To Sync or Not to Sync clocks... Crossfire O/C Question

[jacktesterson]

I have 2 7970's, a Diamond 7970 and an Asus DCII model. I picked up both used off ebayat good prices used, and the 3GB will help down the road at some point I'm sure. This is going to be a fairly long term build due to other financial issues coming my way.

Due to the style of case being used, I want to stay at Stock voltage to keep temps down.

At stock voltage, here are my individual core results (memory not tested yet)


--- Asus 7970 DCII is stable at 1180 MHz Core Stock voltage (Up from 925)

--- Diamond is stable at 1050 core stock voltage.


As you can see, the Asus clocks much better. The Diamond requires voltage bumps to go higher. Its not a great clocker at all.


MILLION DOLLAR QUESTION


Am I better running both cards at 1050 Core in crossfire (Currently Am), or Would I be better off running the Asus at 1180 and the Diamond at 1050 in crossfire? (Which in limited testing seems to work).

Thoughts?

 

[GotNoRice]

You're not really gaining anything by having one card with faster clocks. It will be held back by the slower card.

You could have your faster card as primary, so if you ran in windowed mode or another situation where crossfire does not function, you'll get the best performance possible.

I usually run my GPUs at the same clocks because I'm pretty anal about that sort of thing.

 

[JAG87]

You're not really gaining anything by having one card with faster clocks. It will be held back by the slower card.

FYI, this isn't true. If card 2 is not ready to draw another frame, card 1 will start drawing again, it won't wait on card 2. He will gain something.

 

[jacktesterson]

FYI, this isn't true. If card 2 is not ready to draw another frame, card 1 will start drawing again, it won't wait on card 2. He will gain something.

Good to know.

Will adjust tonight and see how it works out.

 

[3DVagabond]

FYI, this isn't true. If card 2 is not ready to draw another frame, card 1 will start drawing again, it won't wait on card 2. He will gain something.

Yes, this is why "Frankenfire" (Crossfire with 2 different cards) works and you get performance somewhere in between the 2.

 

[notty22]

I can't speak for crossfire or other sli setups, but mine using identical cards/bios's I tested with different clock speeds, just because one was a better o/c .
The card with higher clock speed always showed less gpu usage, this is in games where I would get 99% gpu usage at times. The higher clocked card would show less gpu usage. Almost like load-leveling was taking place.

 

[GotNoRice]

FYI, this isn't true. If card 2 is not ready to draw another frame, card 1 will start drawing again, it won't wait on card 2. He will gain something.

The faster card will be held back. You can see this effect as clear as day if you monitor GPU usage. You will notice that the GPU usage of the faster card is lower.

If there is any benefit, it's minor.

 

[blackened23]

Uneven clockspeeds in SLI is actually common place with the Kepler, and I haven't found a major performance disparity from it. I'd say you're fine with uneven clocks, perhaps you could close the gap a slight bit though?

 

[JAG87]

The faster card will be held back. You can see this effect as clear as day if you monitor GPU usage. You will notice that the GPU usage of the faster card is lower.

If there is any benefit, it's minor.

Yes, perhaps held back of how the card could perform by itself, but nobody is referencing single gpu performance. In tandem, any kind of benefit however minor it is, destroys your argument of being "held back". Held back means performance should remain unchanged from two identically clocked cards, which is not the case.

 

[GotNoRice]

In tandem, any kind of benefit however minor it is, destroys your argument of being "held back". Held back means performance should remain unchanged from two identically clocked cards, which is not the case.

Sounds like you had to create you own definition for "held back". I think it's pretty clear what I meant. If there is any advantage, it's probably canceled out by the additional micro-stutter that mismatched clocks introduce.

 

[jacktesterson]

So after some testing last night, I find it better to run at Sync'd clocks.

There is little improvement with performance, and the Asus card's GPU usage falls back a little too.

 

[3DVagabond]

Just curious, what was better with the clocks sync'd?

 

[jacktesterson]

Just curious, what was better with the clocks sync'd?

Running heaven and 3dMark11 had some microstuter I never experienced when clocks were sync'd.

Besides that, Seen maybe a couple FPS more in most situations. GPU usage was around 90-95% on the Asus too. Not worth it right now. Happy at 1050 crossfire.