Higher speed..lower score...wha.?

[Rondog]

I used 3D Mark01 at:
stock, Athlon @ 2211mhz (FSB-200) and 9600XT at 500/300 = 12640
overclock1: Athlon @ 2447mhz (FSB-222) and 9600XT @ 543/333 = 14700
overclock2: Athlon @ 2535mhz (FSB-230) and 9600XT @ 546/338 = 13000 ish

What could have caused this?

Im at school right now, so I cant run experiments, but when I get home i'll run some with FSB @ 225 etc. To see if the clock is too high. But neither prime95 nor superPI crash, they run successfully.

The board doesnt have an AGP/PCI lock, could this put more stress put on AGP/PCI by raising the FSB? Could this also cause the decrease in performance of the card? Or is it just too high temps to be worth while? Until I get better CPU/AGP coolers, other than the stock ones.

Stock: idle/load 30C/40C
Overclocked: idle/load 37C/50C

What are your thoughts on this issue?

[WonderBread]

i dont think the temps are the problem. how much did you relax the timings on the RAM? i think that the looser the timings the performance will be reduced, but im not quite sure.

[Rondog]

Ram timings were 2-2-2-6 for all the tests.

[WonderBread]

idk then, out of my knowledge. sry man

[XeroHouR]

its possible that maybe the program is continuning to run, but producing errors or hitching. These types of benchmarks are based off of fps typically, aquamark is, and i've gotten lower scores from higher clocks on CPU and GPU combined. Processing bad information causes it to be re-requested, and processed again(correct me if I'm wrong I'm no softare guy), this could result in longer process times, and thus, less performance. My guess though....

[Rondog]

So your saying that there is a point with overclocking that it becomes unworthwhile?

[XeroHouR]

I'd think so, maybe some of our software giants here can help out

[Rondog]

Worked it out, stupid Core/Mem on GPU was erroring far too high and yet it wasnt causing any artifacts. Slowed it slighty, and it benched at 14997, the increase was piss-poor so I didnt publish it.