GTX 670 - today's the big day, right?

[MichaelD]

Doesn't the NDA expire today? I'm surprised there's nothing on the AT Home Page about it. :\ I know there's some big conference where NVidia's supposed to be making "a big announcement". I built my new system two days ago. It needs a new GPU!

 

[imaheadcase]

You are blind. Its on the front page, its about 9 hours from now.

 

[MichaelD]

And you are rude. If you're talking about the article from 25 Apr that is halfway down the page, I saw that 3 days ago. I'm looking for new info on pricing and availability.

 

[notty22]

Reference NVIDIA GeForce GTX 670 Graphics Card Pictured
Here is the first picture of an NVIDIA reference design GeForce GTX 670. For the most part, its design will resemble that of reference design GTX 680. The card pictured below is said to have had Leadtek branding, which later got removed. GeForce GTX 670 is carved out of 28 nm GK104 silicon by disabling one of the card's eight SMX units, resulting in a CUDA core count of 1344. Its clock speeds are lower than those of the GTX 680, with around 950 MHz core base clock, and core boost clock a little over 1 GHz. It features 2 GB of memory across a 256-bit wide GDDR5 memory interface. The latest price speculation puts it at US $399-$429.

Picture from different link, mentioned in the above article.

 

[toyota]

http://videocardz.com/32476/geforce-gtx-660-ti-and-gtx-670-specification-leaked


probably just a rumor but the 670 looks exactly like what I thought it would. the 660ti seems odd though. it makes sense to be a cut down gk104 but why increase the clocks on it over the 670? 660ti sli would probably slaughter the gtx680 for the same price so might be the new killer setup for 1920x1080/1200. 1.5gb of vram would hold the full potential of that setup back in a few rare cases especially at 2560 though.

 

[Subyman]

That pic is still going around? Look how crappy it looks, no way nvidia media looks like that.

 

[skipsneeky2]

The nda is lifted tonight right?

Looking forward to the results of the 660ti which i hope can perform as well as a gtx570,its price is in the current gtx560 ti range,if not more,so i am expecting great things from it,as with the gtx670,lets hope it can handly beat a gtx580 as in retail its gonna be as much as a gtx580,if not more if gouging is involved.

 

[Skurge]

http://videocardz.com/32476/geforce-gtx-660-ti-and-gtx-670-specification-leaked


probably just a rumor but the 670 looks exactly like what I thought it would. the 660ti seems odd though. it makes sense to be a cut down gk104 but why increase the clocks on it over the 670? 660ti sli would probably slaughter the gtx680 for the same price so might be the new killer setup for 1920x1080/1200. 1.5gb of vram would hold the full potential of that setup back in a few rare cases especially at 2560 though.

Looks like the 660ti will be competing with the 7850, but with the cut down memory bus and cut down rops it may be slower, but it's clocked pretty high.

 

[aaksheytalwar]

660ti will probably be $300-350 so 660 sli would probably complete with 680 custom

 

[toyota]

660ti will probably be $300-350 so 660 sli would probably complete with 680 custom

no way will 660ti cost that much.

 

[BD231]

Ugly if true. There'd be no point in buying a 670 over a 680 with such a minor price differential, and if performance speculations are true you can get the same performance for half the cost.

 

[aaksheytalwar]

680 $500 670 $400 660ti $300 660 $250 650 ti $150 and so on. This is pretty obvious IMO considering past trends as well. 670 will be like 15-20% faster than 660ti at most, the difference won't be $400 to $250 for that

 

[DigDog]

$420 . say what. it actually makes sense to go 680.

 

[BD231]

0/10

 

[RussianSensation]

0/10

Really? 0/10?

$399 for GTX670 that's going to have just 1 cluster turned off is a fail? Sure, it may be but that's if you think GTX680 is itself a GTX670 Ti that should have been $399. But that's an argument for another day. Otherwise, GTX670 at $399 will put the breaks on most 7950s.

In some games, HD7950 is pathetic. It needs 30% overclock just to be relevant at $400. I wouldn't be surprised if GTX670 beats HD7970 in some games.

This is better than GTX570 vs. GTX580. The 670 has same memory bandwidth, same amount of VRAM, just 1 SMX cluster turned off and slightly less texture bandwidth (112 vs. 128). In comparison the HD7950 is neutered a lot more vs. HD7970.

$420 . say what. it actually makes sense to go 680.

GTX670 = $399 MSRP
GTX660 Ti is rumored to be $249.

 

[tviceman]

I would really like to see Nvidia *NOT* disable any memory controllers for the gtx660ti. 2 gigs of VRAM has pretty much become the standard. Lower speed ram with all the memory controllers enabled would provide better performance and future proofing, IMO. If Nvidia has a noticeable amount of GK104 chips with defects in one of the memory controllers, I think they would be better off making a gtx660 @ 192-bit, and gtx660ti @ 256-bit.

Also, the price gap between $400 and $250 is awfully large when taking into consideration neither of these two products are halo status. Something still needs to fit in between there at ~$320.

 

[DeeJayeS]

Really? 0/10?


In some games, HD7950 is pathetic.

What other games besides BF3?

 

[MichaelD]

Lots of great info! Thank you, everyone. I haven't been keeping up with the tech behind the cards for a few years now (shader counts, ROPs and that stuff) so I can't speak technically at most of your levels.

$400 (estimated for now...we'll see what it really turns out to be) for a 670? Hmm, is the $100 more for the 680 worth the diff? And based on the chart that RussianSensation put up, you gotta wonder about the 660Ti in SLI.

Bottom line is that I'm excited about this release. I feel I'll have a new GPU setup in about another month...hopefully sooner. Good times.

 

[blastingcap]

Please stop comparing averages. It's much better to compare time series graphs and keeping track of MINIMUM framerates. This goes double now that NVIDIA has a GPUboosting feature that may increase average framerate without doing anything for minimums. Fps vs. time charts at the resolution you are interested in, are what should matter.

 

[RussianSensation]

What other games besides BF3?

There are plenty of other games such as Witcher 2. Reviews are all over the web. GTX680 is ~ 30% faster than 7950. If GTX670 only has 1 cluster disabled, it stands to reason that 7950 will lose in a ton of games:

- Crysis 2
- Batman AC
- SKYRIM
- Dirt 3
- Old Republic
- WOW
- Dragon Age 2
etc.

Luckily for 7950, it has healthy overclocking headroom.

Please stop comparing averages. It's much better to compare time series graphs and keeping track of MINIMUM framerates. This goes double now that NVIDIA has a GPUboosting feature that may increase average framerate without doing anything for minimums. Fps vs. time charts at the resolution you are interested in, are what should matter.

Thankfully Hardware Canucks, Computerbase.de, GameGPU.ru, and other websites actually show recorded segments of the games. So you can see exactly where the game was tested. Also, while minimum frames are important, in some games such as Metro 2033 sequence, the minimums will be inherently low because of how that benchmark is made. At the same time, if you actually compare time series graphs for GTX680, they don't show any unusual maximum framerate spikes vs. HD7970 or serious problems with minimum framerates either. In other words, we can still use averages since GTX680 doesn't seem to exhibit unusual behaviour that would constitute "cheating" avg. FPS by elevating maximums but keeping minimums low. That's unsubstantiated assumption. Of course I agree that time series are helpful, but we shouldn't outright dismiss 98% of reviews out there because they don't have them.

There is nothing magical about GTX680's Turbo Boost. Average numbers coincide with time series graphs in the same review.

Based on that, it's hard to imagine how a stock HD7950 will be competitive with a $399 GTX670.

 

[RussianSensation]

And based on the chart that RussianSensation put up, you gotta wonder about the 660Ti in SLI.

I wouldn't recommend it. $500 setup with 1.5GB of VRAM in 2012 isn't great. While most games today don't yet hit the 1.5GB buffer, I imagine someone like you spending $ on a card will keep it for longer than 1 year? I think HD7850 2GB OC in CF would be better, even if slightly slower. It remains to be seen how GTX660 does since HD7850 is a pretty good overclocker. Also, personally I'd give up 20-30% performance just to have a single GPU.

 

[Don Karnage]

If the 660 Ti gets 1.5GB it'll be a huge fail. Need 2GB Min these days

 

[blastingcap]

There are plenty of other games such as Witcher 2. Reviews are all over the web. GTX680 is ~ 30% faster than 7950. If GTX670 only has 1 cluster disabled, it stands to reason that 7950 will lose in a ton of games:

- Crysis 2
- Batman AC
- SKYRIM
- Dirt 3
- Old Republic
- WOW
- Dragon Age 2
etc.

Luckily for 7950, it has healthy overclocking headroom.




Thankfully Hardware Canucks, Computerbase.de, GameGPU.ru, and other websites actually show recorded segments of the games. So you can see exactly where the game was tested. Also, while minimum frames are important, in some games such as Metro 2033 sequence, the minimums will be inherently low because of how that benchmark is made. At the same time, if you actually compare time series graphs for GTX680, they don't show any unusual maximum framerate spikes vs. HD7970 or serious problems with minimum framerates either. In other words, we can still use averages since GTX680 doesn't seem to exhibit unusual behaviour that would constitute "cheating" avg. FPS by elevating maximums but keeping minimums low. That's unsubstantiated assumption.

There is nothing magical about GTX680's Turbo Boost. Average numbers coincide with time series graphs in the same review.

Based on that, it's hard to imagine how a stock HD7950 will be competitive with a $399 GTX670.

I'm not assuming anything, simply pointing out what gamers already know: you feel downward fps spikes a lot more than you benefit from fps spikes upwards.

In fact it is you who are doing the assuming--that the gtx 680's gpu boost will not unduly affect its averages more than its minimums.

That said, I haven't seen anything that blatantly disproves the assumption; I think in most cases the gtx 680 average is probably no better or worse a measure than any other card without GPU boost. But why even take the risk? It is simply cleaner to use minimums and not even have to worry about such things.

The gtx 670, which is clocked higher and closer to its ceiling, can and should win in most games vs the 7950. (It will probably lose in some games like AvP.) When both are oc'd to the max, I expect it to be a draw in performance. The gtx 670 should still be more efficient, have more features, and better SLI support.

Competition is good and helps consumers by keeping a lid on prices. The enemy of PC gamers is not NVIDIA or AMD so let's stop these red vs green wars. Our common enemy: console gamers.

 

[skipsneeky2]

If the 660 Ti gets 1.5GB it'll be a huge fail. Need 2GB Min these days

If its a gtx570 replacement card,i doubt by the time the card uses 1.5gb in such titles that it would deliver good fps....you would prob end up knocking down settings dropping vram down.

When i had my 7970,i rarely saw 1300mb usage with playable settings in BF3 and we know for a fact the gtx660 ti won't even touch the 7970.

I feel its vram amount is justifiable if it delivers up to gtx570 performance basically giving you a extra 250mb of vram for $50 less then you would get with a $300 gtx570,if anything it keeps the price of the card lower.

 

[realjetavenger]

<snip>

Competition is good and helps consumers by keeping a lid on prices. The enemy of PC gamers is not NVIDIA or AMD so let's stop these red vs green wars. Our common enemy: console gamers.

I'll drink to that.
Oh wait, too late already (hiccup) did