How does core clock and memory clock work on GPU's? What is their purpose?

[Phusius]

I mean, what do they effect exactly, my gtx 680 is at 100+ core and +300 mem.

but i honestly only did that cause some guy said to. what exactly does mem refer to? textures load faster, swap out faster? or is that core clock job?

 

[Stephen.]

From my observation Core clock in relation to the number of shaders if raised it increases the Pixel Fillrate, and for the memory in relation to bus width it increases memory bandwidth. All this changes raise your overall GPU performance hence more FPS

 

[Stephen.]

I hear that some games benefit from high memory clocks, one of those games that people say is Battlefield 3

 

[eidairaman1]

only drawback when overclocking it creates stability issues, so from sounds of it youre new to this, id run the hardware at factory for now then minor increase frequencies slowly

 

[Phusius]

I have had it at +100 core and +300 mem and 32%+ in evga precision for quite some time now and its running fine on all games.

 

[largon]

Any a chip can do a number of things (a fixed number or varying) in a single clock cycle, ie. "width of the device". Chips run at their typical frequency, that is the pace the parts inside can function as designed. "Throughput" is a basic measurement that combines width of the chip and frequency into a figure that tells us how much the chip can do in a given amount of time.

From my observation Core clock in relation to the number of shaders if raised it increases the Pixel Fillrate

Pixel fillrate is governed by the width of the ROPs (raster operators aka. "render back-ends") and core clock. Amount of shaders is not a part of the equation.

and for the memory in relation to bus width it increases memory bandwidth which intern leads to a higher Texture Fillrate.

Memory clock and texture fillrate are not related at all. Memory clock is the frequency the onboard memory chips' I/O bus operates. Texture fillrate is governed by core clock and the width of the texture mapping unit. Onboard memory is not just a static storage for textures, all functional parts of a GPU require access to the memory while they alter the data according to calculations issued by an app.

 

[Stephen.]

Any a chip can do a number of things (a fixed number or varying) in a single clock cycle, ie. "width of the device". Chips run at their typical frequency, that is the pace the parts inside can function as designed. "Throughput" is a basic measurement that combines width of the chip and frequency into a figure that tells us how much the chip can do in a given amount of time.
Pixel fillrate is governed by the width of the ROPs (raster operators aka. "render back-ends") and core clock. Amount of shaders is not a part of the equation. Memory clock and texture fillrate are not related at all. Memory clock is the frequency the onboard memory chips' I/O bus operates. Texture fillrate is governed by core clock and the width of the texture mapping unit. Onboard memory is not just a static storage for textures, all functional parts of a GPU require access to the memory while they alter the data according to calculations issued by an app.

I'm no expert, that's why i indicated from my observation.
Anyway since you have some knowledge on this kind of thing, if you don't mind please explain the following screen shots in details what is happening:

 

[the54thvoid]

Pics 2&4 give the info as to what does what.

With memory unchanged at 1050 but clocks increased you get:

Pixel Fill Rate up,
Texture Fill Rate up,
Memory Bandwidth down.

Why does bandwidth go down? Does it become a bottleneck for faster core?

 

[Stephen.]

Pics 2&4 give the info as to what does what.

With memory unchanged at 1050 but clocks increased you get:

Pixel Fill Rate up,
Texture Fill Rate up,
Memory Bandwidth down.

Nice explanation i have gotten so far,.. i hadn't looked at it like that,...

 

[Stephen.]

Why does bandwidth go down? Does it become a bottleneck for faster core?

Bandwidth goes down if you look at picture 3 1065 Memory Clock compared to 2 & 4 1050 Memory Clock

I did not see the need of running it at 1065Mhz so i maintained 1050Mhz since difference is Like 2Mhz

tho it can run higher than 1065 and still stable

 

[Phusius]

i feel like a nerd now, cheers

edit: officially* feel like a nerd now, cheers