Why does Crossfire run better at higher resolutions?

[AndroidVageta]

Case in point...I have Crossfired 4890's with 1GB of RAM...my question is simply this:

Why do games yield better performance when using higher resolutions with Crossfired enabled?

My monitor is a 22 inch LCD with a max resolution of 1680x1050...playing GTA 4 with the memory limit unlocked with everything set to max (no AA) at 1680x1050 I got an average of about 30fps with smooth to slightly choppy playback depending on what was going on.

So just for the hell of it I hooked up my old Compaq P1210 CRT (which Im going back too now lol) just to see how much GTA 4 at max settings at 2048x1536 would kill my system...HOLY SHIT!!!

Not only does running GTA 4 at this high as shit resolution give me 40fps average but game play is soooo much smoother!

How is running a game at a higher resolution giving me better FPS then at a lower one? This proves true with Crysis too...at max settings I got 35FPS avg @ 1680 vs. 41FPS avg @ 2048...how?

 

[x3sphere]

Not possible really.. might just be a placebo effect. How did you check the average FPS? If you did it during normal gameplay that's a variable test as certain points of a game can be more stressful than others. 35 vs 41 FPS is not a big difference and suggests that you're CPU limited at 1680 given the amount of GPU power at your disposal. Also, CRTs have a faster response time so that could be influencing your perception of smoothness too.

 

[XpediencY]

Running at a higher resolution gets rid of the bottleneck put on by your CPU. Since you have more GPU power vs CPU, putting more stress on the GPU is better than stressing your CPU.

 

[AndroidVageta]

I did the GTA 4 test doing a planned walk through using FRAPS to find my results...Im telling you, GTA 4 is actually FASTER at 2048 then 1680...atleast on my PC it is...I find it odd as well, believe me lol!

 

[sweetreddyd]

Increasing the resolution on GTA IV has a side effect of forcing many of the other options lower.

It's very likely you are running a higher res now, but have another, more taxing detail tuned lower, especially draw distance and texture detail.

I noticed GTA IV ran faster at 2560 than it did at 1920, and then later noticed my texture detail had dropped to medium from high without permission.

 

[kllrnohj]

Running at a higher resolution gets rid of the bottleneck put on by your CPU. Since you have more GPU power vs CPU, putting more stress on the GPU is better than stressing your CPU.

Bottlenecks don't work like that. If the CPU is bottlenecking the game to 30 fps at 1680x1050, then increasing the resolution won't get rid of the CPU bottleneck, it simply makes it more likely that the GPU will become the bottleneck at an FPS lower than 30.

@OP: You also changed your AR, which could mean that the card is doing less work overall since rendering a static sky and ground is pretty simple. Sounds to me like your testing method has issues, though. Like if you left vsync on with your LCD and disabled it with your CRT or something like that. Or you disabled AA or changed the graphics settings (or the game did automatically when you changed resolution)

 

[x3sphere]

I did the GTA 4 test doing a planned walk through using FRAPS to find my results...Im telling you, GTA 4 is actually FASTER at 2048 then 1680...atleast on my PC it is...I find it odd as well, believe me lol!

Like I said, I have a feeling that is just because the CRT has a near instant response time. Input lag can influence perception of smoothness.

With GTA IV, I do not get better frames at 2560x1600 versus 1920x1200 on my setup and I have an 5970 which is in effect a Crossfire card.

 

[bigdogchris]

Not possible really.. might just be a placebo effect..

Completely possible.

1680x1050 is way to short on pixel count for X-fired 4890's; to much overhead. You're better off running just 1.

Now it's not like that in all cases but I run at 1440x900 and used to notice a 40-50% FPS boost when I sli'ed my 8800GTS 512 (I've since moved to single gpu solution).

Truth be told, most people buy way to much GPU for what they really need. I've read a few places that did some reviews on memory. Even with 1080p+ resolution gaming, you don't need over 512MB. Now I don't suggest running out to buy a 512MB card now instead of a higher end GPU because the processing power on those cards are pretty poor. As for the processing power, a single 4890 is actually more than you need for 1680x1050 gaming.

 

[Artificiary]

I can't imagine it making that much of a difference, especially since you're rendering nearly twice the number of pixels at 2048, but maybe the change in aspect ratio has something to do with it?

 

[x3sphere]

Completely possible.

1680x1050 is way to short on pixel count for X-fired 4890's; to much overhead. You're better off running just 1.

Now it's not like that in all cases but I run at 1440x900 and used to notice a 40-50% FPS boost when I sli'ed my 8800GTS 512 (I've since moved to single gpu solution).

Truth be told, most people buy way to much GPU for what they really need. I've read a few places that did some reviews on memory. Even with 1080p+ resolution gaming, you don't need over 512MB. Now I don't suggest running out to buy a 512MB card now instead of a higher end GPU because the processing power on those cards are pretty poor. As for the processing power, a single 4890 is actually more than you need for 1680x1050 gaming.

While it might be overkill, there's no reason why you'd be getting less performance at a lower res due to increased overhead. I've never seen any testing from professional reviews suggest this. Really, the OP's results suggest he is hitting a CPU wall at 1680 and seeing the around the same frames (his testing is variable so it's not going to be 100% accurate) at a higher res seems normal to me, for reasons I explained in my other post.

 

[bigdogchris]

While it might be overkill, there's no reason why you'd be getting less performance at a lower res due to increased overhead. I've never seen any testing from professional reviews suggest this. Really, the OP's results suggest he is hitting a CPU wall at 1680 and seeing the around the same frames (his testing is variable so it's not going to be 100% accurate) at a higher res seems normal to me, for reasons I explained in my other post.

Regardless of what you explain, I've encountered this personally.

Have you ever owned an SLI/X-Fire system?

 

[AndroidVageta]

My testing is not varied AT ALL!

I have GTA 4 edited so the RAM limit is off...I am running GTA 4 at MAX settings (including textures) and a 2048x1536 resolution...all in all using about 1600mb of RAM according to GTA 4...so setting something higher doesnt automatically reduce anything.

Same with Crysis...nothing changed except the resolution.

Running a quad core Xeon 3350 @ 3.2ghz a core.

 

[x3sphere]

Regardless of what you explain, I've encountered this personally.

Have you ever owned an SLI/X-Fire system?

Not a traditional setup with two cards but the 5970 which I am using now utilizes CrossFire to communicate between both GPUs on a single card so I don't see why it'd be any different.

 

[Blackbird0147]

Games use more power when they are running on higher res. I think ATI configured their divers so that higher res causes the second gpu to work more.

 

[infect0]

Games use more power when they are running on higher res. I think ATI configured their divers so that higher res causes the second gpu to work more.

+1

I do not own crossfire but sounds like poor hardware/driver optimization to me. I think that is the most reasonable explanation.

 

[MJZ]

This should not be happening, obviously. It sounds like either methodology flaw in the comparison or a driver issue.

I would be expecting 60+ FPS with low resolutions , not sub 30 FPS .

 

[cannondale06]

it can happen at lower resolutions due to overhead. I have seen cases where a single card will match or even be faster until the resolution is raised. with really fast cards and a mediocre cpu this is probably common. its likely pretty dependent on the actual game and driver being used too.