Core Clock vs Mem Clock vs Shader Clock - A few questions

[MaxAwesome]

Hey guys,

I have an Asus 8800gt (the one with the glaciator cooler) and I have it clocked @

Core: 690mhz (stock 600mhz)

Shader: 1690mhz (stock 1500mhz)

Mem: 1050mhz - 2100mhz effective (stock 900mhz - 1800mhz effective)

This is without a doubt a great overclock, especially for the memory. it totally blew my mind, a 300mhz increase without crashes/errors. It brought 3Dmark score up from 11k to 13k+. Not bad at all :lol: That's above Ultra performance right there

But now this brings me to my question: When overlocking, which of these clock domains will yield more performance? Is is the core clock? The Shaders? Memory?

For example, is it worth it keeping the memory this high? Is it dangerous by the way? Meaning, could this memory overlclock prove stable right now, but in the long run cause the card to crap out? :/

Should I lower the Mem clock and try to increase core/shaders? I'm trying to find out which of these GPU domains is more responsible for the overall performance.

[PennySavre]

For my 8800GTS 640MB the Core overclock made quite a bit more difference than the shader overclock... and this was on a card with only 96 shaders.

I would think the Core overclock would make the most difference. Overclocking the memory and shader would probably make a decent difference too but maybe not quite as much.

[mrhuggles]

agreed

[wolf]

i think youll find that if you oc all 3 by themselves, from most gain to lease you have

GPU
Shader
Mem

Now mem overclocking is a grey area as some games will hugely benefit from more memory bandwidth and others will not.

however i always recommend overclocking the lot at it tends to have a nice synergistic effect.

for example GPU+shader oc may give you ~7% more fps, and the mem oc alone gives you ~3%, together they could give you more than the combined ~10%, ive seen it happen a fair amount, and it makes sense, and the gpu runs faster it will want a faster connection to its frame buffer to keep up with it.

[spearman914]

Core clock acts like a CPU clock except on a GPU. Mem clock is how fast data is transfered to VRAM. Higher res will have a better pay off for Ocing the mem. Shader clock is how fast math calculations are done.

[farlex85]

Core clock is actually the least difference maker in my 8800gts 512. Memory actually seems to yield the biggest gains, shaders next (go the furthest though), then core clock last. The core has to keep up w/ the other two, but if you need to lower your core clock to get higher memory and shader speeds, I would (to an extent, find the max of all clocks, then back off on the core a few notches and up the shader and mem to the next place up).

[spearman914]

Originally Posted by farlex85

Core clock is actually the least difference maker in my 8800gts 512. Memory actually seems to yield the biggest gains, shaders next (go the furthest though), then core clock last. The core has to keep up w/ the other two, but if you need to lower your core clock to get higher memory and shader speeds, I would (to an extent, find the max of all clocks, then back off on the core a few notches and up the shader and mem to the next place up).

What were the oced/stock clocks and ?

[farlex85]

Originally Posted by spearman914

What were the oced/stock clocks and ?

For me? Stock clocks are 670/972/1674 (core/mem/shader) and the highest 3dmark score I was able to achieve was w/ 821/1152/2052. I can bump the shaders down to 1998 and crank up the core to near 850, but I lose some points by doing so. The shaders can only be clocked in 50mhz increments, so imo it's best to find their high point then tweak the core and mem.