Is this causing the problems with the 8xxx series?

[Sekra]

As I've seen no reports of what the situation is with the problems with the nvidia 8xxx series I decided to do some poking around myself. And I think I've found something. I found this PerfSDK package from nvidia which I downloaded hoping it would give me some clue about what is causing the problems with the OpenGL driver since I am one of those poor buggers having to view "Allahs fist" -battle as a slideshow in realtime mode with my GF8800 GTS 640mb card. So I installed the PerfSDK package, did some configuring and fired up CM: SF (BF.com version 1.03). I told the GLExpert (which is a program in the PerfSDK to check problems of OpenGL applications) to post the errors in "console".

This is what I got when I fired up allahs fist and ran it in realtime for a minute or so:

</font><blockquote>code:</font><hr /><pre style="font-size:x-small; font-family: monospace;">OGLE: Category: 0x00000002, MessageID: 0x0081002

Falling back to software because RenderMode != GL_RENDER

(null)

</pre>

[The DesertFox]

Originally posted by Hotti:

As I've seen no reports of what the situation is with the problems with the nvidia 8xxx series I decided to do some poking around myself. And I think I've found something. I found this PerfSDK package from nvidia which I downloaded hoping it would give me some clue about what is causing the problems with the OpenGL driver since I am one of those poor buggers having to view "Allahs fist" -battle as a slideshow in realtime mode with my GF8800 GTS 640mb card. So I installed the PerfSDK package, did some configuring and fired up CM: SF (BF.com version 1.03). I told the GLExpert (which is a program in the PerfSDK to check problems of OpenGL applications) to post the errors in "console".

This is what I got when I fired up allahs fist and ran it in realtime for a minute or so:

</font><blockquote>code:</font><hr /><pre style="font-size:x-small; font-family: monospace;">OGLE: Category: 0x00000002, MessageID: 0x0081002

Falling back to software because RenderMode != GL_RENDER

(null)

</pre>

This message is posted many many many times. So it seems that some (programming?) error is causing the game to fallback to software rendering instead of hardware rendering and thus making your CPU do all the work your GPU should be doing!!! No wonder we get a slideshow there.. Our processors (like my Intel C2D E6400) might be able to process small maps in realtime too with some kind of performance but naturally when you increase the mapsize and having to calculate all the stuff that is happening under the "hood" AND on top of that you make your CPU process the gfx too..

How about that Steve?

Edit:

I found this in the GLExpert userguide:

This first part is about that Category 0x00000002:

</font><blockquote>quote:</font><hr /> NVAPI_OGLEXPERT_REPORT_SWFALLBACK

Category ID : 0x00000002

The SWFALLBACK category controls the reporting of situations where the driver has fallen back to software for at least part of the graphics pipeline. This has obvious performance impacts, and should always be avoided.

[Sekra]

The problem with this is that I've only found WHAT is happening not WHY it's happening. But I hope this can help them find out the reason too.

[The DesertFox]

Originally posted by Hotti:

The problem with this is that I've only found WHAT is happening not WHY it's happening. But I hope this can help them find out the reason too.

Hotti,

according to the latest posts from BF.C on this matter they didn´t even know WHAT is happening. So I think maybe now they hopefully have a starting point to sort this mess out.

cheers

[Sekra]

Also thinking the big picture here a bit further and putting some pieces of the puzzle together I think this problem has been noticed to be closely related to the mapsize so I believe there is something wrong with code in the landscape. The bigger the map the smaller the framerate naturally because it would seem that the ground is being software rendered instead of using our precious high-end gfx card to do that.

Edit:

The only thing that strikes me odd here is the fact that as Steve has stated the problem is more likely in the nvidia driver. Because if the error would be in the game code everyone running the game with nvidia cards would have problems and the slower your processor is the slower frame rates you would get.

I also browsed the open Nvidia Developers forums in which someone reported memory leak propblems with nvidia opengl drivers. Try here:

http://developer.nvidia.com/forums/index.php

[ September 04, 2007, 07:04 AM: Message edited by: Hotti ]

[peleprodigy]

Jeez, this sounds like a rookie thing to do. I'm not calling Hotti a rookie though...hopefully the game programmers can use this information and process a fix pronto!

funny-i had a feeling that it was software rendering , the graphics just lacked that pop that hardware gives

[rune]

I sent the thread to Charles to have a look see. I just sent it, so hopefully he can see if this means something to him.

Rune

[peleprodigy]

now that is pronto

[Dirtweasle]

If that is the problem you guys ought to send Hotti a collection of lithographs or something for un-covering it.

...on second thought; If it was there all this time waiting to be found so easily, why did it take a concerned user to have to find it?

Maybe you should send us all a gift!

[Battlefront.com]

Hotti,

Interesting! Well, it at least explains why the framerate is so dang'd low for some people, but it doesn't explain what's causing it. As you said in your first post, there is a problem since the nVidia slowdown isn't happening for all CM:SF players using the affected cards (for the sake of simplicity, I'll just call it the 8800). Meaning, two customers have 8800, one gets a slideshow and the other gets very good framerates looking at the same exact scenario. If this were a problem with CM then both people should be experiencing the same slow speed. Since that isn't the case, we have to look elsewhere.

The most likely cause is some sort of negative interaction of hardware and drivers. This interaction denies Combat Mission access to hardware rendering when it is asking for it, therefore CM has to accept software rendering. That doesn't tell us much, unfortunately, but it is a start.

BTW, this is Charles' specs. Hotti, you didn't mention what driver and forceware versions you are running. How do yours compare to this?

Windows Vista

GeForce 8800 GTS

Driver version: 7.15.11.5818 (April 12, 2007)

ForceWare version: 158.18

Thanks!

Steve

[ September 04, 2007, 09:24 AM: Message edited by: Battlefront.com ]

[Sekra]

Oh silly me! I will check the driver version right away.

[metalbrew]

Originally posted by Battlefront.com:

ForceWare version: 158.18

I believe 162.18 is the current non-beta driver. Why the older driver?

[Dennis50]

162.22 seems to be latest for 8800 and vista 32 on Nvidia's web site.

Regards,

Gunz

[Sekra]

Well it seems the PerfKIT installs its own "instrumented driver" which gives special access to the GPU. Anyhow here is what I was able to dig up:

Windows XP SP2

Driver Provider: NVIDIA

Driver Date: 2.8.2007

Driver Version: 6.14.11.6316

Digital Signer: Not digitally signed

Before installing I had forceware 162.18.

I tried installing the newest beta Forceware 163.44 but then I couldn't run the PerfKit.

[Sekra]

So I managed to use the kit to get some performance ratings using the Performance monitor supplied with windows XP. Here are two screenshots:

The graphs are explained in both pictures. Again I was running "Allahs Fist" and not even on full graphics settings.

Vert sync: Off

3d model quality: Better

3d texture quality: Better

Antialias/Multisample: Off

High priority access: Off

On the first picture the game is running but on the picture below that there it is waiting in the setup phase for me to press play. As you can see from the purple line the average FPS is around 8 when I'm in the setup phase. Of course there are a few spikes here and there but the thick green line indicates I'm using 100% cpu time almost all the time.

[Dirtweasle]

Wow. Very interesting.

[peleprodigy]

It is frustrating that a forum member used something from the nvidia developer site to find the problem. Is this diagnostic tool available for free?

[Battlefront.com]

Peleprodigy,

Keep in mind that what Hotti identified isn't very helpful to us in and of itself. It just confirms what we already know... some people with 8800 are getting horrible framerates. We're sure it is the drivers so that's still the angle we're looking into.

Steve

[Dirtweasle]

You knew the issue was the game reverting back to software rendering from hardware rendering?

[Douglas Rudd]

I was wondering what the current version of opengl is, so I looked at the opengl.dll in my system and it is this version:

5.1.2600.2180 (xpsp_sp2_rtm.040803-2158)

Is this the most recent, or is older? And if older, where would one get a more recent version?

[Battlefront.com]

Dirtweasle,

You knew the issue was the game reverting back to software rendering from hardware rendering?

Not specifically, no. We suspected that was the case simply because the framerate was so horrid and yet not influenced at all by any of the settings within CM or (for the problematic people) the settings on the card itself. In other words, the speed was consistent with being forced into software mode. The exact reason why is still a mystery to us, and that is what we need to be able to tell what the heck is going on. We're getting closer, though.

As I said, we're sure the problem lies in the drivers. The question is what the problem is. We are going to experiment over the next few days with reverting our troubled system to older drivers and then advancing our behaving system to newer drivers. The interesting thing is that the system that is using the older drivers was made by Alienware and shipped 2 weeks ago to us, yet Alienware chose to include drivers that are almost 6 months out of date. To us that says that Alienware knows something about the state of the current drivers and felt it was better to ship older than newer.

Steve

[Dirtweasle]

OK, well I downloaded FRAPs and fired up the Allah's Fist Scenario that it seems the others were using as a benchmarking tool and am quite surprised how bad the frame rates really are.

Min 0

Max 73

Avg 12.79

The max was when I was at the start screen, and from there it went downhill.

If I begin the sample after the start screen I get

Min 3

Max 31

Avg 6.574

My machine is;

Intel E6750 Core 2 Duo Conroe - (CPU affinity set to CPU 0)

Gigabyte P35C DS3R mobo

XP Pro, with SP2

MSI GeForce 8800 GTS Driver version 6.14.11.6218 and Forceware 162.18

I posted my full specs in the other thread, do you need anything else?

Happy to run whatever test would help.

[SlapHappy]

I think it's also important to mention here that 7XXX (and 6xxx) series card users can and have resorted to older drivers to improve frame rates.

However, there are only a handful of fairly recent drivers available for the 8XXX series, so those users can only go back so far.

[peleprodigy]

what are the best drivers to use for 6xx series cards? i may toss in my old 6800 ultra again...when i tried it last time i used current drivers. the results weren't that much different than my 8800, which now makes sense, because i used the same processor in both cases (since it was software rendering).

[Stirling]

Could there be a motherboard/video card conflict somewhere?