4870X2 3d mark 03 bugged score thread

[massman]

Okay, I've come across the bug myself, although I have no idea what happened or what caused the bug to be produced. Note that this result can be reproduced and that there's no sign of black screens whatsoever. If I didn't know the scores at 375 and 395 FSB, I would consider this 88k run to be 100% legit.

I'll be conducting further testings tomorrow.

[Gautam]

Abnormally high GT2/GT3...but...in XP...

[Viss]

Not sure if 1st one is very bad or second one is too high.... Both XP sinlge card.

800/900 mem @ 833 7-6-5-18 1T:


800/900:

[CCC]

massman>

Rerun each of the runs when you have time at the same fsbs/cpu mhz clocks as in your testing but minimize each run as soon as it starts so its running in the background then have rivatuner’s monitoring proggy open ok and see if at each fsb setting when gt1 goes to gt2 and so on if each time the benchmark 'test' starts are the 3d clocks going to that of 750/900 or staying at 500/900

I cant see a random amount of cpu power causing this.

To say hey I have 4ghz now I only had 3700mhz before so now I can run 40%faster in gt2/3 is crazy unless the AI feature in CCC likes select cpu+mhz #'s more than others

Also rem single card acts like dual card so cpu power makes a world of diff, but it should be consistant form one guy at 600x10 6ghz to the next so my bet is the switching that goes on between tests in 03 from gt2-3 and 3-4 is not always working to set the 2d core clock back to 3d core speed untill gt4 starts

so if you can check this

[Viss]

This could be an issue but it would not explain some of the very high scores we have seen.

[GLENBOY]

the scores sort of indicate that its running at the lowest resolution 800 x 600 in the background maybe a gpu recover or something, my own scores seem to consistent but low and at 4800mhz cards at 800/940 2 cards in crossfire i get in the 129/130k region with good ram timings and speed

[.p4nzer]

I have made some try with my 3870.

I have made 5 tests, so little, but the results was always the same.

The black screen bug affect every card on Vista, but this differential beetwen score at same frequencies seems to be an issue of the 4870x2 driver or of the card behaviour.

Bye

[spl]

my 2003
http://it.youtube.com/watch?v=fzZuf1aHXDw

[massman]

Abnormally high GT2/GT3...but...in XP...

Abnormally high or are the other abnormally low? The more I test, the more I'm convinced that this so-called bug is not really a bug at all. I'll be posting a new graph soon that shows the exact point on which the GT2 and GT3 scores boost.

Rerun each of the runs when you have time at the same fsbs/cpu mhz clocks as in your testing but minimize each run as soon as it starts so its running in the background then have rivatuner’s monitoring proggy open ok and see if at each fsb setting when gt1 goes to gt2 and so on if each time the benchmark 'test' starts are the 3d clocks going to that of 750/900 or staying at 500/900

How can I minimize 3DMark03? The benchmark has to be visible always, no? I'll try to add an extra TFT to monitor rivatuner when I'm running the bench.[/QUOTE]

I have made some try with my 3870.

I have made 5 tests, so little, but the results was always the same.

The black screen bug affect every card on Vista, but this differential beetwen score at same frequencies seems to be an issue of the 4870x2 driver or of the card behaviour.

Bye

The 'black screen bug' is another kind of bug than the one I'm testing here. That one is actually a driver-related bug as the driver crashes and Windows tries to recover the driver to not crash the system. 3DMark03 apparently doesn't notice the crash and continues to run, resulting in higher FPS's.

[.p4nzer]

The 'black screen bug' is another kind of bug than the one I'm testing here. That one is actually a driver-related bug as the driver crashes and Windows tries to recover the driver to not crash the system. 3DMark03 apparently doesn't notice the crash and continues to run, resulting in higher FPS's.

I know.

I wanna only say that I think this differential works only on 4870x2.

Now I don't think it's a bug.

Bye

[CCC]

Abnormally high or are the other abnormally low? The more I test, the more I'm convinced that this so-called bug is not really a bug at all. I'll be posting a new graph soon that shows the exact point on which the GT2 and GT3 scores boost.

This wont say much as then why can guys on air and cold get the bug at fsb from 400-600 doesnt seem to matter with bad or good system ram timings/ratio also and cant be once you pass a select cpu mhz as we have seen this bug happen at 5500mhz but at 6200mhz 620x10 there is no boost at all.

See thats why this is such a pain as some of the best clockers in the world are tring to match this without sucsess or a run to run repeat basis

but an extra lcd hooked up to another monitor would be good

I will do the same tuesday when im off. I also have some new things to try 'outside the black screen bug' as you said massman this is not what we are tring to prove/repeat as that we know it is what it is a bug

this id start calling a Anomaly

[DeDaL]

E8600
Asus P5E3 Premium bios 0505
2*1024 Kingston 14400 (D9GTR)
1* 4870X2

Windows Vista SP1 32bit
nochipset Driver
Catialist 8.8 noCCC


I have ~89500 with Asus TOP bios and ~99800 with ATI referense Bios (the clocks for two runs is same)

I noticed difference on GT2 and GT3.

[massman]

I continued my tests today and I ran into some weird issues to say the least.



The data on the left of the vertical white line is the data of yesterday + extra findings of today. Basicly, I turned on my rig today and checked whether the boost was still there. I ran at 410FSB and the score was still 'bugged', thus I ran tests to find out on what point exactly the score is changing from not bugged to bugged. 404 is not bugged, 405 is bugged. I re-ran these settings and the results were the same.

After that, I tried to go up in FSB, but seemed to be limited by my TG ram, switched to OCZ PC9200's, did a CMOS reset and tried at 430FSB. Surprisingly, the bug was gone, my results were low 80k's. All tests in the 405-420 bug range turned out to be non-bugged with the OCZ's (data is not in graph, I can give you the data if you want).

Thus, I switched back to my TG's and switched cpu cooling. Now I was able to run 430+ FSB as apparently I was not limited by the memory, but limited by cpu temperature. The test data you can find in the graph; don't worry, I re-ran most of the FSB settings and turned out that NONE of the previous settings were giving me a boost!

Being fed up with this evolution, I increase Vcore/Vmem/Vmch and tried higher FSB settings ... surprisingly, the boost showed up again. 450FSB was rather difficult to boot at, due to temperature issues, but I reached 90k, which is about 10k too high.

Now, for what I can see, clearing the CMOS is the only thing that really solves the boost issues. Shutting down your rig doesn't help (yesterday I've shut down at a bit past midnight, first tests today were performed around noon, so that's a 12 hour gap). Maybe the only necessary action is to unplug the PSU from the power circuit, didn't test that.

The boost itself seems to pop up when the cpu is running at it's end due to temperatures. I'm quite sure that is not a cause, but rather a symptom or maybe has nothing to do with it.

[Marvin_The_Martian]

What does your last sentence mean? You say, it's seems to pop up with high cpu temps rather then any fsb settings, but then continue to say it's not the cause but probably a symptom?

You could test if it's cpu temp pretty easilly, get another setup pair it with a crappy hs without a fan and repeat your testing there ( other platform to ensure it's indeed the temp and not a bios setting you were unaware of having changed? ) maybe?

Sorry I just got confused about that sentence

[T_M]

Here's what i think is a non-bugged run?

http://hwbot.org/result.do?resultId=769658

[Gautam]

Interesting...that's a lot higher than I've ever gotten at 4.8GHz.

http://www-personal.umich.edu/~gautamb/135.jpg

You're killing me in every test by 30 FPS or so apparently. :S Yours seems similar to Kinc and spl's. There's gotta be something you're doing that me (and a lot of others) are missing...any ideas what that could be?

(I have a run at 136k with the memory at 1060...so...mem freq isn't the cause)

[massman]

What does your last sentence mean? You say, it's seems to pop up with high cpu temps rather then any fsb settings, but then continue to say it's not the cause but probably a symptom?

You could test if it's cpu temp pretty easilly, get another setup pair it with a crappy hs without a fan and repeat your testing there ( other platform to ensure it's indeed the temp and not a bios setting you were unaware of having changed? ) maybe?

Sorry I just got confused about that sentence

Sorry if I confused you, it's just to clarify the remark on the graph. I seemed to be hitting the workable temperature levels of the cpu everytime the boost occured. BUT, I don't think the temperature is the cause of this anomality, at most something caused by the real cause of the boost ... although that sounds rather dumb to me.

I don't change that much bios settings; only FSB and voltages.

[Marvin_The_Martian]

if it sounds dumb but there is no other visible explantion, test it

Just do a default run get a score, take the fan of the hs and put a hair dryer on it to help those temps abit

If it has anything to do with temps, you should see the same occurance?

[DeDaL]

Hm.....My GT2 and GT3 slow, but GT4 is higher.

And what is valid ?

[Gautam]

Hm.....My GT2 and GT3 slow, but GT4 is higher.

And what is valid ?

I don't even have any idea.

That is pretty weird...are you sure your clocks are setting properly? I get about the same score at stock card clocks.

[DeDaL]

Gautam I flash the cards - 840/970, its no freq play before run, i can run it 20 times and have same result.

my set:
only PCI-E = 118 (it is optimal for my board in 03/05)

[Gautam]

What if you change PCI-E back to 100? And Cat AI standard?

[massman]

if it sounds dumb but there is no other visible explantion, test it

Just do a default run get a score, take the fan of the hs and put a hair dryer on it to help those temps abit

If it has anything to do with temps, you should see the same occurance?

I just got my boost back, cpu was running quite hot/unstable again

[DeDaL]

What if you change PCI-E back to 100? And Cat AI standard?

Absolutely expected

130574

[DMH]

Is it just me or you have 2 different performance levels?