AH2 cpu limited?

[eagl]

First off, this is NOT a gripe or complaint.  I'm happy with the new vid card I got and AH2 looks nice and is playable.  This is supposed to be a discussion, not a gripe session.  I put this in general discussion instead of hardware/software because it's not really aimed at any hardware in particular, and I already confirmed by reading a few virtually ignored threads that some others have noticed something similiar.  Anyhow...

Has anyone noticed that AH2 framerates seem to be cpu limited?  I've been playing around with 2 different vid cards on my system and it seems like I can use wildly different graphics settings and get very little change in framerates.  Here's what I started with and what I did.  I'm also curious as to if HT is coding exclusively on intel processors and maybe inadvertently using intel specific instructions so AMD systems get handicapped, because my results were a bit unexpected.  Or maybe my system just sucks, but I'm not sure about that.

First, system specs.

AMD Athlon XP 2000+ "thoroughbred", 1.67 ghz, 133mhz FSB
512 meg DDR 266
SB Audigy2 ZS
windows XP Home

The rest of the system doesn't really matter much, except that it's all high quality components and stable.  The AGP bus is confirmed working at 4x.

I started with a GF4-4200 and noticed something odd - at 1024x768, I got essentially the same framerates using no FSAA and 2x FSAA.  In AH1, running 2x FSAA would usually result in a 30% or more framerate drop at the same resolution.  Further experimentation with FSAA and ansio/mipmap filter settings seemed to show that I could really work the video card or just let it loaf along, and my framerates stayed almost the same.  The only way to get better framerates was to change the sliders, but even then it was tough to get consistent framerates over 30 when near fields.

I just got a nice new Nvidia 6800 GT today and installed it.  I tested it first with doom3 and it worked exactly as you'd expect - much better image quality, double the framerate even with higher resolution and detail set, everything about it was better.  Then I tried AH2 and got...

No improvement.  I think I might have gotten 3 fps more.  Again, I ran through some tests changing resolutions, fiddling with FSAA and ansio settings, and again it seems like AH2 is badly cpu limited.  My shiny new 6800 GT gets me barely 30 fps sitting in the tower with all three of the sliders set in the exact center and a resolution to 1024x768, which is about what my GF4-4200 would get me.

Another oddity - When running AH1, I could alt-tab out to the system and when back at the desktop, things seemed to respond normally.  Windows would pop up quickly, I could check email and browse the web with AH1 running happily in the background.  I could even surf the web or check email with an AH h2h host running and nobody in the arena (full with me and 7 others) noticed any warps at all.  But when I alt-tab out of AH2, the system responds very very sluggishly.  Even bringing up folder views takes 10-15 seconds.  AH2 is clearly hogging the cpu far more than AH1 did.

So what's the big bottleneck here?  Are all the new graphics features being offloaded to the cpu?  Since dropping all the sliders is the only thing that increases framerate, even more so than dropping resolution, is all that detail being produced by the cpu instead of the video card?

Is it an AMD vs. Intel issue, using SSE2 or whatever instructions?  Are the compiler flags set to P4 uber alles?

Inquiring minds want to know, especially before they spend money on new hardware.  If it's an AMD vs. Intel issue, knowing that beforehand could prevent spending a whole lot of coin on another ineffective upgrade.

[Karnak]

What I do know is that after seeing AthlonXP, Pentium4 and Athlon64 benchmarks my next planned upgrade switched from an AthlonXP 3200+ to an Athlon64 3500+ or 3800+.  There is just no comparison.

The players with Athlon64s seem to get great framerates, so I don't think this is an Intel vs AMD issue.

[humble]

Frame rate is not CPU limited its operating system/monitor limited. You can't generate FPS faster than your monitor will refresh. Windows XP limits FPS to 75/sec for this reason. You can overide that setting but "V-sync" will still limit you to your refresh rate (usually 65-85 FPS). If you turn V-sync off then you'll show FPS in the 100-300 range....BUT....your monitor will only show every 2nd third or fourth frame...or worse hangup/CTDT etc.

All of the newer cards are basically VPU's with a vast majority of rendering onboard the card. The true bottlenecks are in the system memory (2100 vs 3200 for ex) and in the "front side bus" 266 vs 800 or even 1600+ in new AMD socket 939 boards. However this all goes back to your monitor...currently the "true" bottleneck.

[humble]

Hmmm rereading your post you've got other issues. I run a Ti-4200 on a similiar system and I'm pegged at 75 FPS under almost all conditions...in a low furball over a big base I'll drop a bit...but never enough to really notice (lets guess 48-55 FPS)...you have some type of driver/resource issue I think....

[humble]

Originally posted by Karnak
What I do know is that after seeing AthlonXP, Pentium4 and Athlon64 benchmarks my next planned upgrade switched from an AthlonXP 3200+ to an Athlon64 3500+ or 3800+.  There is just no comparison.

The players with Athlon64s seem to get great framerates, so I don't think this is an Intel vs AMD issue.

hehe....

Me to, was pricing stuff today wondering if it makes sense yet...

[Nilsen]

humble

if im correct i think he is asking if AH is a game that needs a fast cpu to render good framerates and not just a good graphics card.

eagl?

[humble]

Originally posted by Nilsen
humble

if im correct i think he is asking if AH is a game that needs a fast cpu to render good framerates and not just a good graphics card.

eagl?

If I'm correct the decision on what resources to use is controlled by the vidio card not the "program".  The "VPU" handles all it can internally and off loads what it cant to the CPU...I'm not a techie but I don't see enough "eyecandy" to cause a massive bottleneck. Since he has a very modest frame rate I'd guess eagl has some other issues to resolve...

[Hyrax81st]

Originally posted by eagl
First, system specs.

AMD Athlon XP 2000+ "thoroughbred", 1.67 ghz, 133mhz FSB
512 meg DDR 266
SB Audigy2 ZS
windows XP Home

The rest of the system doesn't really matter much, except that it's all high quality components and stable.  The AGP bus is confirmed working at 4x.

This is my exact baseline system with the exceptions noted below:

1. I have 1GB RAM DDR 266.
2. I have an ATI Radeon 9800 XT (256MB Ram) with all AA/AS stuff turned off.
3. I am using Omega's optimized drivers for my card and it is running at 8x AGP.


I am loading all textures directly onto Video card with the AH2 video setup option - this eliminated all "stutters" for me. I am generally setup with all 3 graphics sliders set to midway point in-game. I get 50-60 FPS near the ground and if in dogfight up high ( I set sliders to max performance rather than detail, then), I get 100FPS (and yes, my screen refresh rate "hertz" is set higher to allow true frame rate of 100 in-game).

Yes, CPU seems to be important (at least on "lower end" machines like ours ?) - which is why I am using FSAutoStart by Ken Salter. This is a nice utility that helps "shutdown" processes clogging up the CPU that are not necessary for the game. After you exit the game, FS restarts those processes and you are back to where you want to be. As the name implies, he wrote it for Flight Simulator folks, but it can be used for just about anything. I have used this for 3 months now and, overall, I have probably been "disco'd" less, had fewer VOX dropouts and smoother rides in AH2. Check it out at http://www.fs-gs.com/downloads.htm

[eagl]

Humble,

I have my refresh rate set to 85, but in the tower I'm getting 25-30 fps, nowhere near the monitor refresh rate limit.  You say a "similiar" system...  intel or AMD processor?

The amd vs. intel thing could still be an issue even though the new athlon 64 processors are doing so well, because that's a whole new generation of cpu and it incorporates the P4's instructions where the athlon XP's don't.  I've run across many examples where products compiled with pentium 4's in mind simply run like crap on athlon XP processors.  Since the athlon 64 was made after the P4, the athlon 64 can handle those instructions just fine.

When I installed the new card, here's the steps I took -

Uninstalled nvidia drivers
ran detonator destroyer
went into device manager and removed old vid card
shut down, swapped out vid cards
booted up, cancelled out of windows hardware installer
ran latest nvidia detonator driver installer
rebooted a couple times while getting things set up

tried doom3, it works like a champ (very fast, smooth, great image quality) except for tiny pauses during HD accesses, but I don't see those pauses in AH.

Tried AH2, and had my disappointing results.

I'm willing to admit my computer or configuration might be the problem but it's the same config that gave me such great results in AH1 and such consistent results in AH2 as far as seeming to be cpu limited instead of video limited.  My mobo is an ABIT KR7A with the VIA KT266A chipset and although earlier versions of the via chipset had severe problems, I never saw the stability or gross performance problems so I don't think that's the problem.  I've run this system for over a year with quite good and consistent results.  With the GF4-4200 card, I could get 3dmark2001 scores over 10,000 which doesn't mean anything except that my system was performing almost exactly as it *should* given the hardware I have and the scores other people report.

I'll run some benchmarks and toss them in here.  They don't mean much except that a really poor benchmark score would point towards some sort of system problem.

edit - 3dmark2001SE - 12706, 2000 pts higher than my best score with the GF4-4200

3dmark2003 - 9215

So the basic benchmarks don't seem too far off what I'd expect...  Nothing to brag about but not grossly low, right?

[Octavius]

Eagl, I agree.  I've seen similar results on similar rigs.  AH2 on another 1.5ghz AMD with the same 9600xt performs decent after tweaking the sliders a ton (in game).

On my AMD 64 rig, framerates soar with the very same 9600xt.  But at the same time, the bane of my system are trees and lots of them.  Bases, heavy aircraft presence doesn't seem to have a large effect.  

Unfortunately I haven't had the chance to test anything with a higher performing vid card.  I assume that it is the bottleneck with a 64bit CPU.  

Doom3 and AH2 are vastly different games graphically.  Many shaders are present in Doom, it's an indoor, small environment with plenty of textures.  AH2 is 'outdoor' with a vast area to draw.  Many textures here as well.  Throw in a bunch of monsters or aircraft and you'll get slowdowns with relatively slower rigs.    Above all, both games have entirely different engines.  

btw, Doom 3 rocks even with a 9600xt.

[eagl]

I fired up AH1 just to see what I'd get, and at 1280x1024 it's pretty much pegged at my refresh rate of 85.

[eagl]

A final test

Loaded up the game, 1280x1024, 2x FSAA, and same results - 25ish fps in the tower and any time either trees or buildings were in view.  So I tried something different.  In this test, I was in an area of rolling hills, a bit of a mountain on one side, and some water in there too, near a small field and town.

Looking at ground with both trees and buildings visible - 25 fps
Looking at ground with only trees visible, no bldgs - 40 fps
looking at ground with only bldgs visible, no trees - 40 fps
looking at ground with no trees or bldgs visible - 70 fps

So I tried something - I turned the first 2 sliders all the way up to the max, and turned the bottom one to the min.  That way nothing but buildings at the airfield and towns would show up if I climbed some.  I climbed up a bit and flew away from the field and town until only some of the buildings showed up, but no trees, and then I did another test - if I turned and looked out so the buildings were just barely at the edge of the screen but still visible waaay far away, I got 44 fps.  If I added a tiny bit of rudder so the bldgs disappeared outside the edge of the screen, the framerate immediately jumped to around 70 fps.  So even with the buildings very very far away, really only a cluster of dots, they're still sucking around 30% of the framerate.

And I can adjust my resolution up as high as I want, add FSAA and all that, and my framerates remain the same.  That really sounds like it's cpu limited.

I know AH2 is nothing like doom3, it's just weird how the framerates change so dramatically like that even when looking at objects that are far away.  The buildings and trees each eat about 30% of the terrain-only framerate.

I think I'll just end up running with all the sliders almost full right, the resolution at 1280x1024, and 2x or 4x FSAA.  Ansiotropic filtering seems to make the trees shimmer but if I turn the sliders down, the trees disappear so maybe I'll turn some ansio on too.  And hopefully HT can find some way to transfer some of the cpu load onto the vid cards so the buildings and trees don't kill framerate at all resolutions

[humble]

I'm running a XP 2200+ on a Gigabyte 8xagp board with 512 DDR(pc2700)mem and a PNY Ti-4200 VC. My MB is a much better one, is a 400 FSB board. Think it runs at 266 with my chip...but not really sure. Other than that system not to far off. Certainly not by as much as your FPS indicates...

Also went back and checked...my FPS not as good as I thought.
In "1st" tower bout 42 FPS...a bit better if I go out to a field ~47FPS. Normally pegged out in flight/combat at 75 FPS with dips to 40-50 FPS down in the weeds at worst....

[eagl]

What resolution?  What texture size?  Where are your sliders again?  

You're likely running 333mhz FSB (166x2) instead of 266, but I'm not positive because it's possible to make AMD cpus run at all sorts of odd FSB/multiplier combinations and you can even run your memory asynchronously from the cpu bus on some mobos which can either help or hurt depending on the application or game you're trying to run.

It's possible you're making up the performance on both the additional cpu and FSB speed.  133/266 memory is PC2100 if I recall correctly.

I've rechecked, and I'm not running anything in the background that would account for any measurable cpu load.

[Karnak]

With a 2200+ he'll be running at 266mhz on his FSB.