Superpi maddnesss!!!!!!

[cadaveca]

okie, so i get this new stuff...rdx200 and mushkin uccc.

aquamark scores whacked out...over 100k w/x1800xl @ stock...

and these superpi runs :woot: :
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41197&stc=1&d=1134639173
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41193&stc=1&d=1134639035
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41194&stc=1&d=1134639035
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41195&stc=1&d=1134639035
http://www.xtremesystems.org/forums/attachment.php?attachmentid=41196&stc=1&d=1134639035


ideas as to what is going on?

[cirthix]

please fix the image with cpuz, you resized it and its unreadable

[M.Beier]

What OS?!?! - Seems kindda' bugged IMO

EDIT: The AM3 seems fair though, I get 102K at 3.15ghz I think,... With x850XT a bit clocked - but around same performance as x1800XL I guess

[bachus_anonym]

I would like you to let snq (http://www.xtremesystems.org/forums/member.php?find=lastposter&t=82419) (he is the author of the "Mod1.4" version of SuperPi) know about your results and see if he can find out why it happens.

[M.Beier]

One more question, take a watch, go away for 10 mins, and see if the watch on the computer reads 10 mins difference as well... Could be bugged time-reading @ windows, thats an old cheat... - 32M @ 1sec cheat :|

[cadaveca]

time seems correct. at the beginning of aquamark, i get 268FPS as a high, while the "ship" is still in the "furrow". i have never seen over 200 w/overclocked vid, nevermind the superpi times. Coutless reboots leads to the same behavior.

Time is accurate.

[snq]

I would like you to let snq (http://www.xtremesystems.org/forums/member.php?find=lastposter&t=82419) (he is the author of the "Mod1.4" version of SuperPi) know about your results and see if he can find out why it happens.
You called? ;)

I have an idea that it has something to do with Windows timers being off because of overclocking.. But this is a very weird case. I'm using multiple timing methods at the same time and I've never seen them both being off. If one of them would be off, that'd be seen as cheating by super pi.
Can you tell me what you did before you did these runs? And can you see if you can reproduce it?

Also if you can reproduce it, could you check if the windows clock is running too fast/slow?

[M.Beier]

Well, In my 24/7 settings, with high quality I see 202 in start of AM3, so dont think its all wrong :\ - But the SPI is veeery strange

[cadaveca]

was running 255x11 @ 1.4v
rebooted, tried to get ram to 275...didn't work..booted to memtestx86+ 1.65 disk only.


I cut off the IHS, remounted, lost 13c...

i set the FSB to 300, multi to 10, vcpu up to 1.5, vdimm up to 2.7, ht multi to 3

Mem set to 3-4-3-8-7-16-2-2-2-2-3120-1-Auto-0-7-3-8-5-16-disable-diable(8bursts)

primed it for 3 hours

ran super-pi...got 20 or so seconds.

tried 32<...got 18M or so

rebooted, tried 320x11 @ 1.6v, memtest ran a few passes, booted to windows, failed to boot windows, set to 300x10, booted into windows, tried systoll to up FSB, got to 308 using the auto method, back down to 305, ran superpi, for 18 or so seconds. killed processes in systray, then got a 17.7 second run.

[M.Beier]

Prime like in large FFT's ?? - Stable at 3ghz!?

Stepping @ 0541 ??

[snq]

Can you try these 2 things:
- Run original unmodded superpi and see if you get the same times
- Reboot and don't change any timings/clocks and run modded pi

[cadaveca]

cabye 0543 fpmw letter "K" in the serial #, "A" and a "H" under the IHS. "n" on the cpu pcb.
max case 65c, thermal offset 13c(had a really bad "silicone" job on the IHS), 67.6w TDP.

Something is whacked.

I did both, yes same times, and rebooting does not make a difference in times, and making changes in speed affects the final time as it should.

[[cTx]Philosophy]

Something has to be wrong here im sure, i just wanna post to see the outcome..

[snq]

Thanks for testing.
Windows' timers are somehow not running as they should.. But you should see the same thing in other stuff, a lot of games and other apps use the same timing functions as the mod does. It's nearly impossible that this problem happens just with superpi, it's not like it uses some kind of magic timer, uses regular timing functions that are widely used by other apps ;) I'd be suprised if the windows clock runs as it should, because I think it should run a few % too slow.
I'll slap together a test app with different timing methods to see if there's any that actually runs at correct speed. Will post it here when I'm done :)

[snq]

Here's a test app..

It uses 5 different timing methods, please check if any of them are correct (use a stopwatch or phone or whatever to check). Also check if any of them are running away from or falling behind the others. T5 may be ~15 msecs behind, that's normal, as long it doesn't fall behind more than that. T5 will probably also show incorrect results if you changed clock speeds since you started windows.

Press a key to pause it so you can actually read the times ;)

[sin0822]

wow that thing is cool, but my times are oof too and not only the T5 T1 is off by 2 ms and t5 by 3

[snq]

2 msec is quite normal, actually a difference up to 15 msec (which seems to be the resolution of most windows timers except T5) is completely normal. As long as the difference between them stays about the same give or take a few msec the timer should be correct.
But in cadaveca's case they should be off by several seconds after a while.

[cadaveca]

uh, like 300ms after like 5 minutes...if that.

problem is not just with superpi, like i said earlier, aquamark gets way high scores as well. I suspect you are right in this instance, but how to fix it ....re-format?

[sin0822]

yeah try a reformat, or maybe a differnt os?

[snq]

Okay, so the timers were all consistently wrong.. That means that they are wrong already while booting windows (that's when T5 is calibrated).
I hope a reinstall will help, but it might be that somehow the board is banana'd.. You could check in the bios if the clock is running fast there as well.. Or if you are getting too high bandwidth results in memtest or something. Anything that uses timing but not windows.

[cadaveca]

Yes it seems slow in bios...abt 3 seconds.

Further testing led to the problem....my drive makes this board toss it's cookies. i left it in bios for a while, and after 2 hours, only 10 minuts passed in bios. unplugging the drive solves the problem

What's interesting is that flashing the bios fixed the issue temporarily...but it slowly got worse and worse.
Very strange indeed....and i thought i had some monster cpu..lol...

very strange that it slowly happened, adn then finallly last night it got seriously bad..to the point that just 2d was lagging even.

i replaced each part..drives first...reconnected the drive and the problem comes back. I have an identical drive for RADI that i will try later to see if it's a compatibility issue...but it's definately not a windows or software issue.

[snq]

My knowledge kind of stops here, I don't know much about hardware, so I'm not sure how much sense this makes..
But I'm assuming there's some kind of chip that takes care of the timing.. And that chip seems to have gone mad on your mobo, or at least very sensitive to voltage changes (because it works when you unplug a drive). If you have a spare PSU around you could test with that but I doubt it'll help anything.

I'd contact the board manufacturer and see what they have to say about this.
Or maybe someone here has any more ideas now that we figured out that the problem is not with superpi or windows.