Sign in with
Sign up | Sign in
Your question

GTX670/680 core vs memory

Last response: in Graphics & Displays
Share
October 18, 2012 9:23:21 AM

Hi everyone,

This is my first post and probably the first of many. I'm interested in finding out the results of people's overclocks on the gtx670 and gtx680 cards and how those results differ based on core vs memory clocks.

The reason I ask is I have a gigabyte gtx 670 which boosts to 1228mhz on the core (not too high) but the memory is overclocked to 7200mhz (stable in heaven and no crashes or anything in any games (max temp so far of 67C)).

The memory seems to be making more of a difference for me. When testing on heaven, my fps only increased by about 3 over stock when I boosted the core clock (+55mhz) but when I boosted the memory I got a massive increase even after a small overclock. My theory, based on my card, is that the memory overclock 'appears' to make more of a difference to a stronger core clock. Even at 6600mhz i was getting a huge boost in performance.

My friend has a gtx670 overclock to 1290mhz boost and 6500mhz memory and I smoke his card even though he has a much higher core clock (asus DCUII). In heaven the results are about 2 - 3 fps difference.

My heaven benchmark results (would post a pic but at work atm!):

Avg FPS: 52.4
Min fps: 25 (ish)
Max fps: 134

My buddy manages about 49.9 average with a similar system.

So to conclude my mammoth wall of text, are your cards providing similar results? Is this limited to the gigabyte brand or is my card just an odd one!

Cheers,

Reaper.

More about : gtx670 680 core memory

October 18, 2012 9:31:03 AM

A memory overclock will benefit your card, if you are limited by the memory bus of your card. A general trend (though not true in every case) is that when you are running games at very high resolutions a higher memory overclock will increase your frame rates, but usually won't at low resolutions. An analogy for this is having a balloon connected to a pipe. Squeezing the balloon harder is like having a faster core clock, enlarging the pipe is like having faster ram.
October 18, 2012 9:38:39 AM

theLiminator said:
A memory overclock will benefit your card, if you are limited by the memory bus of your card. A general trend (though not true in every case) is that when you are running games at very high resolutions a higher memory overclock will increase your frame rates, but usually won't at low resolutions. An analogy for this is having a balloon connected to a pipe. Squeezing the balloon harder is like having a faster core clock, enlarging the pipe is like having faster ram.


Hi Liminator and cheers for the reply,

I only play at 1080p, probably should have mentioned that before! So it's not like I'm playing at a really high resolution.

I'm just interested in other people's real world results as most people seem to focus on core clocks over memory, whereas in my own experience this isn't necessarily the way to get the fastest gpu.

In response to your rather great explanation; I'm wondering if maybe the cards at stock are quite limited by their RAM? Most kepler cards seem to boost way past their supposed 'boost' clock but obviously the RAM doesn't budge. Maybe this is something OCer's should look at a little more?

Just a thought!

Cheers,

Reaper
Related resources
October 18, 2012 9:38:44 AM

theLiminator said:
A memory overclock will benefit your card, if you are limited by the memory bus of your card. A general trend (though not true in every case) is that when you are running games at very high resolutions a higher memory overclock will increase your frame rates, but usually won't at low resolutions. An analogy for this is having a balloon connected to a pipe. Squeezing the balloon harder is like having a faster core clock, enlarging the pipe is like having faster ram.


Hi Liminator and cheers for the reply,

I only play at 1080p, probably should have mentioned that before! So it's not like I'm playing at a really high resolution.

I'm just interested in other people's real world results as most people seem to focus on core clocks over memory, whereas in my own experience this isn't necessarily the way to get the fastest gpu.

In response to your rather great explanation; I'm wondering if maybe the cards at stock are quite limited by their RAM? Most kepler cards seem to boost way past their supposed 'boost' clock but obviously the RAM doesn't budge. Maybe this is something OCer's should look at a little more?

Just a thought!

Cheers,

Reaper
October 18, 2012 9:40:27 AM

Sorry double post :( 
October 18, 2012 10:11:23 AM

Geforce GTX 670 vs Geforce GTX 680
Power Consumption (Max TDP)
Geforce GTX 670 170 Watts
Geforce GTX 680 195 Watts
Difference: 25 Watts (15%)

Memory Bandwidth

Performance-wise, the Geforce GTX 680 should theoretically be a little bit superior to the Geforce GTX 670 in general.

Geforce GTX 680 192256 MB/sec
Geforce GTX 670 192000 MB/sec
Difference: 256 (0%)
Texel Rate

The Geforce GTX 680 will be a lot (about 26%) more effective at texture filtering than the Geforce GTX 670.
Geforce GTX 680 128768 Mtexels/sec
Geforce GTX 670 102480 Mtexels/sec
Difference: 26288 (26%)
Pixel Rate

The Geforce GTX 680 is a bit (about 10%) faster with regards to FSAA than the Geforce GTX 670, and also should be capable of handling higher resolutions more effectively.
Geforce GTX 680 32192 Mpixels/sec
Geforce GTX 670 29280 Mpixels/sec
Difference: 2912 (10%)


Core Speed 670 Gtx - 915 MHz
680 Gtx - 1006 MHz
!