[Hardcorp] GALAXY GTX 660 Ti GC OC vs. OC GTX 670 & HD 7950

[hokies83]

$330AR OC'd gtx670 on a gtx680 PCB with Borderlands 2 makes everything else look stupid right now: http://www.newegg.com/Product/Produc...82E16814162107

Fixed.

Fixed Again

 

[RussianSensation]

RS, what about FP16 performance, wasn't Kepler supposed to have and advantage there? That's one area I have little information on. This is a real question--I'm curious.

Not sure if serious. Addressed this at least 3x already in different threads. You keep asking me the same questions and I answered them in those other threads and you keep asking them again and again? Why?

You are still stuck on Kepler vs. GCN and trying to nitpick apart what has been shown already over and over? Both architectures have their positives, already outlined. Take a game with GPU cloth, tessellation, extreme particles, watch for Crysis 3 when GTX680 >>>>> HD7970. I don't design the games. Put in things that Kepler is good at and GCN is going to be slower. Put in things GCN is good at and Kepler is slower. Wait for Borderlands 2, Crysis 3, Medal of Honor Warfighter before writing off Kepler architecture as you seem to want to do. While it is true that GCN is better for compute in games, Kepler has its trump cards as well. It's a matter of time before some games allow Kepler architecture to shine as well. I know you are trying really hard to paint Kepler as a failed architecture this round and validate your recommendation in that thread for an HD7850 over the 670. Sorry, I don't agree. GTX670 OCed keeps up with an HD7950 OCed overall. The main difference is AMD has a price/performance advantage.

If GTX670 drops to $310-320, I wouldn't hesitate for a second to recommend it over the 7950:
- lower power consumption
- similar OCed performance
- can still handle MSAA unlike 660Ti
- free BL2 coupon (Sleeping Dogs still hasn't been added by AMD...)

 

[blastingcap]

Not sure if serious. Addressed this at least 3x already in different threads.

You are still stuck on Kepler vs. GCN and trying to nitpick apart what has been shown already over and over? Both architectures have their positives, already outlined. Take a game with GPU cloth, tessellation, extreme particles, watch for Crysis 3 when GTX680 beat up the HD7970. I don't design the games. Put in things that Kepler is good at and GCN is going to be slower. Put in things GCN is good at and Kepler is toast. Wait for Borderlands 2, Crysis 3 before writing off Kepler architecture.

Actually, you discussed stuff other than FP16 earlier in the thread, but not FP16. I was just curious if you had heard anything different in the last several months. No need for eye rolls, those are bad for your eyes.

Also, you are dead wrong in your edit of your post, specifically: "I know you are trying really hard to paint Kepler as a failed architecture this round and validate your recommendation in that thread for an HD7850 over the 670." Please don't insinuate any such motive on my part. I told you that I was simply curious. If pricing were equal I'd prefer NV cards but they are not, so I've bought AMD for the last few years.

If I seem fixated on that 7850 thread it's because it was one of the last threads I remember prior to my absence from these forums for several months, that's all.

 

[RussianSensation]

Ya, you gotta let that thread go and move on. We just had a different point of view. My main point is both GCN and Kepler have advantages. It looks like GCN is in the 'lead' right now because last couple games that have come out which were demanding have been AMD Gaming Evolved titles: Sniper Elite V2, Sleeping Dogs, Dirt Showdown.

We have to be fair and see what happens when features that benefit from Kepler architecture are used in other games, say Borderlands 2, Crysis 3, etc.

What are your thoughts on how Kepler architecture will adapt for Crysis 3 compared to GCN?
Mine are in post #25: http://forums.anandtech.com/showthread.php?t=2266150

I think GTX680 > 7970 in Crysis 3.

 

[Dark Shroud]

How does it explain an overclocked 7970GHE being slower than a stock card as illustrated by RussianSensation?

Granted the average and max went up but the Min dropped way off.

That test was done on a stock card with the blower fan design. So it's hitting the powertune limit.

 

[AnandThenMan]

I think GTX680 > 7970 in Crysis 3.

TWIMTBP will make sure of that, at least at game launch. But we've seen time again when AMD is given the opportunity to tune in drivers, the tables turn. Plus the 8000 series will be what Kepler is going up against, unsure if Nvidia will have a refresh by then.

 

[SlowSpyder]

AMD's newest drivers seem to have really given the Radeon 79xx's a nice kick in the pants. And overclocked 7950 is a monster at ~$300. Newegg has a few for $299.99 with free shipping, custom cooled cards are a bit more.

The 7950 is priced about the same as the 660Ti, but appears to be able to beat it more often than not.

 

[blastingcap]

I think GTX680 > 7970 in Crysis 3.

Of course. TWIMTBP

 

[SlowSpyder]

RussianSensation,
can you stop your marketing posting sessions, pls?

Repeating this "directcompute" stuff is ridiculous. In none of the 3 games GK104 shows a "directCompute" problem.

AMD's problem is some kind of Amdahl's law: A too narrow front-end for too many processing units. That's why you seeing stuff like downsampling (ogssaa) and forward+ (brute force rendering technique for more light sources) in more and more gaming evolved titles.

The only problem GK104 has is that the chip has not enough compute performance.

RS is a straight shooter. I've been reading his posts for years, he certainly calls it like he sees it, and I've seen him praise Nvidia in the past too. I don't think any other poster here posts as many informative, information/benchmark backed posts than RS.

 

[tviceman]

TWIMTBP will make sure of that, at least at game launch. But we've seen time again when AMD is given the opportunity to tune in drivers, the tables turn. Plus the 8000 series will be what Kepler is going up against, unsure if Nvidia will have a refresh by then.

GK110 wil be out. Nvidia refreshed each Fermi chip 7-8 months after initial releases. Kepler will have no issues getting refreshed in time to compete with AMD's next line up. I think I speak for most people.here when I say I'd be shocked if AMD will be able to stay competitive with fully functional GK110 GeForce card. The shader count difference between GK110 and GK104 is significantly larger than GF110 and GF104. It's just a question of when Nvidia will be able to deliver fully operational GK110 GeForce cards.

 

[hokies83]

Only reason Amd is doing so well is Nvidias lack of drivers for the 600 series.. when there released the tables will turn..

Meh im down a Gpu.. and have a G1 Sniper 3 so im gimped atm... But when i was running SLI i have the 20th Highest SLI / Xfire 3d mark score on OCN... i expect to move up to the top 10 when i get my card back and do some tweaks..

 

[VulgarDisplay]

Only reason Amd is doing so well is Nvidias lack of drivers for the 600 series.. when there released the tables will turn..

Just like AMD was doing so bad and the worst company because they didn't have drivers for the 7000 series until 12.7 beta.

I'm kinda interested to see if AMD's new driver and developer relation campaign is going to turn the tables in the GPU landscape. I'm really loving the things they are doing lately, and the changers in the leadership at AMD seem to be the "catalyst" (pun intended) that are moving things forward for AMD's graphics division.

I truly hope they can keep it up and win some mindshare as well as marketshare from gamers out there. If they compete on more levels with nvidia we all reap the benefits.

 

[blackened23]

Only reason Amd is doing so well is Nvidias lack of drivers for the 600 series.. when there released the tables will turn..

Meh im down a Gpu.. and have a G1 Sniper 3 so im gimped atm... But when i was running SLI i have the 20th Highest SLI / Xfire 3d mark score on OCN... i expect to move up to the top 10 when i get my card back and do some tweaks..

A lot of existing titles are heavily optimized already - so I wouldn't expect miracles, although i'm sure they can tweak performance somewhat. I'm not disagreeing that drivers don't make a difference, nvidia can definitely make some headway on some recent titles (such as The Secret World and Guild Wars 2) for example. But for older titles like crysis 2, metro 2033 they're mostly optimized already.

Main thing I'd like to see in the next round of drivers is some better SLI optimization in The Secret World, perhaps SLI optimizations in Sleeping Dogs with better SSAA performance. That would be asking a lot though since SSAA makes any GPU chug.

 

[blackened23]

Just like AMD was doing so bad and the worst company because they didn't have drivers for the 7000 series until 12.7 beta.

I'm kinda interested to see if AMD's new driver and developer relation campaign is going to turn the tables in the GPU landscape. I'm really loving the things they are doing lately, and the changers in the leadership at AMD seem to be the "catalyst" (pun intended) that are moving things forward for AMD's graphics division.

I truly hope they can keep it up and win some mindshare as well as marketshare from gamers out there. If they compete on more levels with nvidia we all reap the benefits.

I'd say nvidia has the upper hand here for now, although AMD seems to be trying to change that. Right now nvidia has a lockdown on unreal engine games for the most part while AMD has optimizations for all Square Enix games. And then there's borderlands 2 which I presume will be a huge title.

Square Enix games as of lately have been absolutely great (DX:HR, Sleeping Dogs, etc) but they haven't caught on quite as much as say, Batman: AC. That bet may pay of for AMD because Square has produced a ton of quality games lately, although nvidia as well seem to be behind a lot of great titles as well (they're very good at hedging their bets on which games will catch on)

 

[AnandThenMan]

GK110 wil be out.

Source?

Sea Islands taped out quite some time ago, we could see the first version as early as December.

 

[RussianSensation]

RS is a straight shooter. I've been reading his posts for years, he certainly calls it like he sees it, and I've seen him praise Nvidia in the past too. I don't think any other poster here posts as many informative, information/benchmark backed posts than RS.

Thanks for the support SlowSpyder! I try to stay neutral for both brands. I am not always right, but I try to contribute where I can. For example, when AMD cards had driver issues in Guild Wars 2 Beta, I put up some info on that for better or worse; and then other posters later provided updates that the performance was fixed. When Fermi was much faster in tessellated scenarios last generation, such as in Crysis 2, I made sure to note that for prospective buyers. It's just sharing of information, with no monetary involvement if AMD or NV sell more cards based on what I say. People have also mentioned that I promote bitcoin mining to sell more AMD cards, but what I am promoting with that functionality is an option for gamers to upgrade to a faster modern GPU and maybe try and save some $ to ease the cost of gaming with newer hardware. There is no conspiracy theory.

Meh im down a Gpu.. and have a G1 Sniper 3 so im gimped atm... But when i was running SLI i have the 20th Highest SLI / Xfire 3d mark score on OCN... i expect to move up to the top 10 when i get my card back and do some tweaks..

I can understand an appeal about having a top score in a popular benchmark that is shared by enthusiasts on a forum. However, 3DMark11 is not a game. Thus, unless its performance is reflective of how videocards perform in real world games, it's a synthetic benchmark and at best a showcase of some possible graphical features that may or may not be implemented in future titles. For example, at Xbitlabs, the 1324mhz GPU boosted MSI GTX680's Lightning scores 4094 in 3dMark11, or a commendable 16% lead over the 1250mhz HD7970, and yet the Lightning loses in games overall at high resolutions in the same review.

At TechPowerup, at 2560x1600 HD7970 Toxic and GTX680 are tied at 8.2 fps in 3dMark11, with Toxic at 1200mhz Lethal Boost only having a 6% advantage (8.7 vs. 8.2 fps) . In 18 games, HD7970 Toxic is actually 12% faster, while the Lethal Boost is 19% faster than a GTX680 at 2560x1600.

GTX680 vs. GTX580 in 3dMark11 at 2560x1600 = + 58% (8.2 vs. 5.2 fps)
GTX680 vs. GTX580 in 18 games at 2560x1600 = + 31% (89% vs. 68%)

What do all of these examples tell me about 3dMark11? It's not a very accurate predictor of how videocards will actually perform in modern video games. All it's telling me is how videocards will perform specifically running 3dMark11 test suites. Furthermore, while you say that NV can simply improve drivers, which I do not deny that they can, it can also be true that NV optimized highly for 3dMark11 and that the performance delta between the 680 and 7970 in 3dMark11 could actually be a function of NV's better driver optimization specifically for 3dMark11, or perhaps 1 of the test suites in 3dMark11 is using some graphical feature that runs much faster on the GTX680 (but I haven't researched the test suites in detail to confirm if say they are very heavy on tessellation). It could actually be the case that it is 3dMark11 that is the outlier, not the real world gaming performance. I am actually much more inclined to believe that 3dMark11 is not representative of GTX680's performance, but the best case for a 680.

 

[hokies83]

Thanks for the support SlowSpyder! I try to stay neutral for both brands. I am not always right, but I try to contribute where I can. For example, when AMD cards had driver issues in Guild Wars 2 Beta, I put up some info on that for better or worse; and then other posters later provided updates that the performance was fixed. When Fermi was much faster in tessellated scenarios last generation, such as in Crysis 2, I made sure to note that for prospective buyers. It's just sharing of information, with no monetary involvement if AMD or NV sell more cards based on what I say. People have also mentioned that I promote bitcoin mining to sell more AMD cards, but what I am promoting with that functionality is an option for gamers to upgrade to a faster modern GPU and maybe try and save some $ to ease the cost of gaming with newer hardware. There is no conspiracy theory.



I can understand an appeal about having a top score in a popular benchmark that is shared by enthusiasts on a forum. However, 3DMark11 is not a game. Thus, unless its performance is reflective of how videocards perform in real world games, it's a synthetic benchmark and at best a showcase of some possible graphical features that may or may not be implemented in future titles. For example, at Xbitlabs, the 1324mhz GPU boosted MSI GTX680's Lightning scores 4094 in 3dMark11, or a commendable 16% lead over the 1250mhz HD7970, and yet the Lightning loses in games overall at high resolutions in the same review.

At TechPowerup, at 2560x1600 HD7970 Toxic and GTX680 are tied at 8.2 fps in 3dMark11, with Toxic at 1200mhz Lethal Boost only having a 6% advantage (8.7 vs. 8.2 fps) . In 18 games, HD7970 Toxic is actually 12% faster, while the Lethal Boost is 19% faster than a GTX680 at 2560x1600.

GTX680 vs. GTX580 in 3dMark11 at 2560x1600 = + 58% (8.2 vs. 5.2 fps)
GTX680 vs. GTX580 in 18 games at 2560x1600 = + 31% (89% vs. 68%)

What do all of these examples tell me about 3dMark11? It's not a very accurate predictor of how videocards will actually perform in modern video games. All it's telling me is how videocards will perform specifically running 3dMark11 test suites. Furthermore, while you say that NV can simply improve drivers, which I do not deny that they can, it can also be true that NV optimized highly for 3dMark11 and that the performance delta between the 680 and 7970 in 3dMark11 could actually be a function of NV's better driver optimization specifically for 3dMark11, or perhaps 1 of the test suites in 3dMark11 is using some graphical feature that runs much faster on the GTX680 (but I haven't researched the test suites in detail to confirm if say they are very heavy on tessellation). It could actually be the case that it is 3dMark11 that is the outlier, not the real world gaming performance. I am actually much more inclined to believe that 3dMark11 is not representative of GTX680's performance, but the best case for a 680.

I do not think the 680 is any better for 3dmark then a 7970.. the highest 7970 on ocn is #12 on the score board..

But in the Quad fire vs Quad SLI Quad fire is on top followed by 2x 690s.

 

[RussianSensation]

I do not think the 680 is any better for 3dmark then a 7970.. the highest 7970 on ocn is #12 on the score board..

But in the Quad fire vs Quad SLI Quad fire is on top followed by 2x 690s.

I think you missed the point of what I am saying. Someone might have scored high in 3dMark11 on a water-cooled 7970 that reached 1350-1400mhz. 3dMark11 is so off the mark for actually comparing GPUs for games, that it's basically worthless. GTX680 has a huge lead in 3dMark11 over a stock 580 and beats a 7970 easily but that % lead does not actually translate to real games. In other words 3dMark11 tells us nothing about how say GTX580/680/7970 videocards stack up against each other in games. In 3dMark11 a 680 is 58% faster than a 580, so does that mean the 680 is 58% faster on average in DX11 games than a 580 is? Of course not!

At TechSpot, it took a 1200mhz 7970 to barely beat a reference 680. In real world games, a 1200mhz 7970 = Sapphire TOXIC and it was 19% faster than a 680.

This is why I still don't understand why people even look at 3dMark11 scores. They are meaningless to compare videocards. The only purpose they serve is stress test your GPU, see possible cool effects that haven't yet been used in games and e-peen. The current 3dMark11 version is probably the worst in recent years since it's really inaccurate for trying to extrapolate AMD vs. NV gaming performance in actual games.

 

[SlowSpyder]

Only reason Amd is doing so well is Nvidias lack of drivers for the 600 series.. when there released the tables will turn..

Meh im down a Gpu.. and have a G1 Sniper 3 so im gimped atm... But when i was running SLI i have the 20th Highest SLI / Xfire 3d mark score on OCN... i expect to move up to the top 10 when i get my card back and do some tweaks..

Do you have a source to back that up, that Nvidia's drivers are holding back 600 series performance? I know that, generally speaking, over time drivers improve performance. But do we know that Nvidia has a big performance bumping driver in the works? How do we know that Nvidia's drivers aren't already allowing the 600 cards to perform about as well as the architecture can be reasonally expected to? I'm not saying it cannot happen, but we can't say a big performance jump unleashed by drivers will happen either.

When the 7900 cards launched, the drivers were pretty unpolished. I know, I was an early adopter. The normally fantastic (for me) AMD drivers were not nearly as good as the drivers for my 5870's. You could see that there were a lot of areas that needed improvement from a functionality standpoint (disappearing cursor, won't wake from sleep, etc.) So it was reasonable to expect that the performance of the cards was just as crippled due to the early drivers. I think Nvidia's card launched with much better drivers than the 79xx cards, so who knows how much more can be tweaked out.

 

[blackened23]

I think you missed the point of what I am saying. Someone might have scored high in 3dMark11 on a water-cooled 7970 that reached 1350-1400mhz. 3dMark11 is so off the mark for actually comparing GPUs for games, that it's basically worthless. GTX680 has a huge lead in 3dMark11 over a stock 580 and beats a 7970 easily but that % lead does not actually translate to real games. In other words 3dMark11 tells us nothing about how say GTX580/680/7970 videocards stack up against each other in games. In 3dMark11 a 680 is 58% faster than a 580, so does that mean the 680 is 58% faster on average in DX11 games than a 580 is? Of course not!

At TechSpot, it took a 1200mhz 7970 to barely beat a reference 680. In real world games, a 1200mhz 7970 = Sapphire TOXIC and it was 19% faster than a 680.

This is why I still don't understand why people even look at 3dMark11 scores. They are meaningless to compare videocards. The only purpose they serve is stress test your GPU, see possible cool effects that haven't yet been used in games and e-peen. The current 3dMark11 version is probably the worst in recent years since it's really inaccurate for trying to extrapolate AMD vs. NV gaming performance in actual games.

Pretty much. 3dmark11 does not match real world performance, and nvidia has a heavy lead in 3dmark11 generally speaking - while AMD does better in Heaven.

There are people obsessed with benchmarking synthetics and thats a sub-culture that i'll never understand. Don't really see the point of overclocking your system to insanity to get 200 more points in 3dmark11, its ridiculous. And doesn't help real games performance.

 

[Olikan]

When the 7900 cards launched, the drivers were pretty unpolished. I know, I was an early adopter. The normally fantastic (for me) AMD drivers were not nearly as good as the drivers for my 5870's. You could see that there were a lot of areas that needed improvement from a functionality standpoint (disappearing cursor, won't wake from sleep, etc.) So it was reasonable to expect that the performance of the cards was just as crippled due to the early drivers. I think Nvidia's card launched with much better drivers than the 79xx cards, so who knows how much more can be tweaked out.

i still think that amd drivers are unpolished...

heck!...we have yet to see VCE in action from gpus (not cpu based on trinity)

 

[SlowSpyder]

Pretty much. 3dmark11 does not match real world performance, and nvidia has a heavy lead in 3dmark11 generally speaking - while AMD does better in Heaven.

There are people obsessed with benchmarking synthetics and thats a sub-culture that i'll never understand. Don't really see the point of overclocking your system to insanity to get 200 more points in 3dmark11, its ridiculous. And doesn't help real games performance.

I get the chasing a benchmark score thing, I can see how tweaking and playing with your hardware to achieve a certain score can be as fun as any other hobby with computers. But, as RS said, we really cannot use synthetic benches to try and prove anything as far as gaming goes.

RS did a nice job pointing out the 680 vs. 580 scores. If the 3DMark scores were leaked early for the 680 before it launched, and we went by that alone, we would be lead to believe that the 680 is going to be almost 60% faster than the 580. Obviously that's not the case.

 

[SirPauly]

If they compete on more levels with nvidia we all reap the benefits.

Imho,

Indeed! Ideally, have a desire for both IHV's rolling up their sleeves, working hard to try to give reasons why the PC gaming platform as a whole is compelling. With heated competition on multiple fronts, hopefully dollar/product value, innovation, immersion and gaming experiences may be improved over-all.

nVidia and AMD may have to offer compelling reasons why a potential or prospective gamer should game on the PC platform and compete with each other, too.

 

[RussianSensation]

Of course. TWIMTBP

There are titles where AMD does better despite them being TWIMTBP. The main reason I think is that the game is looking to use all the core technologies of Kepler, specifically the graphical techniques/features where Kepler is faster than GCN. As others have said though by Feb 2013 we may already have HD8000 series and maybe GTX700 series too. GPU wars incoming, can't wait!

 

[RussianSensation]

i still think that amd drivers are unpolished...

heck!...we have yet to see VCE in action from gpus (not cpu based on trinity)

VCE has already been tested and it works on desktop HD7000 series GPUs just as well as it does on GTX600 series for video transcoding - which is to say terrible, compared to native CPU for rendering quality or in regard to speed compared to QuickSync.

QuickSync = speed
CPU = quality
AMD/NV transcoding = very poor for speed and quality; in other words worthless in their current state.