Overclocking and benchmarking: GTX 670 vs GTX 680

Antiflash

Limp Gawd
Joined
Jan 12, 2009
Messages
150
I first had a GTX 680. OC version from Gigabyte. It is one of the few with custom PCB in addition to its custom cooler.
When the GTX 670 was lunch last week I also bought the same OC version from Gigabyte. It uses the exact same PCB and cooler from the GTX 680. In fact it has a sticker on the PCB to cover the 680 model number that was printed by the PCIe connector.
p1030912gk.jpg

First I considered myself lucky with the Gigabyte GTX 680, I was able to overclock it to 1293mhz (core clock) that is above average for that card. I have tested three GTX 680: (1). This gigabyte, (2)EVGA Superclocked that only reach 1202mhz, and (3) a MSI Twin Frozr OC that was able to get to 1254mhz.
INTRO
First a few clarifications regarding overclocking this cards (nothing new)
I will not report clock offsets, only max clock. I have found that under load, the max clock is very stable. If you are not power constrained (increase power target) every card will reach its max clock under load and stay there without fluctuation.
This max attainable clock is different for each card and independent of the “nominal boost clock” that you get in the specifications of the card. In my experience all GTX 6x0 that I have tested go over the nominal boost clock to a intrinsic max clock that is variable from card to card but it does not change under any load.
The rule of gold is to try to maintain temp lower than 70C. As soon it gets to that temperature your max clock is decreased by 13mhz, and another 13mhz at 80C
When you increase the offset the max clock is increased by the specified amount until it is not stable anymore. But you are actually forcing your intrinsic max clock to any clock you desire.
Don’t bother with volgate, when your card is under load at its max clock it will use its maximum available voltage automatically. You cannot force any higher voltage.

OVERCLOCKING RESULTS

Gigabyte GTX 680 OC
Max Clock: 1293Mhz (20.7% increase over its base clock 1071mhz –Factory OC)
Max Memory: 3204Mhz (6.6% increase over 3003mhz)

Gigabyte GTX 670 OC
Max clock 1354mhz (38.5% increase over base clock of 980Mhz- Factory OC)
Max Memory 3404Mhz (13.3% increase)

As it turns out that 670 is an good overclocker. Is the advantage in clocks enough to cover for its disable functional units in comparison with the 680? Let’s see.

BENCHMARKS RESULTS

For all tests I used the following HW and Drivers
Procesador: i7-930 @ 3.52Ghz. MB: Gigabyte X58A-UD3R, RAM 6GB, 1600Mhz CAS 7-8-7
Nvidia drivers 301.34

UNIGINE HEAVEN
Settings: Tessellation Extreme, Anisotropy 16x, AA 8x, Resolution 1920x1080. Windowed mode.

Heaven Results for GTX 680 @ 1293/3204[/I]
Average FPS 50.9
Scores 1281
Min FPS 31.9
Max FPS 132.6

Heaven Results for GTX 670 @ 1358/3404
Average FPS 51
Scores 1286
Min FPS 32.7
Max FPS 131.3

3D Mark 11
Settings: “Basic” Performance test.

3DMark 11 Heaven Results for GTX 680 @ 1293/3204
3DMark Score P10101
Graphics Score 11204

3DMark 11 Results for GTX 670 @ 1358/3404
3DMark Score P9862
Graphics Score 10799

BATTLEFIELD 3
Settings All Ultra except MSAA (disable) only FXAA en High. Resolution 2560x1440
bf3n.jpg

BF3 Results for GTX 680 @ 1293/3204
Average FPS 68.6
Min FPS 43.7
Max FPS 113.4

BF3 Results for GTX 670 @ 1358/3404
Average FPS 67.5
Min FPS 40.8
Max FPS 111.4

TEMPERATURE AND POWER CONSUMPTION

Not the best "load" test but for simplicity I will use "Heaven" to compare temperature and power consumption between the cards.
In both cases Heaven was running in its loop for at least 10 minutes (all max at 1080p windowed) After the temperatures had stabilized I ran the benchmark with the results shown above (both cards with similar scores)
It does not matter to the test as you will see, but it is important to mention 680 fan duty was limited @ 85% and for the 670 it was unlocked up to 100%.
Another unusual characteristic of this cards is that they have one 8 pin 6 pin power connectors, the reference cards have two 6 pin power connectors. That means this card can draw more power from the PSU, that is somehow reflected in the power adjustment you can make in afterburner. It only goes up to 111%. But it does not matter because overcloked and loaded they hardly ever reach 100% (and normally much lower).

`Power and temp Results for GTX 680 @ 1293/3204
Clock 1293mhz (no fluctuations thru the test)
Max power 83%
Max Temp 62C
Max fan duty 79%
System power consumption (max as read in a kill-a-watt) 353Watts

Comment: as I mentioned above the fan duty does not reach its limit of 85%

Power and temp. Results for GTX 670 @ 1358/3404
Clock 1358Mhz (No fluctuations thru the test)
Max power 91%
Max Temp 69C
Max fan duty 93%
System power consumption (max as read in a kill-a-watt) 386Watts

Comment: It is close but in Heaven the temperature never reaches 70C and the clock is stable at 1358. Running greater loads as BF3 unfortunatelly the temperature can go up to 74C (even using 100% fan duty) so that game it gets down-clocked to 1345 most of the time.
What is interesting is that contrary to what I would expect, even without overclocking (only the factory overclocking) the GTX 670 always consumes more power than the 680. For example without overclocking (offset =0) the GTX 670 clocks itself to 1215Mhz and its max power is 85% in Heaven. The GTX 680 with offset=0 goes up to 1176Mhz and its power is only 77%.


CONCLUSIONS
From reviews we already knew that a Overclocked 670 could beat an stock 680. But comparing a good overclocked 680 with a good overclocked 670 shows that performance are very close.
Just check the forums to confirm that the 670 is a better overclocker than the 680. The difference in overclock overhead almost makes up for disabled SMX in the 670 providing 680 performance at a 80% of its price.
This specific Gigabyte 670 is a great deal because:
  1. It’s a custom design but it does not cost more than the reference cards
  2. It uses the same PCB design than its big brother. Just the better power delivery gives you better opportunity to attain a higher overclocks
  3. The cooler is better way better than reference, 10C to 20C cooler, a critical point if you don’t want to loose 13Mhz when reaching 70C. (unfurtunatelly I still reached up to 73C in BF3)
 
Last edited:
Thanks for the comparison! The 670 is one heck of a great card. Makes the 680 seem pointless for an extra $100.00.
 
Nice comparo. Nice to see that max clock achieved matters. I'm getting my 2nd card soon for SLI action.
 
thats a pretty beastly overclock for the gigabyte 670...most of them i see get around mid to high 1200's...I hope mine is a 1300 and up card
 
And this is why some 680 owners are cursing the video card gods. Imperceptibly close performance for $100 less per card.
 
That is a ridiculous overclock on the 670. Probably the highest I've seen.
 
the chart does not jive when the framerates listed. red, which is the 670, has the lowest minimum framerate on the chart yet in you say the 680 gets lowest minimum. and your chart shows the 680 with the highest max framerate yet you say the 670 has the highest.
 
the chart does not jive when the framerates listed. red, which is the 670, has the lowest minimum framerate on the chart yet in you say the 680 gets lowest minimum. and your chart shows the 680 with the highest max framerate yet you say the 670 has the highest.

I noticed that too.... maybe he labeled them wrong :rolleyes:
 
That is a ridiculous overclock on the 670. Probably the highest I've seen.

Exactly. No overclocking is ever a sure thing. Once you get a card that is supposed to overclock well and it doesnt you will never be happy. I didnt even think twice about spending the extra $100 for the 680. Just check out the 670 overclocking thread. Of course this is all coming from a 680 owner ;)
 
the chart does not jive when the framerates listed. red, which is the 670, has the lowest minimum framerate on the chart yet in you say the 680 gets lowest minimum. and your chart shows the 680 with the highest max framerate yet you say the 670 has the highest.

Ups, sorry, my mistake. It's been corrected now. For BF3, the 680 is in average 1 frame faster than the 670. The chart was correct, the numberes were backwards.

The problem with this test is that I could not get the 670 to stay below 70C (on the Gigabyte cooler @ 100%), it was running between 70C-73C. Effectivly it was automatically down clocked from 1358mhz to 1345mhz during most of the test. nVidia should seriusly consider modify this beheaviour. It was ok to run a 480 at 90C+ stock but now you start limiting the clocks at 70C!

Another mistake is that the is in the picture of the cards. They are sitting on their wrong boxes. You can notice it because the 670 has a white sticker close to the PCIe connector to correct that the PCB was printed with the 680 model.
 
Interesting. I have the reference 670 and it seems to max out at around 1300mhz as well, I've had it up to 1315mhz but it crashes occasionally. I've settled on 1215mhz due to using a 450w SFX PSU.
 
The windforce 680 for some reason has performance anomolies, Check out this review from OC3D:

http://www.overclock3d.net/reviews/gpu_displays/gigabyte_gtx680_windforce_review/1

Other factory OC'ed boards even at the same clock speeds are much faster. Maybe it has a faulty BIOS or something?

What you see there is a reviewer that lacks the understanding how this cards works.
Nvidia sell you the cards with two clocks
Base clock and boost clock. The first is the lower any 3D application will run. The second is just a guideline of what clock you will usually see when not power or temperature constrained.
For practical purposes you can ignore the "boost clock". It is only a way to tell you: your card will be usually autoclock higher than its base clock, and usually should be at around this "boost clock". Manufactures can't test every chip and print different specifications for individual cards, so they just define a conservative "boost clock" for a card segment and that GPUZ reads from the BIOS and it is the same for all cards that share the same segment. (reference, OC, superduper OC)
The thing is, each individual chip will boost at whatever clock its internal algorithms tell them. This is usually higher than the "nominal boost clock" and is different for each card.
What you see in that review is that the chip in the zotac card will reach a max clock that is higher than one in the gigabyte card.
The reviewer just compare what GPUZ tells about how much he has been able to overclock the "boost clock" and concludes that they are overclock at similar clocks. When he increase the offeset, the meaningless "boost clock" is increased accorndingly. (as well as the base clock and the real max clock). Even when GPUZ shows that they have the same overclocked "boost clock" in reality they are clocked at their intrinsic max clock that is different for each card.
The only way to check the real max clock is looking at the monitoring window of afterburner or presicionX and see dinamically what clocks the card is reaching. Even GPUZ can do this but not in the main window where it shows you the "boost clock", but in the monitoring window.
As disconcerting as it sound, with this technology you play the chip lottery every time you buy a card. Even the same model from the same manufacturer can perform quite differently. And I'm talking out of the box performance, not taking into account any overclocking. That is because, by design, they overclock themself automatically, and we all know, not all chips are made equal.
 
Last edited:
What you see there is a reviewer that lacks the understanding how this cards works.

It's like 10 fps slower (about -14%) than a stock 680 in BF3 ... something is clearly wrong with either the card, the system, the driver version, in-game settings, etc. A small difference in peak boost clock is not going to change the results that drastically.
 
I have tested three GTX 680: (1). This gigabyte, (2)EVGA Superclocked that only reach 1202mhz, and (3) a MSI Twin Frozr OC that was able to get to 1254mhz.

Just curious, how the heck you managed to get all those cards and what you ended up doing with them?

I was going to get an MSI Twin Frozr 680 but it was never in stock. Glad I ended up with the Gigabyte 670 OC, haven't had a chance to play with OC yet, been too busy enjoying games on the new system :).

Great post/analysis BTW.
 
It's like 10 fps slower (about -14%) than a stock 680 in BF3 ... something is clearly wrong with either the card, the system, the driver version, in-game settings, etc. A small difference in peak boost clock is not going to change the results that drastically.

I have to concede that something is wrong with that card. But I do not think it is common for all Gigabyte cards. I was able to test my Gigabyte GTX 680 against the MSI Twin Froz but only in Heaven and 3Dmark11 and they performed quite close clock for clock. Unfurtunatelly I did not run any comparison in BF3 and I no longer have the MSI to re-test.
 
Wow awesome post. Looks like the GTX 670 with a good aftermarket cooler is just the business.
 
Do you have the ability to test in triple monitor? I wonder if the extra Shaders and hardware may help the 680 there. If not then maybe I should have went the 4GB GTX 670 route since that's cheaper than a GTx 680 2GB

This test is very interesting. I hope everyone reads this and decides to buy GTX 670s so I can have a easier time finding GTX 680s :p:D
 
What you see there is a reviewer that lacks the understanding how this cards works.l.


Dude, not to sound snide, but Most hardware editors are not morons. These are intelligent people that have been in the "game" a long time and are familiar with the technology behind the silicon! OC3D has been around a long time and he knows what he's doing. The card just has weird performance anomolies there, which other OC'ed cards did not have.

The zotac AMP! and reference OC'ed 680s in his test benchmarked faster. I don't get it, this isn't the first website review i've seen which showed weird performance issues with the gigabyte windforce 680. I suspect this is why the OP's 670 outperformed the 680, that should not
happen at similar clock speeds, the 670 is minus an entire SMX cluster.
 
Last edited:
this isn't the first website review i've seen which showed weird performance issues with the gigabyte windforce 680.

If there are other reviews reporting this issue with this specific card that should somehow confirm that something is wrong Gigabyte OC version and not something only affecting that review. Really interested to have additional info, are you able to provide more links? If so, I can get the MSI card back from a friend next week to discard this is affecting my card.
 
that review was just weird. there were even games where the oced Zotac AMP card was getting beat by the stock Zotac AMP card. there might even be some other oddities but I noticed that just form skimming the article.
 
Nice information there.

I have two MSI 670s, and haven't had much time yet to test them.
All I can say so far is these 670s are very fast.

I run three monitors in Surround, so I can provide some input on performance once I,ve had some time.

I do not have any 680s laying around to compare them to, unfortunately.....:eek:

But.....I may be able to run some comparisons using two Crossfired HD 7970s?:cool:
 
That's an outstanding overclock on the 670. Not everyone is going to be lucky enough to get that!
 
I was going to get the GTX 670 until I read this review..

I mean, even at that insane overclock, it's still weaker than the GTX 680's mundane OC, it takes 15% more power, it's significantly hotter, and for what - 50-100$ less (consider the special editon)? Not worth it for me.
 
I was going to get the GTX 670 until I read this review..

I mean, even at that insane overclock, it's still weaker than the GTX 680's mundane OC, it takes 15% more power, it's significantly hotter, and for what - 50-100$ less (consider the special editon)? Not worth it for me.

I came to the same conclussion. Allready have one GTX680. Was going to sell it and had 2 gtx670's on the way. Cancelled the GTX670's and ordered an extra GTX680. Between the husttle of having to sell the GTX680 online and the news of people having problems with their 670's I figured the extra hundred bucks was not that big of a deal. What I really wanted was a 690 but this will do the trick. May have to rearange my case a bit to fit them both though.
 
I was going to get the GTX 670 until I read this review..

I mean, even at that insane overclock, it's still weaker than the GTX 680's mundane OC, it takes 15% more power, it's significantly hotter, and for what - 50-100$ less (consider the special editon)? Not worth it for me.

670 "insane" overclock and then 680 "mundane"? They are both respectively good OC. I Haven't seen quantitative evidence but reading the forums people seems to be getting low twelve hundreds or the 680 and close to 1300 for the 670. Both OC here are about 50mhz above " average".
Valid point on the power, and very unexpected because 670 is supposed to have lower TDP, after all it has fewer Cuda cores.
 
Two GTX 680s I personally tested did 1325 stable. Truth is it's the luck of the draw for all of them. But probably not fair to compare a super overclocked 670 to a moderately overclocked 680. You never know what you're gonna get. If I bought a 670 expecting it to overclock like this based on a review (I wouldn't - I'm not that ignorant - but some are) and it didn't, I'd be p!ssed.
 
Last edited:
Isn't it odd that the 670 Gigabyte model with the same cooler is the same price as the reference? How is that possible? They had to have cut some corners, like with the heatsyncs and fans maybe?
 
Isn't it odd that the 670 Gigabyte model with the same cooler is the same price as the reference? How is that possible? They had to have cut some corners, like with the heatsyncs and fans maybe?
maybe they have a ton of those pcbs and coolers but not enough gtx680 chips.
 
670 "insane" overclock and then 680 "mundane"? They are both respectively good OC. I Haven't seen quantitative evidence but reading the forums people seems to be getting low twelve hundreds or the 680 and close to 1300 for the 670. Both OC here are about 50mhz above " average".
Valid point on the power, and very unexpected because 670 is supposed to have lower TDP, after all it has fewer Cuda cores.

This exactly, and it is 135 dollars less too. I would say 1220ish is average for a 680 and 1300ish for the windforce gigabyte 670. They end up being effectively the same speed for a lot less cash, especially for sli.
 
Is it? You've taken it apart then and posted pictures? I'd love to see them.

The Windforce 680 and 670 have the same PCB and cooler to my knowledge which is probably why they perform so close to each other. Gigabyte has a real winner this generation with the Windforce 670.
 
The Windforce 680 and 670 have the same PCB and cooler to my knowledge which is probably why they perform so close to each other. Gigabyte has a real winner this generation with the Windforce 670.

That is correct, it's been compared by reviewers and it is literally the exact same PCB.
 
Back
Top