NVIDIA GeForce GTX 670 Reviews

[Krizby87]

@bhavv: you can try disable boost clock on the driver (power management mode) and try for the maximum oc. Should have waited until overvolt is usable when you made your bet :-/

Here is with my GTX 680
No Oc


Mem oced to 6800Mhz


200 Pts (1.8%) increase with 400Mhz overclock (13%), seems like the GK104 in not really bandwidth craving that much, then again I'm benching at 1280x720.

[Mungri]

Im trying to do the tests for you, but 3Dmark11 keeps getting this error in the final combined test:

Workload work failed with error message: Process exited unexpectedly

I cant seem to get it to work.

[Squuiid]

Im trying to do the tests for you, but 3Dmark11 keeps getting this error in the final combined test:

Workload work failed with error message: Process exited unexpectedly

I cant seem to get it to work.

This is a bug in 3Dmark11. Change the option "GPU Count" from Automatic to be either 1 or 2 depending on if you have 1 GPU or 2 if you have a GTX590/690 or SLI, etc.
This will stop this error from occurring.

[LostHawk]

Im trying to do the tests for you, but 3Dmark11 keeps getting this error in the final combined test:

Workload work failed with error message: Process exited unexpectedly

I cant seem to get it to work.

I had the same problem with my new 3770k @4.6GHz build yesterday. Perfectly stable with LinX.
Same error message in final combined test and also in Vantage final tests.
I had to raise the cpu voltage a little bit and problem solved... (+0.015v)

Results: 3d11: P11553, 3dVantage: P38171 (gtx580s here not 6x0s)

[Mungri]

You know I've never liked reference design coolers. I always saw them as low quality / inferior compared to all the custom built cards / coolers that the AIBs made, but compared to the reference GTX 670, all the rest of these reference coolers look like the most high end and premium quality coolers you can get:

[RPGWiZaRD]

^ Wow, with GTX 560 Ti cooler on GTX 670 it would probably already never pass 60C load haha. Not suprising that manufacturers try to save as much up on costs as possible and GTX 670 runs very cool and stays under 80C with that cooler so what problem is there (from Nvidia point of view)? :P Still for a $399 card you'd expect a little better cooler though for sure.

[rusty]

I need this http://wccftech.com/zotac-geforce-gt...tion-detailed/

[Mungri]

^ Wow, with GTX 560 Ti cooler on GTX 670 it would probably already never pass 60C load haha. Not suprising that manufacturers try to save as much up on costs as possible and GTX 670 runs very cool and stays under 80C with that cooler so what problem is there (from Nvidia point of view)? :P Still for a $399 card you'd expect a little better cooler though for sure.

The card could run very cool, but at 70 degrees it throtles maximum boost clocks by -12 Mhz, and voltage by 0.025v, and at 80 degrees -25 Mhz and 0.05v.

Nvidia made a card with these slight thermal throttles at 70 and 80 degrees, but put such a poor heatsink on it that it runs past 80 degrees at stock speeds and activates the 25 Mhz and 0.05v throttle.

If they used a better heatsink, even just the 560 ti cooler, it would be staying under 60 degrees at stock speeds like the Gigabyte Windforce does. My KFA2 stays under 70 at 1250 Mhz, which is enough to prevent the overclocked frequency from reducing. It doesnt matter at all for games as there wont be any difference, but when your trying to run benchmarks to see what the card is capable of, its no good that the GPU boost feature automatically reduces the clocks.

I'll be setting my CPU and GPU speeds back to factory settings now anyway as I'm done stressing my card. Lots of people have been returning their KFA2s to OCUK to replace with a Gigabyte Windforce or Asus non top because at stock speeds and even with a full 75% fan profile, the card cannot keep itself under 70 degrees, while the Gigabyte Windforce stays under 60 degrees at 1250 Mhz and stock fan settings.

They went and increased the prices too as a result on the gigabyte and asus cards, the KFA2 is still £360, the Gigabyte £370, and the Asus non top is £380? Thats just daft when the GTX 680 reference design is only £399.

I need this http://wccftech.com/zotac-geforce-gt...tion-detailed/

I wouldnt reccomend that one, seriously. Its the exact same cooler as the KFA2 / Galaxy card but with an extra heatpipe:

http://i.imgur.com/WuB7h.jpg

Its not as good as the direct heatpipe contact coolers from Asus and Gigabyte.

[Mungri]

Bhaav ? what is the official turbo boost of this card ? 1200mhz ?

Depends on the base clock of each model, and some are varying. The average boost clocks that most people are getting are:

915 Mhz base, 1084 Mhz boost
980 Mhz base, 1189 Mhz boost
1006 Mhz base, 1215 Mhz boost.

But I've seen people on OCUK who have bought either the Gigabyte or KFA2 that are getting lower than those, and one person even got a better KFA2 with 1228 Mhz boost clock, and it was overclocking to 1359 Mhz! In all these cases, a GPU-Z screenshot confirmed that the base clock was as advertised (980 or 1006 Mhz), but those people with lower boost clocks had gotten chips that werent stable up to 1200-1250 Mhz.

If you dont have the boost clocks mentioned above, which happened with very very few people only, I'd send the card back. Everyone that got a Gigabyte or KFA2 card working at the correct boost clocks above managed a minimum stable overclock of 1250 Mhz. People with lower boost clocks were getting instability when clocking to, or a little over 1200 Mhz.

In most cases, buying the 980 or 1006 Mhz base clock cards should guarantee you a 1200+ Mhz chip, but a few that slip past quality control checks are not boosting up to or stable at the correct boost clock fequency.

I'd imagine from the look of current results that fewer than 1% of chips are clocking past 1300 Mhz, and the same number are incapable of 1200 Mhz (on the 980 and 1006 base clock cards that is). When buying a 980 Mhz (Gigabyte / Asus TOP), or 1006 Mhz (KFA2 / Galaxy, Zotac, Palit, Gainward), you should be getting a handpicked 1215-1250 MHz capable chip at the least, unless the card slipped through being tested.

I'd also definitely stick to either the Gigabyte or Asus card if getting one of these custom designs, their direct contact heatpipe coolers are around 10 degrees lower than the KFA2 / Galaxy cards are, giving around 60 degree temperatures on the Gigabyte at 1250 Mhz. I havnt seen any user results from the Asus card yet, and wouldnt trust the review results as official as they use cherry picked cards, but the cooler that Asus use is the same quality as Gigabyte's with one less fan.

From the information I've seen so far from across reviews and user results:

KFA2 / Galaxy - 81 degrees at stock settings, 71 degrees at maximum fan speed, <66 degrees up to 1250 Mhz after a cooler reseat to elimitate throttling (my own results).
Asus - 70 degrees at stock settings (turn up the fans a little and enjoy no throttling, results from techpowerup).
Gigabyte - 60 degrees at stock fan profile with the card clocked up to 1250 Mhz (results from users such as Krizby in this thread).

[nascasho]

I have a 680 vs 670 run on 3DM11, talk about close!

So far the suckers been going through 4hrs of BF3, gonna get some D3 torture later though.

[Mungri]

I have a 680 vs 670 run on 3DM11, talk about close!

So far the suckers been going through 4hrs of BF3, gonna get some D3 torture later though.

Any chance you could test them clock for clock (same boost not base clocks)? If you dont have your 680 anymore, just run the 670 at the same frequencies to the results you already have.

It would be nice to see how much, or how little difference the disabled SMX unit makes.

A GTX 670 and 680 with the same base clocks (1006 Mhz) perform almost identically due to the 670s higher boost clock.

[Lanek]

Bhaav can you check the Asic quality of your 670 ? ( with GPU-z, it work for AMD cards, should work too on Nivida one's ).


If i have ask you what was the boost clock on yours, is due to the fact it differ on lot on the 680 retail cards... due to Asic quality, whatever is write, some do some do 1058 some 1084 and ofc the best 1110
mhz . ( dont get the right number in head, will correct them if needed when i got the article of hardware.fr , note all press sample was 1110mhz ones ) ..

So basically this will ofc rely on the performance throttling on the boost clock, but ofc certainly with the 125% limit. It is well possible many 670 non reference goes higher due to the Asic quality and specially as the tdp limit is higher or same of 680 on some case, with 1 smx disabled they go higher.

Ofc temp influence tdp... so the limit is maybe faster reach. On certain model ..

Now i will be honest with you, when i have seen the gigabyte and Asus, the choice was easy, higher pcb and cooler quality. the choice is easy. ( specially for the difference on price )

[nascasho]

Any chance you could test them clock for clock (same boost not base clocks)? If you dont have your 680 anymore, just run the 670 at the same frequencies to the results you already have.

It would be nice to see how much, or how little difference the disabled SMX unit makes.

A GTX 670 and 680 with the same base clocks (1006 Mhz) perform almost identically due to the 670s higher boost clock.

Both were run at 1260/3444 in AB. Don't have my 680 anymore as I was waiting for one that would make sense to WC it.

During the run I was monitoring the core freq and temp with the latest AIDA64 and it was locked at the 1260/3444 (6888) and 1.175V. Same with the 680, but I only have a results that I was trying to compare to my 480 I had.

Have a few Crysis and metro runs too if you want me to compare.


Also made this earlier yesterday:

http://www.youtube.com/watch?v=B9sd_HJLVo4&hd=1


EDIT: Also, wanted to check my ASIC and GPU-Z read 100%? Guessing it's not reading it correctly for these cards yet?

[Lanek]

Both were run at 1260/3444 in AB. Don't have my 680 anymore as I was waiting for one that would make sense to WC it.

During the run I was monitoring the core freq and temp with the latest AIDA64 and it was locked at the 1260/3444 (6888) and 1.175V. Same with the 680, but I only have a results that I was trying to compare to my 480 I had.

Have a few Crysis and metro runs too if you want me to compare.


Also made this earlier yesterday:

http://www.youtube.com/watch?v=B9sd_HJLVo4&hd=1


EDIT: Also, wanted to check my ASIC and GPU-Z read 100%? Guessing it's not reading it correctly for these cards yet?

Lol yes surely, as it is quite impossible to get 100%..

What soft ware you using for recording? fraps ?

[Chickenfeed]

NCIX got in a decent amount of stock on the Gigabyte 670 OC so I cancelled my Asus preorder. For $10 over reference Gigabyte is offering one hell of a card so it looks hard to go wrong.

[cold2010]

What soft ware you using for recording? fraps ?

MSI Afterburner

[Lanek]

i was not find the quality excellent when use MSI, but this was a long time ago, will try it again. so i have allways use fraps ( with the big problem of fraps who is the size of the files )

[Mungri]

I didn't know that about the Asic quality, my boost clock is 1215 Mhz like most people are reporting for this card. Good GPU, high quality PCB, shameful cooler performance. MSI video recording quality is perfect, just remember to change the slider from the default 75% which will blur everything to 100% and compress it after.

[Lanek]

I didn't know that about the Asic quality, my boost clock is 1215 Mhz like most people are reporting for this card. Good GPU, high quality PCB, shameful cooler performance. MSI video recording quality is perfect, just remember to change the slider from the default 75% which will blur everything to 100% and compress it after.

yep will try it... I own fraps since a long time and the quality is good, but for long recording, it is better to have a lot of place on the HDD you use for recordings.... MSI for recording on the fly should be a bit more practical.

[PatRaceTin]

would AMD bring 7970 price to 670 level (as 6970 vs 570) ?

[Ice009]

Hey guys, been thinking of upgrading from my SLI GTX 580 3GB MSI Lightning Xtreme Edition cards to 2 x GTX 670.

Is it worth it? And how would the performance compare in 2D Surround with a 5760 x 1200 resolution?

Another reason I was thinking of changing over is because of the power usage. Would it make much difference or would the extra I pay for the GTX 670s cards nullify the power savings.

Also, what is the deal with the throttling on the GTX 670s. I've seen it mentioned that they throttle when they reach certain temps? Does that happen on the GTX 680s too?

[Chickenfeed]

The power savings would be offset by the cost of upgrading but if want want to be more environmentally friendly theres that angle...
They will be no doubt be cooler runnner and quieter as well however. You'd be looking at 150-200 watts less at load. Obviously you'd get better performance as well.

When people are refering to throttling thats kind of misleading. Basically the gpus boost functionallity is very similar to the Turbo functionallity in Intel cpus. If there is thermal and tdp headroom, the card can clock higher, if not it will run lower but no lower than its base clock, no different than a cpu. I wouldn't call this throttling whatsoever... It's about increasing efficiency by keeping the card as close to tdp thus as busy as possible.

Given you are using 3GB buffer cards now, I'd try to get a hold of 4GB 670s given your intent to use surround.

[Ice009]

The power savings would be offset by the cost of upgrading but if want want to be more environmentally friendly theres that angle...
They will be no doubt be cooler runnner and quieter as well however. You'd be looking at 150-200 watts less at load. Obviously you'd get better performance as well.

When people are refering to throttling thats kind of misleading. Basically the gpus boost functionallity is very similar to the Turbo functionallity in Intel cpus. If there is thermal and tdp headroom, the card can clock higher, if not it will run lower but no lower than its base clock, no different than a cpu. I wouldn't call this throttling whatsoever... It's about increasing efficiency by keeping the card as close to tdp thus as busy as possible.

Given you are using 3GB buffer cards now, I'd try to get a hold of 4GB 670s given your intent to use surround.

Thanks a lot, appreciate the info, so the power savings would be negated by the cost of upgrading, which is pretty much what I thought. It basically comes down to performance then. I'd prefer the 4GB cards for Surround, but for some reason, in Australia, they seem to jack the prices up a lot higher for the 4GB cards. The 2GB GTX 670s here range from $480-$550. The 4GB cards are probably going to cost $600, which is out of my price range for two of them. If we had prices closer to the US prices I'd have two GTX 680s already.

So it's pretty much between the 670 2GB SLI Vs GTX 580 3GB SLI. What do you think? Not worth it, or would it still be a worthwhile upgrade?

So for the throttling it won't go lower than the base clock even if it hits 80 plus degrees? Thanks for explaining that, I thought it might actually down clock under the base clock. So if I want to OC them and keep them there I need cards with good cooling. What cards would be best for SLI? They'll have a two slot gap between them on my MB.

[bamtan2]

side note: I still havent seen good evidence that the smaller boards are any worse than the bigger boards. I would love to be proven wrong, but it still looks to me like the board isnt a constraint. the cooler on it maybe, but not the board.

Hey guys, been thinking of upgrading from my SLI GTX 580 3GB MSI Lightning Xtreme Edition cards to 2 x GTX 670.

in less than 12 months a gtx 780 will make both of those configurations look poopy

[Ice009]

side note: I still havent seen good evidence that the smaller boards are any worse than the bigger boards. I would love to be proven wrong, but it still looks to me like the board isnt a constraint. the cooler on it maybe, but not the board.



in less than 12 months a gtx 780 will make both of those configurations look poopy

Yeah, sure it will, but I'm not waiting 12 months to upgrade. What do you think I should do now? Stick with the 580s or try 670s? I just wonder if the 2GB of RAM will be a hindrance in Surround Vs my 3GB cards.