670 Overclocks

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
Sorry I meant the voltage limit, the power limit is set at 122, sometimes Furmark can push it as much as 133.
 

Praxis1452

Platinum Member
Jan 31, 2006
2,197
0
0
Oh I just ran across this but Unigine Valley actually drives my TDP to 100% at 70C so (200W). If I hadn't set the power target to 200W it would actually be TDP limited. I'm shocked it has a higher TDP than furmark but that's what I'm getting.
 

24601

Golden Member
Jun 10, 2007
1,683
39
86
These kinds of tests are usually pretty pointless unless you set the power limit to higher than the card will ever hit.

This is because the IMC and the ROPs are usually tied to the core clock speed, meaning if you have it going down (from the card downclocking from hitting its TDP limits) it's going to skew your results down.
 

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
These kinds of tests are usually pretty pointless unless you set the power limit to higher than the card will ever hit.

This is because the IMC and the ROPs are usually tied to the core clock speed, meaning if you have it going down (from the card downclocking from hitting its TDP limits) it's going to skew your results down.

The power limit was set to the max it allowed me, 122. I made sure to monitor my clockspeeds with GPUz whilst running the tests and they were all at 1163, apart from Arkham city which occasionally clocked down to 1150. Probably because it had to spread the power over more units to compute the extra Physx.
 

atomheart

Member
Sep 9, 2012
41
0
0
At stock power settings, my gtx 670 sits at 1240mhz for the core. It's a galaxy GC model from the first batch of cards on release.
 

Praxis1452

Platinum Member
Jan 31, 2006
2,197
0
0
At stock power settings, my gtx 670 sits at 1240mhz for the core. It's a galaxy GC model from the first batch of cards on release.

I have a reference Galaxy GC as well. I think it has the longer 680? pcb too though I don't remember and I don't feel like opening my case since it's a pain.
 

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
Ok for those who are interested I've managed to get 1241MHz out of the GPU by increasing the voltage to 1.2v.

In order to do this I had to use the Keplar bios tweak program that Praxis mentioned, whilst having NVFlash in the same directory

At first I ramped up the voltage to just above 1.175 to see what it would do but it often lead to screen freezes as the actual voltage as going quite a bit higher. After raising the voltage again (just in case) and many restarts I decided to set the maximum voltage in the bios to 1.175, giving me a solid 1.200v when in game.

This meant once I'd tweaked my settings in Asus's util, I was able to gradually ramp it up to an offset of +150 which gave me a very stable 1241MHz, at no point does the voltage or frequency deviate from this during benching or gameplay like it did frequently when I gave the chip too much voltage or power, although now my fans are running at around 85% just to keep the chip below 70. Thankfully Asus's design doesn't make this unpleasant.

All the games I've tested it with do get improvements without artifacting, but i guess I will have to see if it is completely stable as I play games the whole way through.

On the other hand the memory latency speed factor has increased 5 points from 13.3 to 13.8, equaling some of the worst in my original chart so one day I may go back to the and start testing the memory all over again.

Saying that, Arkham City gave me the best result of 61FPS the first time I benched it, usually I have to bench it a few times as it sometimes it starts off at 37FPS instead of the 77 I know it's capable of. Also the minimum framerate is often skewed by the frames between scenes when the screen goes black. It will often leave a scene with a decent score and go into the next one with a much lower minimum framerate which lowered the average by a few fps so although it is not the most ideal benchmark, it still represents how a game with a lot of features has to balance them.



Edit: well that was short lived, as soon as I fired up Ungine Heaven and it got to the heavy tessellated dragon scene, it just froze again.

I tried down-clocking and it seemed happy at 1202, which also means I can lower the voltage to 1.187, which is good in a way as my card will probably last longer and it will cost less in the long run to run it. The maximum frequency I can get at this voltage is 1202 which I'm still happy with, it doesn't seem to deviate from this.
 
Last edited:

Termie

Diamond Member
Aug 17, 2005
7,949
48
91
www.techbuyersguru.com
Sounds like you have an average 670. People don't talk about average overclocking, just huge overclocking, but yeah, your card is like a lot of cards in regard to its limits.
 

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
Sounds like you have an average 670. People don't talk about average overclocking, just huge overclocking, but yeah, your card is like a lot of cards in regard to its limits.

Well I started the thread to get an idea of what memory clocks people where using and eventually it gave me enough info to get the most out of my card, which average or not i'm still very happy with, and i've learned a lot about how these cards work. Once I readjust my memory to suit the new overclock, I feel a Crysis 3 run coming on :biggrin:
 

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
After retesting the latency, it seems that once I overvolted, the bios must have loosened the timings as I'm getting higher speed factor results. So after retesting I've found that 7096 MHz gives me the best bandwidth to latency ratio at frequencies just below where I am prepared to go and games are running very smoothly. Overall I am very happy with this overclock.

For those looking for higher results, this is more of an informative thread people can learn from than an e-peen thread.

On another note, I've found if i set the new(ish) 3D mark fire demo to the extreme setting but at my monitors native resolution, I get exactly the same framerates!
 

Deders

Platinum Member
Oct 14, 2012
2,401
1
91
Sorry for the triple post but I've just got some new information for those wanting to find out the safest maximum memory clock for their card:

As I mentioned before Metro Last Light is a good benchmark that shows up lots of memory errors, I'd also like to mention that despite Metro being clear of issues, Crysis 3 had a memory related issue:

Just as I came out into the open after the lift drops in the Hydro Dam level, I cloaked and aimed a grenade towards the 1st 5 enemies that congregate as you emerge from the cage, and found that the aiming line wasn't the only line on the screen. So after knocking my memory clock down a couple of notches again this artifact disappeared and I can carry on playing safer in the knowledge that I'm not doing any (or at least as much) damage to the Vram.
 

gold333

Junior Member
May 4, 2009
4
0
0
Just found this thread, see if anyone can advise on Kepler overclocking.

Basically I have tried EVGA Precision, ASUS GPU Tweak and MSI Afterburner.

This is a review of all with a question at the end.

The card is: ASUS 670 TOP
The game is AC4: BF

So the problem is that using every setting locked (voltage, core&mem mhz, adaptive/performance in control panel, etc.) While everything is locked the game will show different FPS values (fraps) at a specific location depending on the instance I start it.

This 670 defaults to 1241 mhz (gpu-z) out of the box and I can run it at 1293.4mhz (gpu-z). I had to scale mem clocks down to 6200mhz (gputweak) or 1550 (GPU-Z) to stop it crashing.

I was seeing wild FPS fluctuations until I set the minimum voltage to 1150. This was because it doesn't throttle as much between the max of 1175 and 1150 but alot between 1175 and the default 0987 mV (sadly ASUS tweak and Precision wont allow any higher minimum volts than 1150, I'm talking to them to find out why.) Even that 0.025mv means 156mhz and around 4fps unnecessary fluctuations. I wish there was a way to force these cards to run at 1175 and 1175 only.

So ASUS gpu tweak was the best. It just worked. Every mhz set stayed constant with GPU-Z. 1190 in GPU tweak was 1293.7 in GPU-Z everytime.

EVGA Precision was an absolute nightmare. One boot +55 core (one wonders what that +55 starts from) was 1241mhz (gpu-z)the other boot it was 1306mhz (in gpu-z) I cant believe how people can overclock using EVGA Precision X. It was crazy as another time +77mhz was 1297. Also the K-Boost mode (forcing the boost to full even in 2D (so when unloaded.) Was a nightmare. 2 out of 5 times it would restart the machine when turning off. Also the set minimum voltage didnt stay constant, etc. I hate EVGA Precision.

MSI Afterburner is essentially the same program (down to the tool tips). It sucks even more because you cant even set a minimum voltage in Afterburner. How else are we supposed to stop the voltage (and therefore FPS) fluctuations?? Afterburner is so bad it can't even monitor GPU voltage.

So the question is how can we lock kepler cards to always operate at 1175mV and stop arbitrary fps fluctuations.
 
Last edited:

guskline

Diamond Member
Apr 17, 2006
5,338
476
126
My twin EVGA 670 FTWs under water in rig 1 below are OC'd by EVGA Precision with the following settings:
Power target 145%
GPU clock +60
Mem clock +415

GPU-Z shows GPU base at 1066 (stock 1006) and memory 1760 Mhz (base 1552). These are settings that I have run tons of benchmarks on w/o problems.

Again these are the FTW edition based on the GTX680 PCB with OC'd core.
 
Last edited:

Johnny4

Member
Nov 12, 2013
71
0
0
I have gotten it to +150 Core offset and +400 Mem offset @ 1.175 V Stable and I think thats as high as mine can go because a +155 it starts to fash in Unengine