A2 no longer benefits from 5870 crossfire

[node_runner 0]

It seems that A2 isn't getting any performance benefits from crossfire when using Radeon 5870 cards. I have 3 Radeon 5870 cards in a crossfire configuration. Every other game I own has an incremental performance boost from 40%-80% each time I activate another one of the cards. But with A2, I get no performance benefits at all over using a single 5870.

I have tried four versions of the Catalyst drivers so far with no luck: 9.10, 9.11, 9.12, and 9.12-hotfix. I'm using the 1.05 version of A2 and running the "Benchmark 1" scenario mission to gauge performance.

Here is an example of crossfire scaling on my system using Dirt2 as a benchmark:

Crossfire disabled (using a single card): 55fps avg.

Crossfire enabled using x2 cards: 90fps avg.

Crossfire using x3 cards: 115fps avg.

When I run the "benchmark1" scenario in A2:

Crossfire disabled (using a single card): 52fps avg.

Crossfire enabled using x2 cards: 51fps avg.

Crossfire using x3 cards: 51fps avg.

I realize this may have more to do with ATI/AMD drivers than anything else and that BIS may not be able to do anything about it. I've also posted this problem on the ATI forums, but I want to post here in case there is anything BIS can do to notify ATI/AMD of the issue so that it can be addressed in a future driver update. I'm not sure what else I can do to get the attention of the right people to have this looked at, but any suggestions are appreciated.

Has anyone else experienced similar issues with the 5800 series cards using crossfire? Am I the only one having this problem? Every other game and benchmark I have seems to scale properly. A2 is the only problem.

Thanks.

*Edit: I didn't post this in the existing performance issues thread because on a single card A2 performs fine. I don't think this is actually performance issue but something related to a driver bug with ATI or a bug/addition which needs to be made to A2 so that it works with 5800 series crossfire.

Edited January 14, 2010 by node_runner
addition

[kklownboy 43]

well it looks like your CPU limited....

As for the "Benchmark1" its not how i play the game.

[no use for a name 0]

It could also be because the game could be more CPU dependent

[node_runner 0]

That isn't the problem. I'm running a quad core i7 @3.8Ghz and the CPU is not stressed during the benchmark. I know how to benchmark my PC. The problem is definitely either driver related or a problem with A2, most likely driver related.

[kklownboy 43]

So.... its more that its Vsync related relative to your Dirt2 benching. I assume you have Win7,or Vista, which doesnt have the working option to turn off Vsync with A2. To see the benefits of your multiGPU, try doing a test case in the editor. As for the "Benchmark1" With my settings*, and QuadFire(4870X2s) i see 41fps. With one card's two GPUs(CF on and Cat_AI off), its 21fps, with just one GPU(no CFX,and Cat_AI off) i get 21fps... Sooo id almost say your right untill i use my test mission in the editor, where i can see where the low frames are or where more GPU(s) help, like being able to use 8XAA with all fourGPUs ect... Its kinda odd where ill see the same frame rate at the same spot/view between 4gpus and 2gpus, but then i bring the Scope up or Zoom and or increase filltering(AA) the 4gpus keep or increase the fps. Same goes for VD i can increase the VD with 4gpus and the fps doesnt change much..(5fps) untill i get too much VD.

We do have different cards, and there could be a issue with A2 and 5XXX series @CFX..? What are your settings/res ect?

*

1920/1440 90hz

CCC_AF_16X

VH_Tex

Defaut VRAM

VH_AF

Normal(4X)AA [High(8X)AA-35fps_Benchmark1]

VH_Terrain

VH_Object

VH_Shadow

Disabled_PP

VD_3800~

100% Fillrate

HDRprecission=32

sceneComplexity=1000000

[node_runner 0]

I appreciate your help, I really do. It's nice of you to take the time post suggestions. But my performance is driver-related, and isn't going to be affected that much by vsync. Maybe the difference of like 4-5 fps, but that is about it.

As far as testing it on another island instead of benchmark 1, of course results will vary from mission to mission. The performance is heavily related to how many AI there are, how many explosions are happening, where you are on the terrain, how zoomed you are, etc, etc. There is no doubt that my FPS will go up if I take out all of the AI and look around by myself on Utes on a blank map. That isn't really the point.

Also, I'm not complaining that the performance is bad. I'm not really looking for tips on how to get a better framerate by modifying the game settings. The game settings themselves don't really matter. Whether I have all of the settings maxed or not, I should still be getting better performance with 3 GPU's than with 1 GPU, but the performance is always the same, independently of what the video options are set to.

I have done more than just run benchmark1. This is after weeks of playing the game in various single player and multiplayer session, and there just really isn't any difference with crossfire on or off.

The 5000 series has had crossfire problems for some time now. This is a known issue with this series of cards. I'm sure that it is a driver-related problem. It has been cleared up with most games now, but it is still an issue with A2. I wasn't really looking for a solution on these forums, so much as a "Hey, if anyone at BIS is watching, please pester anybody at ATI if you can so they can fix this sooner than later" kind of thing and to see if there are others like me who are trying to run the 5800 series as a crossfire setup with A2. Maybe if there are others we can get enough people complaining about the issue to ATI/AMD and it will be a priority on their list of things to fix/improve in the next driver release.

[rekrul 7]

Arma2 doesn't support multi GPU support unfortunately. However, I'm not sure it's BIS fault, it might be a driver issue.

Edit: I have two 4890 or whatever, in xfire and there's no increase in performance. I can turn more gfx features on without loss in FPS, that's the only benefit...

Edited January 15, 2010 by Rekrul

[node_runner 0]

Arma2 doesn't support multi GPU support unfortunately. However, I'm not sure it's BIS fault, it might be a driver issue.

Edit: I have two 4890 or whatever, in xfire and there's no increase in performance. I can turn more gfx features on without loss in FPS, that's the only benefit...

That is what I was afraid of. Are you sure about this? I hope this isn't true, because if that is the case, then A2 is unlikely to ever take advantage of multiple GPU's and I'd have to just keep my fingers crossed for OA.

Can anyone else confirm this? Anyone with nVidia cards in SLI mode? If you run with the exact same graphical settings with and without SLI/Crossfire, is there a performance difference?

If it is just the drivers then I can at least have some hope and continue to pester ATI/AMD. If the game itself just doesn't support it then pestering BIS is going to be my only chance. Considering how much performance this game demands it would seem counter-productive not to support multiple GPU's.

[eble 3]

I think from the main tests done A2 is a monster to transfer info from your HD to your GPU etc.

I believe a few people have found putting the whole of arma2 to memory increased performance, unfortunately just adding additional GPU will not see the increases we do in other games.

A2 is a data hog and it needs to transfer vast amouts to run smooth.

Not sure A2 doesn't support multi GPU as I have an SLI card it is faster running both GPU than just one.

but then again mines a Nvidia card and not Ati

[node_runner 0]

I am running A2 on a dedicated intel x25M SSD. I know that currently, A2 has some fairly poor memory management and doesn't support x64 addressing, nor will it address over 2GB of memory. That is at the top of my wish list for OA. But I dunno. I have a pretty rockin' system here. Quad-core 3.8Ghz, dedicated SSD, etc. You would think that even though it is a data hog, A2 would still show performance increases for multiple GPU's.

Lastly, I don't think the texture data that A2 loads would keep it from utilizing the GPU's. It has to load all that data from the hard drive anyway, no matter how many GPU's I have. I'm not sure I understand why adding another GPU would suddenly cause A2 to start loading twice as much data from the HDD.

[pathetic_berserker 4]

If you want to compare to an SLI setup.

When I first got my hands on A2 I was running 2 8800GT's in SLi with a E6700 dual core (gave about 20-35fps) and found there was a about a 3-7 fps increase betwen running 1 card or 2. Then I upgraded to a Q9650 CPU and my FPS jumped another 10fps.

I now have a single GTX285 and run with evrything around high settings except for post process (off) and AA (med). Post Process effects are not to my taste except for screen shots, and the gain with AA is limited when runnng the game at my screens native res of 1600 x 1080, it already looks pretty good for gaming purposes, I get 40-60 fps depending on what and where.

Edited January 15, 2010 by Pathetic_Berserker

[[frl]myke 14]

@node_runner

could you please post your ingame quality settings? I also guess that your CPU is bottlenecking, besides the known Vsync bug.

A way to determine if it is your CPU bottlenecking: change settings on Viewdistance, Terrain Detail and Object Detail. You'll surely get pretty different FPS.

Now countercheck: leave the above mentoined settings to something rather low to medium and start playing around with other settings. Namely 3D Resolution is a nice test candidate. If you can up and down with settings without significant FPS changes, then it's a clear sign your CPU can't provide data fast enough.

Just to repeat:

Viewdistance, Terrain detail and Object detail are related to CPU performance.

All other settings are related to GPU performance.

Special case: Shadow detail. From very low to normal, shadows are calculated by CPU. High and very high are rendered by GPU. So with a decent graphic card you might get indeed better FPS with high shadow settings rather than with low.

By knowing this, you might restart your testing and judge yourself. And don't let yourself get fooled that CPU load doesn't go up to 100%, also a well known issue with ArmA 2.

[Yapab 10]

You are CPU limited (yes even with 3.8Ghz Quad), this can be clearly seen by the 1 fps drop when you run Crossfire.

Crossfire takes CPU calculations and your CPU performance drops, therefore your benchmark drops a little.

Unfortunately it looks like CPU is only running at say 40% or 60%, this may be true when you combine the work load of all cores and average the usage out, however ARMA2 has 1 thread which most other threads depend on.

What happens is that this one thread is maxing out one of your cores out (even at 3.8) and the other threads have to wait for that thread... Because Windows in most cases distributes thread load over cores to control temperature it looks as if none of your cores are maxed out, but one is and the others have to wait.

Not much you can do unfortunately until ARMA2 engine gets optimised for quads and that one thread gets split up. I would honestly run ARMA2 with only 1 GPU for the moment.

Yapa

[kklownboy 43]

Unfortunately it looks like CPU is only running at say 40% or 60%, this may be true when you combine the work load of all cores and average the usage out, however ARMA2 has 1 thread which most other threads depend on.

Yapa

When logging the Benchmark1 relative to logging "real"game play, The benchmark isnt fully taxing the CPU/cores and my GPUs, (CPU cores stay around 45%, GPUs 47%...) In real game play i get all four cores and GPUs near 9X%, and never less than 76% when gaming.When loading games, or looking at the MAP they drop down, but when ingame moving shooting, its full on usage.

I log with Everest. CPU is @4.1 I7_965

---------- Post added at 11:45 AM ---------- Previous post was at 11:25 AM ----------

Arma2 doesn't support multi GPU support unfortunately. However, I'm not sure it's BIS fault, it might be a driver issue.

wrong, it works with CFX and SLI, i can see it by changing my CF and CAT-AI settings.. But if you cant hit your Refresh anyways at what ever your Display rez is ... well you will still see the same fps do to droping below Vsync. I Double my fps from one GPU to two gpus on my 4870x2, as should your pair of 4890s, from one to two. But if your @19/12 60hz with JUST two 4890s your in low fps /Vsync.

Edit: I have two 4890 or whatever, in xfire and there's no increase in performance. I can turn more gfx features on without loss in FPS, that's the only benefit...

Well there ya go the CF profile for A2 has been in the driver for over a year now, and your getting some benifit. Been running CF or SLI since A1 has came out, and i always double my fps from one GPU to two.From 2gpu to 4gpu it wont double.(25%~)

Now if node_runner could post the rest of his specs and settings.

[Leon86 13]

I have 3 Radeon 5870 cards in a crossfire configuration.

:eek:

I'm sure ATI will eventually fix it with a new driver. Meanwhile I'd say just enjoy the excellent performance of just one 5870. I know it's more then twice as fast as my card (an 8800gtx).

[Scrub 0]

Like was said before - crush all options, bells and whistles, and then try checking FPS on crossfire or single. Data volume/CPU bottlenecks are then proven rather than assumed.

If you can turn on more post-processing and not lose FPS, it's the extra GPU providing the number crunching without volumes of data moving. If you can turn all off on Utes, and see a FPS change between one or two GPUs, it is the data/HDD limitation.

My thoughts anyway.

[Leon86 13]

Arma2 uses vsync by default so your fps will never go above 60 unless you force that off in the catalyst control centre. If you want to load your gpu and not your cpu you should put terrain and object detail an viewdistance on sortof medium settings (those load both cpu and gpu) then put your 3d resolution on 200% put postproccessing on highest and maybe even throw in some AA, then put anisotropic filtering on the max (my pc refuses that on 200% res, not enough mem I guess). those all are gpu intensive settings that dont really load the cpu. then run your benchmarks again and see if it made a difference.

[node_runner 0]

Now if node_runner could post the rest of his specs and settings.

I have a big post coming, with detailed/painstaking benchmarks to put these claims to the test. I've been working on them all day. I'll post them as soon as I'm done.

[kklownboy 43]

it is alot of time and work.. and not playing. look forward to your effort.

[Dead3yez 0]

@node_runner

What is your Framerate when there are 0 AI units in chernarus? :)

Then repeat with about 100 AI units far from you.

[rekrul 7]

wrong, it works with CFX and SLI, i can see it by changing my CF and CAT-AI settings.. But if you cant hit your Refresh anyways at what ever your Display rez is ... well you will still see the same fps do to droping below Vsync. I Double my fps from one GPU to two gpus on my 4870x2, as should your pair of 4890s, from one to two. But if your @19/12 60hz with JUST two 4890s your in low fps /Vsync.

Well there ya go the CF profile for A2 has been in the driver for over a year now, and your getting some benifit. Been running CF or SLI since A1 has came out, and i always double my fps from one GPU to two.From 2gpu to 4gpu it wont double.(25%~)

Now if node_runner could post the rest of his specs and settings.

I get xfire to work (and I'm running newest drivers), but there's no real increase in FPS on my system. The only thing is that I can increase is the quality without any FPS loss, which I cannot do on when not running xfire. Also, even at high settings my GPU usage is rarely over 40% on either GPU.

I'm currently running 1680x1050 with 100% 3D and 5k view distance. View distance is the only thing I really want, I don't care about the eye candy. I am intrigued by the post regarding the threads though as one CPU core is always higher than the others, but never at 100% so I'm not sure if there's a bottleneck there. (I'm currently running a Q9550 2.81GHz OCed to 3.4 GHz and 8GB RAM with 4GB RAMdisk.)

Edit: My point was that I believe ATI has to optimize their drivers for full usage, but might be dependent on BIS when it comes to multi-threading.

[Yapab 10]

I'm running the same CPU as you, Q9550 but at 3.82Ghz and GTX275 SLI.

I get the same result, CPU is at about 60% and both GPU's are only at about... 60% too, rarely do they get maxed out.

This is on Very High settings for everything. SLI does indeed work, however the game does not use quad core, and is limited by even a 3.8ghz cpu.

Yapa

[bangtail 0]

I think from the main tests done A2 is a monster to transfer info from your HD to your GPU etc.

I believe a few people have found putting the whole of arma2 to memory increased performance, unfortunately just adding additional GPU will not see the increases we do in other games.

A2 is a data hog and it needs to transfer vast amouts to run smooth.

Not sure A2 doesn't support multi GPU as I have an SLI card it is faster running both GPU than just one.

but then again mines a Nvidia card and not Ati

Yah, this is an ATI 5xxx problem as you most definitely see a significant difference with more than one Nvidia GPU.

I've played A2 on my 5970 a couple of times and was very unimpressed.

Hope they sort out the drivers for you OP :)

PS : There is a 10.1 beta driver - don't know if you've tried them?

http://downloads.guru3d.com/ATI-Catalyst-10.1-Beta--Windows-7-%7C-Vista-(32%7C64)-download-2455.html

Edited January 18, 2010 by BangTail

[node_runner 0]

Sorry that this is taking me so long. I had some problems where I started to get very inconsistent results and I had to re-do a ton of my benchmarks due to arma2.cfg issues. I'm not completely done yet, but I've finished enough of it to get a good idea of how much the CPU and GPU's are related and how much they affect performance at least on my machine. I'll update the Google Doc as I finish more of the benchmarks, but please bear with me because I have to run each of those benchmarks multiple times to make sure it is consistent and something odd isn't happening.

Setup

My hardware specs are:

-Intel Core i7 Quad Core CPU @2.6Ghz stock

-Water cooled cpu, capable of pretty good overclocks.

-x3 Radeon 5870's

-6GB Triple Channel DDR3 Corsiar memory at 1600mhz.

-Windows 7 x64

-Windows installed on a pair of Velocirator HDD's in raid 0

-Arma2 installed on dedicated Intel X25M 80GB SSD

-Single 27" monitor with a native res of 1920x1200

Software Info:

-Arma2 version 1.05 vanilla (no mods used for benchmarks).

-ATI Catalyst drivers version 9.12 (non-hotfix version).

-ATI Tray Tools 1.6 Beta (used to monitor FPS and force-disable vsync in Arma2, as well as monitor usage of each GPU)

-Built in Windows 7 performance monitor used to monitor CPU, memory, and SSD usage.

ArmA2 Configuration:

-Hyperthreading disabled in BIOS

-Vsync forced off using ATI Tray Tools mentioned above

-Catalyst AI set to "Advanced"

-Executable started with the following options: "-winxp -maxmem=2047 -cpuCount=4 -nosplash"

-Arma2.cfg modified to specify the localvram settings depending on how many GPU's I'm using for my test:

mem settings with 1 card: 536870912

mem settings with 2 cards: 1073741824

mem settings with 3 cards 1610612736

Two different sets of video settings used for benchmarks:

[Low Quality]

Draw Distance: 500

Texture Detail: Low

Video Memory: Default (uses size specified with localvram in arma2.cfg)

Anisotropic filtering: Disabled

Antialiasing: Disabled

Terrain detail: Very low

Objects Detail: very low

Shadow Detail: disabled

Posprocess effects: Disabled

Resolution: 960x600

[High Quality]

Draw Distance: 1600

Resolution: 1920x1200

Texture detail: very high

Anisotropic filtering: very high

Antialiasing: high

Terrain Detail: very high

Objects detail: very high

Shadow Detail: very high

Post Process: very high

Benchmarks used:

-Benchmark1 (tests zoomed in battle during the day)

-Benchmark2 (tests zoomed out battle at night)

-Arma II Mark (custom mission designed to perform multiple tests and give a final benchmark index at the end).

Testing Methodology

Because people seem to have different opinions on how well A2 takes advantage of multiple GPUs, how limited it is by CPU's, etc., I wanted to set up benchmarks to put those claims to the test. I wanted to run all three benchmarks on both High Quality settings and Low Quality settings, testing them on 1 GPU, 2 GPU's, and 3 GPU's. But since a lot of folks have been saying it is a CPU limitation, I also wanted to run all sets of benchmarks before and after overclocking my CPU. Some of the benchmarks have my system running the CPU stock at 2.6Ghz. Another set has them running overclocked at 3.3Ghz. If the CPU really is the bottleneck, then Crossfire *should* show more of a performance advantage with my CPU overclocked.

While I ran each test, I also had monitors up to watch the following statistics:

-CPU load of each core

-GPU load on each video card

-Memory usage

-%Idle time of the SSD that A2 is installed on

My thought was that I should be able to see the GPU's get utilized more once my CPU was overclocked. I ran each benchmark, and make some short notes on what I saw with the different monitoring tools.

Results

The raw FPS numbers of my benchmarks so far can be viewed on this Google Docs spreadsheet:

https://spreadsheets.google.com/ccc?key=0AnNLQZzfYb01dGFPZ0RmR3MtdmF1cGx4MUcxNW9WVGc&hl=en

[rekrul 7]

Just a correction. The manufacturers states that even though you xfire more cards, you don't get access to the 2nd or 3rd card's memory bank so the memory available should be consistent, no matter how many cards are in use. Also, I don't know why.