Odd OCing performance limit

[86sixsix]

Hi all...

During my messing with a cobbled together AMD box...

Athlon X2 7750 BE, Asus M2N32 WS Pro, XFX nVidia GTX260, 4GB DDR2... yadda yadda...

I've hit an odd limit where no matter what clocks I increase - my benchmark results aren't going up as much as I'd expect...

The chip's at 200*15(3000)...the card's at 700core,1490shader,1275ram...the ram's at 4-4-4-12 2T...

If I up the graphics card...the benches don't improve (3dmark vantage, 06, 05... you name it)... and game frame rates don't improve any.

If I up the chip... either to a higher multiplier *16 (3200)...same story... (pi calcs don't get any faster) ...and the same in games.

...or if I hammer the FSB and run at 250*12 or *13 ...same story (even though the ram's now at 500...) ...same in games.

The ram controllers produce near enough the same results whether I'm ganged or unganged.

Regardless of what I change...the bench results and game FPS don't waver more than 3-5% and I put that down to Windows being twaddle...

I can't get the ram to play ball at 3... so that's out the window...

Has anyone any ideas of what's choking it up... I'm on a geriatric board... but I can't see that with this system the bus has gotten full...

EDIT: Stock, the system is at 2700 (fsb 200)... the card's 640 core (can't remember the ram/shader speeds)... Since I went past about 2950 on the chip all's I end up getting is a warmer room...

Answers on a postcard... the wackier the ideas the better...

 

[biscuitboy]

3-5% isnt a bad increase in FPS for a graphics card OC mate, What score u getting on SuperPi 1m? U wont see massive increases with small overclocks in Super Pi tbh and the AMD chips dont perform amazingly in SuperPi. Around 20s is about the right score for ur chip.

Are you running in the Right PCI-E slot at x16? Wont make a massive difference, but might free a couple of Fps.

Run the obv spyware checks and check in Msconfig that programs arent hogging any Resources etc...

 

[86sixsix]

No, that's not what I mean...

SuperPi's wombling along at 25 and a little bit... stock 2700 takes 30 seconds... 3000 takes 25 seconds... 3300 takes 22 seconds... 3600 takes 21 seconds...

200*15(3000) takes 25 seconds...

250*12(3000) also takes 25 seconds...

203*16(3250) takes 22.5 seconds...

250*13(3250) also takes 22.5 seconds...

i expected the faster northbridge, way faster ram (and lower latency) to produce faster benchmark results...

looking at the results above...there is no point me going over around 3700 as the sums won't get added up any quicker...

i'm pretty new to windows and it doesn't behave like any other OS i've got...freeBSD, ultrix, SCO all respond to clocks and ram throughput in a linear fashion...

hard faults and pages are well within sensible limits...i doubt there's spyware...tested with a new drive...new windows...and no net...to be sure.

the only common factor i can think of is the hypertransport...but surely a simple pi sum won't flatten it out... whether it's at 1000 or 1250...

or is it that superpi / windows is simply coded poorly and not optimised for us AMD people?

 

[86sixsix]

...never mind...i've just stuck 7 in a VM and ran it through a debugger...it doesn't seem to be issuing/producing very many halt states...

evidently it's just the OS load on the system. it begs the question why that task manager app says 0% on the cores when it plainly isn't...

...opening 3500 instances of notepad, minimising them all and hiding the taskbar brings the system to a crawl...surely it shouldn't make any odds?