps, which unfortunately i'd have to run Fraps for (& didn't bother with this time), is lower during botmatches as well.

And if you assume it's a CPU limitation, you'd be wrong, unless task manager lies.

Not one of the four cores in my quad ever even hits 100% during those runs.

Anyway though, PM me if you guys have more UT3 benching questions; i don't want to take this thread off topic, sorry,

[AlexV]

Originally Posted by -n7-

I don't want to take this thread off topic.

But here's just a simple example in DM-ShangriLa.
1920x1200 maxed flybye
53.29 FPS average, 99.57 percent of time spent > 30 FPS

1920x1200 maxed botmatch
45.63 FPS average, 95.58 percent of time spent > 30 FPS

Botmatch is roughly 85% of the performance of the flybye.

And minimum fps, which unfortunately i'd have to run Fraps for (& didn't bother with this time), is lower during botmatches as well.

And if you assume it's a CPU limitation, you'd be wrong, unless task manager lies.

Not one of the four cores in my quad ever even hits 100% during those runs.

Anyway though, PM me if you guys have more UT3 benching questions; i don't want to take this thread off topic, sorry,

That might be correct for some runs, but for others they end up faster, and the bots sometimes do retarded things like stare at the blank sky. Botmatch was tested initially, but was sacrificed for the sake of more accurate repeatability...although the way of testing UT3 will evolve in the future, based on your feedback. Thanks!

[AlexV]

Originally Posted by TheNGTV

Err... in STALKER, shouldn't the "16 RBEs" be 16 "TMUs"?


Because RBEs definitely don't do AF.

Yes it should, that's a typo we missed as this has been in limbo for quite a while. As soon as the editor-in-chief gets on it it'll be fixed. Thanks!

[AlexV]

Originally Posted by -n7-

But now, sorry, but i have to do some bitching here due to a pet peeve.

It seems nearly every damn review i see does silly UT3 fly-byes
And then they explain that z0mg, the game is CPU limited.
That's such BS though.

Conceptually I agree with you, but, as I said above, botmatch was tested, and it didn't change things significanty. A few FPS at most, and based on some other testing done it did like a faster CPU to a certain extent. The best way of testing UT3 would be actually FRAPSing through a botmatch I think, instead of using UT3Bench(which is an awesome tool, but the botmatch testing part varies too much IMHO)-wouldn't you agree?

[icecold1983]

the game is more cpu limited in gameplay than fly bys.

[Skinner]

It's not only missing GPU-power. Edit, I mean CPU-power!
I have a lot of experience with (quad)CF and I can tell you its the CF profile.
When renameming UT3 exe to FEAR, the min fps is much, MUCH, higher. So the CF profile for UT3 and many other games is far from finished.

COD4 has an bug when runing with CF and mediuim anumated water. When setting water to mostly static= low, CF produces much higher framerates.

One thing which is showing it's ugly head with quad CF, is when AI or any other form of (CPU) action is coming along, the tandem will start to choke.

I hope people, ATI, game devs, will look with an sharper eye to solve this, cause CF is much potential.

[SSXeon]

Originally Posted by Skinner

COD4 has an bug when runing with CF and mediuim anumated water. When setting water to mostly static= low, CF produces much higher framerates.

Its not just CF, did it on my 2900XT and HD3870, for some reason low just runs faster.

As for the profiles, this has been asked many times with almost no response. I hope they are working on this for the R7xx series.

[AlexV]

Originally Posted by SSXeon

Its not just CF, did it on my 2900XT and HD3870, for some reason low just runs faster.

As for the profiles, this has been asked many times with almost no response. I hope they are working on this for the R7xx series.

The COD4 waterbug has been mentioned in the review. It did not affect the run used. It's also apparent in other titles-similar symptoms, albeit not related to the presence or absence of a water shader.

Personally, I haven't experienced the CPU+QuadCF choking in titles which have proper profiles, and I've been extensively using it for quite a while. Could guidelines for replicating this be provided?

[Skinner]

Originally Posted by Morgoth Bauglir

Personally, I haven't experienced the CPU+QuadCF choking in titles which have proper profiles, and I've been extensively using it for quite a while. Could guidelines for replicating this be provided?

Try Time shift, openingsscene (2 fps) and UT3, Warmap Marketplace, it's dips badly, less then 30 fps.

[AlexV]

Originally Posted by Skinner

Try Time shift, openingsscene (2 fps) and UT3, Warmap Marketplace, it's dips badly, less then 30 fps.

Both of those aren't related to AI/CPU usage, but are known issues WRT profiles. Timeshift works fine in that scene with a single X2, and UT3 isn't yet optimally profiled in the drivers, and has issues with some maps AFAICT.

I've asked about ways of replicating this because I've only seen it mentioned by a source that comes to somewhat odd conclusions at times, like this one or one about the X2 and Jericho based on the heat output of the card-both of which are the results of misunderstandings.

As I said, most of the times it's an issue WRT a profile being suboptimal/breaking down, and since generally the most complex interactions happen with NPCs/AI controlled entities on-screen, the tendency might be to pen down the behavior on AI/CPU load causing QuadCF to falter. AFAICT, that's incorrect:I've been through games that were better at threading than even UT3(DiRT, for example), and in spite of high CPU usage they scaled quite well to Quad, and had no odd slowdowns.Cheers

[Skinner]

What' s WRT ?

[friskyolive]

Originally Posted by Skinner

What' s WRT ?

With Regards To?

[AlexV]

Originally Posted by friskyolive

With Regards To?

Yes.