Asus A8R-MVP & Cat 6.2 Bug

jpinard · Feb 16, 2006, 05:18 AM

This problem has only been duplicated on Asus A8R-MPV boards thus far. If you have a Crossfire-based motherboard, or an Asus-A8R MVP, running the tests on page 6 would be helpful.

---------------------------------------------

I'll get straight to the point. I've finally isolated the problem with the CAT 6.2 drivers. It's with one file, the D3D driver aka ati3duag.dl_ in an install package, and named ati3duag.dll once it's properly installed in your /windows/system32/ directory.

Catalyst 5.13 reported normal test scores in 3dMark and cpu limited games, and CAT 6.2 showed a marked decrease. Understanding this, I decided to set out and find the exact file causing the issue. First I ran both sets of drivers without CCC installed and had the same results: the cat 6.2's were eating up 20-30% extra processing compared to the 5.13 drivers. Understanding it was not a CCC issues, nor a CAT AI issue etc, I decided to see if there was something else causing the problem. So I ran 69 tests in a row... that's right - SIXTY-NINE individual custom drivers installs to see what file(s) was causing the cpu performance hit. My methodology that isolated the D3D file was as follows:

* A base install of Win XP SP2, no sound card, mobo drivers, .net framework, system fully updated. I then took the individual setup directories and would do a fresh ATI driver install. I'd force the driver with CAT 5.13, and introduce CAT 6.2 files on install at a time. After testing to see if I could reproduce the cpu performance hit, I'd re-load the saved ghost image file and add one more. Through this painstaking and slow process of elimination I was able to complete 68 file replacements just fine with no cpu degradation. Only when the 6.2 version of the ati3duag.dl_ file was introduced - either on it's own or with multiple 6.2 files would it create the processor hit.
* Afraid someone might come in and say "it's your fault!", I decided to take it further. I made mixed installation packages of CAT 5.13/CAT 6.1, CAT 6.1/CAT 6.2, and of course CAT 5.13/CAT 6.2. I'd used all the driver files for one package, and just substituted the ati3duag.dl_ file. It's important to understand, that to properly register the dll, you can't just plop a different dll in the system directory, you need to install it via the setup driver install package - which is what I did (also note WinXP will give you a non-certified whql driver warning when mixing driver files upon running the installation - this is proper). Moving along, I found that using the Catalsyt 6.1 version of ati3duag.dl_ with the rest of the Catalyst 6.2 driver package yielded perfect cpu results. Thus re-inforcing the fact, something was changed/introduced with CAT 6.2 driver that will reduce performance in games that are cpu bound.
* I'm sure the next thing someone will ask, is, "why aren't more people reporting this?" I wondered this myself and set out to discover why. I ran SuperPi and Prime 95 benches to see if the processing performance hit happens all the time and discovered it does NOT. IT ONLY HAPPENS WHEH THE 3D DRIVER IS BEING ACTIVELY UTILIZED BY A 3D PROGRAM. This is why 3dMark 2003 and 3dMark 2005 shows the effects better than anything else (their cpu tests). 3dMark's 2006 test is much too slow and inaccurate to show this (after all, what's -20% of 1 or 0 fps)? The second reason is this may only be an issue with X1900 series cards or a combination of Crossfire motherboards with X1xx series cards. Plus factor in the fact no one bothers to looka t individual results, they only look at the final score (or bench in video-bottlenecked first person shooters) and you the issue goes missed by all, only noticed by the most meticulous individual.

If anyone has any doubts whatsoever, I am providing ALL the files, ALL the custom driver installation packages I made, ALL the test results, and most important, ALL the certified comparative results on ORB - which will show there's been absolutely no tampering with the evidence (see further down). I strongly urge people to "look beyond the final score"! Benches in Fear, Far Cry, and Quake are worhtless for more cpu intensive titles and it'd be a travesty for ATI or anyone to say sacrificing 20-30% of your processor for slightly better graphic performance is worth it. IN case you're unaware, there are plenty of games that are cpu bottlenecked not video bottlenecked. Titles such as Rome Total War, Silent Hunter III, Civ IV, Falcon 4 AF, Lock-On Modern Combat... in fact almost all miilitary sims will be cpu-bound for people with an X1900 series card. I'd like to add I'm not saying "everyone will have this issue". That's far from true. NForce or VIA mobos, & lower end video cards may not exhibit the issue. But anyone with a Crossfire motherboard (ex Asus A8R-MVP) and an ATI X1k series card should be concerned.

Some people have openly wondered in my other threads why I'm so tenacious about this issue? The answer - when you're sick most of the time, and finances are terribly tight... you need to get your money's worth. This hardware is much too expensive to have somethign like this happen. Of all the biiter irony in my search, I discovered you can't find comparative results on Futuremarks website (except my own) because they did not update their ORB database to accept X1900 tests until the day CAT 6.2 was released. Hence everyone saw an increase in their personal 3dMark scores on their computer and just called it good - never bothering to look at the individual numbers.

OK anough chatter (sorry) now for all the files, pictures, and proof. As I mentioned above, you can easily see the D3D dll for CAT 6.1 and 5.13 work just fine, the problem is only introduced with the CAT 6.2 D3D dll file.

Screenshot of graph taken out of this Excel chart: [color=red]http://www.jeffpinard.com/ATI%20cpu%20bug%20testing%20series%20-%203dMark%202003.xls[/color]

In the following screengrab, you can see the cpu chart from 3dMark 2005 results. Once again, the D3D dll for CAT 6.1 and 5.13 work just fine, the problem is only introduced with the CAT 6.2 D3D dll file. The Excel workup is here: [color=red]http://www.jeffpinard.com/ATI%20cpu%20bug%20testing%20series%20-%203dMark%202005.xls[/color]

Earlier I mentioned mixed driver install packages to test this all out. It's a big zip file since it contains several mixed drivers, but you can download it here to verify I did it correctly (I did). But if you want to test this out too, feel free to grab them (87Meg) [color=red]http://www.jeffpinard.com/Mixed%20Driver%20Package.zip[/color]

In case anyone has concerns looking at my Excel files form 3dMark, you can see a very nice/easy comparison through Futuremark's ORB. It's an excellent chart, and if you scroll down a bit you can see overall 3dMark scores are up, but at a great expense to the cpu. Just look at the blue shaded bars and these are the tests utilizing the CAT 6.2 D3D dll. One blue bar is normal CAT 6.2, and the other blue bar is CAT 5.13 with only the CAT 6.2 D3D file in it. The other tests show you can use every CAT 6.2 file, only substituting an earlier version of the D3D dll and cpu performance is fine (with CAT 6.1 and CAT 5.13). Once again this re-infores an issue with just rhe D3D dll of CAT 6.2

Futuremark 3dM03 comparative view: [color=red]http://tinyurl.co.uk/vx89[/color]
Futuremark 3dM05 comparative view: [color=red]http://tinyurl.co.uk/5y7v[/color]

I mentioned above I ran SuperPi and Prime95 to see if the performance hit was "all the time", or "just in D3D games". You can download my entire SuperPi folder with all the test results and proof pictures of the successful driver mixing and testing. You can correlate times and test results and see they're 100% accurate. An example of the proof pics (sorry they're blurry due to compressing the screen dimensins, but the full uncompressed pics are in the zip file:

[color=red]http://www.jeffpinard.com/SUPER_PI%20and%20Prime95.zip[/color]

ex:

Once again these tests help illustrate that the problem is isolated to running a D3D app tha's cpu bound.

If ATI reads this, you can pm me and I will be more than happy to carry out any additional testing or configuring you'd like. My goal here is to try and find a solution... I've done the best I can without writing my own drivers. and yes, it's the honest to God truth I've spent 50+ hours on this.

knuth · Feb 16, 2006, 06:12 AM

Whether you're right or wrong, you've done some excellent work.

jpinard · Feb 16, 2006, 07:09 AM

I finally got a response from another MVP user and here's what he had to say,

Quote:

"Oh, I see what ya mean. I bet you have a single core processor.

If so, then what you have is not a mainboard issue. Its actually a well documented fact of ATI's video driver. They released versions with dual core enhancements. This is great if you have a dual core/processor or hyperthreaded system. Unfortunately they only have one version and the extra threads they introduced in the optimizations tend to clog up the scheduler and hurt single core performance.

In any application that does not utilize Direct3d (since OpenGL is still unoptimized) you will not be affected.

Unfortunately, my current system has only used the multithreaded drivers, so I have no way to compare. My x1600 won't run on the older ones. My previous configuration had ran on both, but it was hyperthreaded and unaffected. And I wasn't really that worried anyhow.

Unless you actually fell the slowdown in Rome or whatever then don't sweat it. 3dMark isn't the final say on your CPU performance."

If there's any truth to this dual core optimization at the expense of single cores, I beseech ATI to release seperate drivers for duel core vs. single core users. In fact what he wrote is exactly in-line with my hypothesis above. The ony problem here is cpu intensive games ARE affected. If games weren't affected I could care less about my 3dMark scores.

I forgot to post my basic system specs:

Asus A8R-MVP Bios rev 402
AMD FX-53 (single core)
2 Gig 3200LL Patriot Memory (1x1 in dual channel mode)
ATI X1900XTX
Mitsubishi Diamond Pro 2060u 22" CRT
Creative Labs X-FI Xtrememusic sound card
Seasonic S12 500W PSU
Win XP SP2 fully updated
Directx 9.0C Feb 2006 build

beezee · Feb 16, 2006, 10:44 AM

You are one tenacious S-O-B

I applaud your efforts, excellent work!!!

BIG_FREAK · Feb 16, 2006, 01:50 PM

Man That is some sick stuff you have done some sweet work thanks for the great read

Ib Al-Musaheb · Feb 16, 2006, 02:59 PM

Definetely big respects from me ...

H3X4D3C1M4L · Feb 16, 2006, 03:15 PM

Hails

Matth · Feb 16, 2006, 04:07 PM

The "it's a dual / single core issue" argument also seems to be a moot point, as 5.13 and 6.1 are both dual core optimized drivers, as well as 6.2.

Maybe in 6.2, they got carried away with throwing what would be "spare" dual core CPU power at any slight improvement, to the detriment of overall CPU usage, in which case, never mind two versions of the driver, it should be made to auto-adapt.

merlinxi · Feb 16, 2006, 04:26 PM

Finally! AMAZING WORK.

MindlessOath · Feb 16, 2006, 06:33 PM

i still leave this to an optimization (or whatever a bug) that only occors on chipset/motherboards of your flavor, ie either your chipset or the fact that its crossfire.

why? how many people did this test with the same motherboard as yourself?

its the only thing that makes sense. ati created something to make aa differnt or what not and only if you had that board/chipset. mabe its a bug too.

i really really doubt this is a single core/dual core issue, but it can play a roll in this, but i doubt its a huge influnance here.

~~Sanafabich~~ · Feb 16, 2006, 06:36 PM

nice!

jpinard · Feb 16, 2006, 06:37 PM

Thank you for the kind words.

If ATI in their lab, sets up a Crossfire mobo (preferably Asus A8R-MVP), a single core AMD64, and an X1900 - they should be to easily replicate these results.

tech.Me · Feb 17, 2006, 12:53 AM

It's reproducable on my Athlon 2400+.

In RBR, I can get 75fps, no sweat. That's with vsync on, BTW. Omega 4.12s.
With 5.13s and higher, that's a 45-ish fps. Now that's a problem.

With that in mind, I just need to dig out the 4.12s from sites and replace the damn file - it should work.

jpinard · Feb 17, 2006, 03:44 AM

In the interest of being as exceedingly thorough as possible, leaving nothing to chance - I decided to disable (via BIOS) all my USB devices, I disabled my onboard LAN, and disabled my firewire device. I forgot to mention eariler my default configuration has on-board sound disabled, serial ports disabled, the game port disabled, and the midi port disabled.

With the above settings, I re-ran the tests and got the same cpu hit with the CAT 6.2 D3D driver in use. Of course, with no LAN connection I could not submit my results to ORB - hehe. So the results are just in Excel if anyone should have need of that. The reason for disabling all those devices was the theory someone presented of "maybe there's a possability of competing USB or LAN instructions impeding the cpu with CAT 6.2." At any rate, USB/Network/Firewire instructions/devices can be scratched off the list of possible conflicts.

I do have one more thing to test that I hand't thought of eariler... Tinkering around I was trying to see if I could get the constant streaming .net framework messages piling up in the event log. I didn't get it until I had BOTH .net 1.1 AND 2.0 installed. But I did reproduce it (with CCC 6.12 installed - did not check this possability with a different driver). This led me to wonder if by some bizarre circumstance .net framework isn't playing nice with the CAT 6.2 in D3D mode on some systems. I highly doubt it, but I'll try it.

jpinard · Feb 17, 2006, 05:00 AM

YES! We've pegged the bug! insane_words has duplicated my test results. This is a very important finding as it eliminates the possability of it being a dual/single core processor issue.

Since X1900 users with NForce motherboards have not duplicated these results, and Crossfire based motheboard with X1900 cards HAVE... it means the following:

Crossfire-based motherboards (maybe only Asus A8R-MVP oqwners, since I've yet to see data from any DFI owners) are affected. Potentially it is ONLY Crossfire motherboard owners. So far this has only been duplicated by X1900 series owners. I've not really seen anyone with an X1800 or lower try the same tests yet. So I have no idea if it's just the X1900 or if it's everyone with an X1XX series card.

insane_words also confirmed the overall 3dMark score is up, but at great expense to the processor. Once again, this is misleading to owners of ATI hardware - and is probably the reason more Crossfire-based motherboard owners haven't reported it yet.

It should be a simple matter now for ATI to replicate the problem. Fire up an X1900 video card on an Asus A8R-MVP. Bench with 3dMark 2003 using CAT 5.13 and then CAT 6.2 . Do not use 3dMark 2006 as the test. Once again -20% of 0 fps or 1 fps zero fps is negligible and stastically insignificant. 3dMark 2005 and 3dMark 2003 are much better indicators since a cpu-bound frame rate can be held in territory that doesn't fall in the standard deviation. Id' also like to reiterate, the performance hit is only seen when the D3D mode is ACTIVE. Benching without running a 3d app will not show the problem.

If anyone has a DFI Crossfire board and an X1900 please due these tests so we can see if it's all our boards, or just the Asus A8R MVP. As new Crossfire motherboards hit the market this month... we should see if they are effected as well.

If anyone has an Asus A8R-MVP and an X1800 or lower, please test so we can see if it's X1900 users only, or everyone with the new breed of ATI PCIe cards.

Thanks to everyone that has read, offered input/theories, and helped with this issue.

Here is insane_words post:

Quote:

Ok...the results are in.

First off, prior to each benchmark I uninstalled any ATI display drivers on the system, rebooted into safe mode and ran DriverCleaner, rebooted again to install the driver I would be testing, and then rebooted a final time to run the benchmark.

All benchmarks are 3DMark03. Full tests were run (freeware version) but I'm only posting the results from the game tests and CPU tests.

6.2's with 5.13 D3D
GT1: 373
GT2: 143
GT3: 103
GT4: 110
CPU1: 106
CPU2: 19

Total Score: 17191
CPU Score: 1017

6.2's with 6.1 D3D
GT1: 374
GT2: 143
GT3: 104
GT4: 110
CPU1: 106
CPU2: 19

Total Score: 17241
CPU Score: 1016

5.13's with 6.2 D3D
GT1: 364
GT2: 144
GT3: 103
GT4: 111
CPU1: 101
CPU2: 15

Total Score: 17218
CPU Score: 896

Standard 6.2 Install
GT1: 375
GT2: 144
GT3: 104
GT4: 111
CPU1: 101
CPU2: 15

Total Score: 17272
CPU Score: 898

So, the 6.2's do kill your CPU Score in 03. But they bump your overall score. At least, that's what they're doing for me.

Here's a link to my 6.2 Standard Install results: [color=#0000ff]http://service.futuremark.com/compare?2k3=4626432[/color]. Wish I could post them all for you, jpinard, but like I said, I'm using the freeware version.

Also, I did find out why I was getting that redraw lag in IE and explorer windows. Apparently, it's a X2 thing? If I set the affinity for IE and explorer for a single core, the lag is completely gone. Go figure. So it wasn't a CAT issue after all.

Thanks again for the comments and suggestions. Lemme know if you need any other info or benchmarks from me jpinard.

Vital System Specs:
Fully Patched Windows XP Pro w/ SP2
80 gig Raptor 10K RPM SATA HDD
1gig Corsair XMS PC3500 @ DDR400 2 2 2 5 Dual Channel
AMD Athlon 64 X2 4400+
HIS Radeon X1900XT
Asus A8R-MVP w/ latest BIOS and drivers
Audigy 2
Antec True Power II 550W PSU

[color=red]http://www.rage3d.com/board/showthread.php?p=1334180170#post1334180170[/color]

f0rbez · Feb 17, 2006, 05:57 AM

I'm sort of confused. What processors is this effecting?

Also, theres a tweak in ATT that allows you to disable multithreading. Have you tested that?

The_Neon_Cowboy · Feb 17, 2006, 06:13 AM

......... just wait till tomarrow night i'm off to bed

jpinard · Feb 17, 2006, 06:13 AM

Quote:

Originally Posted by f0rbez

I'm sort of confused. What processors is this effecting?

Also, theres a tweak in ATT that allows you to disable multithreading. Have you tested that?

It apparently affects any kind of processor on a Crossfire motherboard. My last post (the one right above yours) supercedes all of my previous theoritical speculation.

Neon, I don't want to be rude, but if you took the time to read all the posts in order in this thread, you'd see clear as day the problem and the proof is there. Your counter-speculation is not grounded in reality... and just because you don't like something - doesn't mean it's not valid.

Now, if you have a Crossfire motherboard and an X1900 video card, you can positively contribute to this thread by publishing benchmark results and testing in cpu limited games. Until you do that however, I'd appreciate you not throw this thread off course with pure opinion not grounded in fact or backed up with proof.

I'd also like to reiterate - which I've already written up above - that the processor performance drain does not happen unless the d3d dll is fully utilized. Therefore benches like Sisoft Sandra would be useless. I don't want to get in a flame war. So lets just leave it to the data.

daedal · Feb 17, 2006, 01:48 PM

Well I applaud your efforts. Respect for all the work you're doing.

MindlessOath · Feb 17, 2006, 02:54 PM

Quote:

Id' also like to reiterate, the performance hit is only seen when the D3D mode is ACTIVE. Benching without running a 3d app will not show the problem.

so try this while running the d3d test with ati tool and running cpu tasks.

also - i told you so.

jpinard · Feb 17, 2006, 08:02 PM

Quote:

Originally Posted by MindlessOath

so try this while running the d3d test with ati tool and running cpu tasks.

also - i told you so.

Mindless Oath - please elaborate on which tests you want me to try. I'll be happy to do them

Quote:

also - i told you so.

<---- what's that supposed to mean?

Zardon · Feb 17, 2006, 08:52 PM

I have let ATI know about this, but I havent had an answer. Just so you know im not ignoring the thread .....

jpinard · Feb 17, 2006, 09:58 PM

Quote:

Originally Posted by Zardon

I have let ATI know about this, but I havent had an answer. Just so you know im not ignoring the thread .....

I didn't think that at all. Thank you Zardon.

BWX · Feb 17, 2006, 10:12 PM

So this only affects x1xxx cards and only affects those on Crossfire mobos?

No one else is affected for sure?

Chaos · Feb 17, 2006, 10:18 PM

hummm too bad I dont have an X1900 or I would have to fire up my new crossfire board.

jpinard · Feb 17, 2006, 10:33 PM

Quote:

Originally Posted by BWX

So this only affects x1xxx cards and only affects those on Crossfire mobos?

No one else is affected for sure?

I'm sorry I don't know the answer to that. In all the posts, I've only seen one other set of tests replicated (with an Nforce4 board) and they did not have this same result. So in poor scientific fashion I assumed it was only Crossfire boards with x1xx cards.

I'd like to apologize to everyone if any of my posts were negative to individuals. Sorry Neon Cowboy if you were offended by my response above. I'm trying very hard to keep a level head - I've been 100% open and honest. I simply do not understand why anyone would be angered over this. It's just an issue that needed some attention/fixing - that's all.

BWX · Feb 17, 2006, 10:46 PM

I think my system is pretty CPU bound in general.. I might do a few tests just see what happens.

Jerunk · Feb 17, 2006, 11:31 PM

I havn't read the whole thread, but stuff like this is definitely appreciated.

Exceededgoku · Feb 17, 2006, 11:46 PM

Thanks for all this information, it is a real eye opener. ATi have really disappointed me lately...

jpinard · Feb 18, 2006, 03:37 AM

I mentioned earlier I did some research into user complaints about .net framework spamming (unrelated to ATI drivers) their system and their event logs. I decided it was worth trying a 6.2 driver-only install with no .net framework. I'm not sure how valid this is... the driver-only package installed the specific X1900 graphic device driver just fine, but could not install multimedia drivers (WDM NDF). But, I ran the tests anyways and the results for CAT 6.2 were the same. Higher graphic scores for everything, but much lower cpu scores. A comparison of "CAT 5.13 with CCC & .net framework" vs. "CAT 6.2 with no .net framework" :[color=red] [color=red]http://tinyurl.com/a59pe[/color] [/color]No sound card installed, just to make sure there was no conflict.

In case there was any doubt - I'm not complaining about the graphic performance at all. I think the overall video performance gains in CAT 6.2 are wonderful.

However, it's a different story when the processor is the bottleneck. And in that case, any gains made - and a lot more - is lost.

I'm going to throw this thought out there... anyone think there's a possability the MVP switch card utilizing the second PCIe slot could be causing issues with CAT 6.2? I know it's probabaly a ridicuous thought...but I thought I'd ask. So far the only people able to replicate this are those with this type of system and an X1900. I'd be curious to see if anyone with a Crossfire board using 2 cards in Crossfire mode - would show the same cpu hit.

Feb 16, 2006, 05:18 AM	#1
jpinard DriverHeaven Addict Join Date: Jul 2003 Posts: 278 Rep Power: 0	Asus A8R-MVP & Cat 6.2 Bug This problem has only been duplicated on Asus A8R-MPV boards thus far. If you have a Crossfire-based motherboard, or an Asus-A8R MVP, running the tests on page 6 would be helpful. --------------------------------------------- I'll get straight to the point. I've finally isolated the problem with the CAT 6.2 drivers. It's with one file, the D3D driver aka ati3duag.dl_ in an install package, and named ati3duag.dll once it's properly installed in your /windows/system32/ directory. Catalyst 5.13 reported normal test scores in 3dMark and cpu limited games, and CAT 6.2 showed a marked decrease. Understanding this, I decided to set out and find the exact file causing the issue. First I ran both sets of drivers without CCC installed and had the same results: the cat 6.2's were eating up 20-30% extra processing compared to the 5.13 drivers. Understanding it was not a CCC issues, nor a CAT AI issue etc, I decided to see if there was something else causing the problem. So I ran 69 tests in a row... that's right - SIXTY-NINE individual custom drivers installs to see what file(s) was causing the cpu performance hit. My methodology that isolated the D3D file was as follows: * A base install of Win XP SP2, no sound card, mobo drivers, .net framework, system fully updated. I then took the individual setup directories and would do a fresh ATI driver install. I'd force the driver with CAT 5.13, and introduce CAT 6.2 files on install at a time. After testing to see if I could reproduce the cpu performance hit, I'd re-load the saved ghost image file and add one more. Through this painstaking and slow process of elimination I was able to complete 68 file replacements just fine with no cpu degradation. Only when the 6.2 version of the ati3duag.dl_ file was introduced - either on it's own or with multiple 6.2 files would it create the processor hit. * Afraid someone might come in and say "it's your fault!", I decided to take it further. I made mixed installation packages of CAT 5.13/CAT 6.1, CAT 6.1/CAT 6.2, and of course CAT 5.13/CAT 6.2. I'd used all the driver files for one package, and just substituted the ati3duag.dl_ file. It's important to understand, that to properly register the dll, you can't just plop a different dll in the system directory, you need to install it via the setup driver install package - which is what I did (also note WinXP will give you a non-certified whql driver warning when mixing driver files upon running the installation - this is proper). Moving along, I found that using the Catalsyt 6.1 version of ati3duag.dl_ with the rest of the Catalyst 6.2 driver package yielded perfect cpu results. Thus re-inforcing the fact, something was changed/introduced with CAT 6.2 driver that will reduce performance in games that are cpu bound. * I'm sure the next thing someone will ask, is, "why aren't more people reporting this?" I wondered this myself and set out to discover why. I ran SuperPi and Prime 95 benches to see if the processing performance hit happens all the time and discovered it does NOT. IT ONLY HAPPENS WHEH THE 3D DRIVER IS BEING ACTIVELY UTILIZED BY A 3D PROGRAM. This is why 3dMark 2003 and 3dMark 2005 shows the effects better than anything else (their cpu tests). 3dMark's 2006 test is much too slow and inaccurate to show this (after all, what's -20% of 1 or 0 fps)? The second reason is this may only be an issue with X1900 series cards or a combination of Crossfire motherboards with X1xx series cards. Plus factor in the fact no one bothers to looka t individual results, they only look at the final score (or bench in video-bottlenecked first person shooters) and you the issue goes missed by all, only noticed by the most meticulous individual. If anyone has any doubts whatsoever, I am providing ALL the files, ALL the custom driver installation packages I made, ALL the test results, and most important, ALL the certified comparative results on ORB - which will show there's been absolutely no tampering with the evidence (see further down). I strongly urge people to "look beyond the final score"! Benches in Fear, Far Cry, and Quake are worhtless for more cpu intensive titles and it'd be a travesty for ATI or anyone to say sacrificing 20-30% of your processor for slightly better graphic performance is worth it. IN case you're unaware, there are plenty of games that are cpu bottlenecked not video bottlenecked. Titles such as Rome Total War, Silent Hunter III, Civ IV, Falcon 4 AF, Lock-On Modern Combat... in fact almost all miilitary sims will be cpu-bound for people with an X1900 series card. I'd like to add I'm not saying "everyone will have this issue". That's far from true. NForce or VIA mobos, & lower end video cards may not exhibit the issue. But anyone with a Crossfire motherboard (ex Asus A8R-MVP) and an ATI X1k series card should be concerned. Some people have openly wondered in my other threads why I'm so tenacious about this issue? The answer - when you're sick most of the time, and finances are terribly tight... you need to get your money's worth. This hardware is much too expensive to have somethign like this happen. Of all the biiter irony in my search, I discovered you can't find comparative results on Futuremarks website (except my own) because they did not update their ORB database to accept X1900 tests until the day CAT 6.2 was released. Hence everyone saw an increase in their personal 3dMark scores on their computer and just called it good - never bothering to look at the individual numbers. OK anough chatter (sorry) now for all the files, pictures, and proof. As I mentioned above, you can easily see the D3D dll for CAT 6.1 and 5.13 work just fine, the problem is only introduced with the CAT 6.2 D3D dll file. Screenshot of graph taken out of this Excel chart: [color=red]http://www.jeffpinard.com/ATI%20cpu%20bug%20testing%20series%20-%203dMark%202003.xls[/color] In the following screengrab, you can see the cpu chart from 3dMark 2005 results. Once again, the D3D dll for CAT 6.1 and 5.13 work just fine, the problem is only introduced with the CAT 6.2 D3D dll file. The Excel workup is here: [color=red]http://www.jeffpinard.com/ATI%20cpu%20bug%20testing%20series%20-%203dMark%202005.xls[/color] Earlier I mentioned mixed driver install packages to test this all out. It's a big zip file since it contains several mixed drivers, but you can download it here to verify I did it correctly (I did). But if you want to test this out too, feel free to grab them (87Meg) [color=red]http://www.jeffpinard.com/Mixed%20Driver%20Package.zip[/color] In case anyone has concerns looking at my Excel files form 3dMark, you can see a very nice/easy comparison through Futuremark's ORB. It's an excellent chart, and if you scroll down a bit you can see overall 3dMark scores are up, but at a great expense to the cpu. Just look at the blue shaded bars and these are the tests utilizing the CAT 6.2 D3D dll. One blue bar is normal CAT 6.2, and the other blue bar is CAT 5.13 with only the CAT 6.2 D3D file in it. The other tests show you can use every CAT 6.2 file, only substituting an earlier version of the D3D dll and cpu performance is fine (with CAT 6.1 and CAT 5.13). Once again this re-infores an issue with just rhe D3D dll of CAT 6.2 Futuremark 3dM03 comparative view: [color=red]http://tinyurl.co.uk/vx89[/color] Futuremark 3dM05 comparative view: [color=red]http://tinyurl.co.uk/5y7v[/color] I mentioned above I ran SuperPi and Prime95 to see if the performance hit was "all the time", or "just in D3D games". You can download my entire SuperPi folder with all the test results and proof pictures of the successful driver mixing and testing. You can correlate times and test results and see they're 100% accurate. An example of the proof pics (sorry they're blurry due to compressing the screen dimensins, but the full uncompressed pics are in the zip file: [color=red]http://www.jeffpinard.com/SUPER_PI%20and%20Prime95.zip[/color] ex: Once again these tests help illustrate that the problem is isolated to running a D3D app tha's cpu bound. If ATI reads this, you can pm me and I will be more than happy to carry out any additional testing or configuring you'd like. My goal here is to try and find a solution... I've done the best I can without writing my own drivers. and yes, it's the honest to God truth I've spent 50+ hours on this. Last edited by jpinard; Feb 24, 2006 at 08:34 PM. Reason: Better title explanation...

Feb 16, 2006, 01:50 PM	#5
BIG_FREAK DriverHeaven Junior Member Join Date: Sep 2004 Location: Fredericton.nb.ca Posts: 73 Rep Power: 0	Man That is some sick stuff you have done some sweet work thanks for the great read __________________ [COLOR=black][/COLOR]

Feb 16, 2006, 02:59 PM	#6
Ib Al-Musaheb DriverHeaven Lover Join Date: Dec 2005 Posts: 190 Rep Power: 0 System Specs	Definetely big respects from me ... __________________

Feb 16, 2006, 04:26 PM	#9
merlinxi The Paranoid Cook Join Date: Feb 2004 Location: New Jersey Posts: 655 Rep Power: 0	Finally! AMAZING WORK. __________________ 700 Watt Toughpower Cable Management Power supply -- Asus P55 LE Motherboard -- Intel Core i7 860 2.8GHZ Quad Core -- 6Gb Dominator Ram DD3 8 Latency Timings -- Windows 7 64bit -- Gigabyte GTX470 Superclocked Video Card -- 1.5 TB Hard drive, 1TB hard drive, 250GB Hard drive, and a partridge in a pear tree.

Feb 16, 2006, 06:33 PM	#10
MindlessOath DriverHeaven Senior Member Join Date: Jul 2003 Posts: 1,121 Rep Power: 0	i still leave this to an optimization (or whatever a bug) that only occors on chipset/motherboards of your flavor, ie either your chipset or the fact that its crossfire. why? how many people did this test with the same motherboard as yourself? its the only thing that makes sense. ati created something to make aa differnt or what not and only if you had that board/chipset. mabe its a bug too. i really really doubt this is a single core/dual core issue, but it can play a roll in this, but i doubt its a huge influnance here. __________________ ATI Radeon X800XT PE <---- LOOK UPDATED!! AMD Athlon 2Ghz, nforce2, 512mb ddr3500 LINKS U SHOULD KNOW!!!! DriverCleaner - use every time u install new drivers! ATI driver feedback - to report driver problems Tacticalcenter - my Gaming website and forums ATI 3rd party Optimized Driver Compairison

Feb 16, 2006, 06:12 AM	#2
knuth DriverHeaven Junior Member Join Date: Oct 2004 Location: Johannesburg, South Africa Posts: 73 Rep Power: 0	Whether you're right or wrong, you've done some excellent work.

Feb 16, 2006, 10:44 AM	#4
beezee DriverHeaven Junior Member Join Date: Dec 2004 Posts: 51 Rep Power: 0	You are one tenacious S-O-B I applaud your efforts, excellent work!!!

Feb 16, 2006, 03:15 PM	#7
H3X4D3C1M4L DriverHeaven Extreme Member Join Date: May 2005 Posts: 6,794 Rep Power: 0	Hails

Feb 16, 2006, 04:07 PM	#8
Matth Flash Banner Hater Join Date: Jun 2002 Location: UK Posts: 3,426 Rep Power: 93 System Specs	The "it's a dual / single core issue" argument also seems to be a moot point, as 5.13 and 6.1 are both dual core optimized drivers, as well as 6.2. Maybe in 6.2, they got carried away with throwing what would be "spare" dual core CPU power at any slight improvement, to the detriment of overall CPU usage, in which case, never mind two versions of the driver, it should be made to auto-adapt.

Feb 16, 2006, 06:36 PM	#11
~~Sanafabich~~ BANNED Join Date: Mar 2004 Posts: 511 Rep Power: 0	nice!

Feb 16, 2006, 06:37 PM	Thread Starter #12
jpinard DriverHeaven Addict Join Date: Jul 2003 Posts: 278 Rep Power: 0	Thank you for the kind words. If ATI in their lab, sets up a Crossfire mobo (preferably Asus A8R-MVP), a single core AMD64, and an X1900 - they should be to easily replicate these results.

Feb 17, 2006, 12:53 AM	#13
tech.Me DriverHeaven Lover Join Date: Apr 2005 Posts: 139 Rep Power: 0	It's reproducable on my Athlon 2400+. In RBR, I can get 75fps, no sweat. That's with vsync on, BTW. Omega 4.12s. With 5.13s and higher, that's a 45-ish fps. Now that's a problem. With that in mind, I just need to dig out the 4.12s from sites and replace the damn file - it should work. __________________ AMD Athlon XP 2400+ \| Windows XP Pro \| ATI 9600XT 128MB \| Samsung 120GB 5400RPM \| Saitek Cyborg Evo

Feb 17, 2006, 03:44 AM	Thread Starter #14
jpinard DriverHeaven Addict Join Date: Jul 2003 Posts: 278 Rep Power: 0	In the interest of being as exceedingly thorough as possible, leaving nothing to chance - I decided to disable (via BIOS) all my USB devices, I disabled my onboard LAN, and disabled my firewire device. I forgot to mention eariler my default configuration has on-board sound disabled, serial ports disabled, the game port disabled, and the midi port disabled. With the above settings, I re-ran the tests and got the same cpu hit with the CAT 6.2 D3D driver in use. Of course, with no LAN connection I could not submit my results to ORB - hehe. So the results are just in Excel if anyone should have need of that. The reason for disabling all those devices was the theory someone presented of "maybe there's a possability of competing USB or LAN instructions impeding the cpu with CAT 6.2." At any rate, USB/Network/Firewire instructions/devices can be scratched off the list of possible conflicts. I do have one more thing to test that I hand't thought of eariler... Tinkering around I was trying to see if I could get the constant streaming .net framework messages piling up in the event log. I didn't get it until I had BOTH .net 1.1 AND 2.0 installed. But I did reproduce it (with CCC 6.12 installed - did not check this possability with a different driver). This led me to wonder if by some bizarre circumstance .net framework isn't playing nice with the CAT 6.2 in D3D mode on some systems. I highly doubt it, but I'll try it.

Feb 17, 2006, 05:57 AM	#16
f0rbez DriverHeaven Lover Join Date: May 2004 Location: s jersey, us Posts: 176 Rep Power: 0	I'm sort of confused. What processors is this effecting? Also, theres a tweak in ATT that allows you to disable multithreading. Have you tested that?

Feb 17, 2006, 06:13 AM	#17
The_Neon_Cowboy HardwareHeaven Extreme Member Join Date: Dec 2002 Location: U.S.A. Posts: 16,009 Rep Power: 92 System Specs	......... just wait till tomarrow night i'm off to bed __________________ Last edited by The_Neon_Cowboy; Mar 19, 2006 at 08:33 AM.

Feb 17, 2006, 01:48 PM	#19
daedal Uber Coffee Drinker Join Date: Mar 2005 Location: Gatineau, Quebec Posts: 2,250 Rep Power: 0	Well I applaud your efforts. Respect for all the work you're doing. __________________

Feb 17, 2006, 08:52 PM	#22
Zardon DriverHeaven Founder Join Date: May 2002 Posts: 32,480 Rep Power: 179	I have let ATI know about this, but I havent had an answer. Just so you know im not ignoring the thread .....

Feb 17, 2006, 10:12 PM	#24
BWX unplugged Join Date: Nov 2002 Location: USNY Posts: 19,669 Rep Power: 110 System Specs	So this only affects x1xxx cards and only affects those on Crossfire mobos? No one else is affected for sure? __________________

Feb 17, 2006, 10:18 PM	#25
Chaos Number Nine Join Date: May 2002 Location: Nova Scotia Posts: 5,268 Rep Power: 167 System Specs	hummm too bad I dont have an X1900 or I would have to fire up my new crossfire board. __________________

Feb 17, 2006, 10:46 PM	#27
BWX unplugged Join Date: Nov 2002 Location: USNY Posts: 19,669 Rep Power: 110 System Specs	I think my system is pretty CPU bound in general.. I might do a few tests just see what happens. __________________

Feb 17, 2006, 11:31 PM	#28
Jerunk DriverHeaven Newbie Join Date: Oct 2005 Posts: 12 Rep Power: 0	I havn't read the whole thread, but stuff like this is definitely appreciated.

Feb 17, 2006, 11:46 PM	#29
Exceededgoku DriverHeaven Junior Member Join Date: Jan 2005 Location: Lincolnshire, UK Posts: 87 Rep Power: 0	Thanks for all this information, it is a real eye opener. ATi have really disappointed me lately... __________________ Vista Ultimate x64 edition rocks my world! Makes XP (and lower) look like dried poo stains on the wall! Core 2 Duo E6600 @ 3.2Ghz (400Mhzx8) 1.52V (set in bios, 1.47v real) \| 2GB GeIL PC6400 4-4-4-12 @ 1000Mhz 5-5-5-15 \| Gigabyte DQ6 @ 1600Mhz \| Sapphire 512MB X1900XTX @ 689/774 \| Enermax Liberty 620W \| Silverstone TJ07-B with custom watercooling \| Dell 2005FPW 3dmark05 - 13140 \| 3dmark06 - 6698 \| SuperPi 1M - 15s

Feb 18, 2006, 03:37 AM	Thread Starter #30
jpinard DriverHeaven Addict Join Date: Jul 2003 Posts: 278 Rep Power: 0	I mentioned earlier I did some research into user complaints about .net framework spamming (unrelated to ATI drivers) their system and their event logs. I decided it was worth trying a 6.2 driver-only install with no .net framework. I'm not sure how valid this is... the driver-only package installed the specific X1900 graphic device driver just fine, but could not install multimedia drivers (WDM NDF). But, I ran the tests anyways and the results for CAT 6.2 were the same. Higher graphic scores for everything, but much lower cpu scores. A comparison of "CAT 5.13 with CCC & .net framework" vs. "CAT 6.2 with no .net framework" :[color=red] [color=red]http://tinyurl.com/a59pe[/color] [/color]No sound card installed, just to make sure there was no conflict. In case there was any doubt - I'm not complaining about the graphic performance at all. I think the overall video performance gains in CAT 6.2 are wonderful. However, it's a different story when the processor is the bottleneck. And in that case, any gains made - and a lot more - is lost. I'm going to throw this thought out there... anyone think there's a possability the MVP switch card utilizing the second PCIe slot could be causing issues with CAT 6.2? I know it's probabaly a ridicuous thought...but I thought I'd ask. So far the only people able to replicate this are those with this type of system and an X1900. I'd be curious to see if anyone with a Crossfire board using 2 cards in Crossfire mode - would show the same cpu hit. Last edited by jpinard; Feb 18, 2006 at 03:49 AM.