OFFICIAL KEPLER "GTX680" Reviews

[Despoiler]

Here come the excuses.

680 is 50 dollars cheaper yet still beats down the 7970. Where's my Violin....

I'm pretty sure you don't know what excuse means. I made a valid point. No one is trying to compare clock per clock. That is the only way to see who's architecture is better. Price has nothing to do with that. You might want to read the rest of my original comment. Now that we have competition (Nvidia is late again and has been late since AMD's 5k series) the prices will come down on AMD's side also.

Yes, and that matters because the underlying architecture, and therefore performance-per-clock, is identical!!!

See above.

 

[zebrax2]

Impressive i must say. Seems like NVidia has the performance/mm2 crown this generation

It's interesting how NVidia went for a little bit less compute performance in exchange for huge die savings.

 

[MrTeal]

GPU boost is a great idea, just like Turbo Boost from Intel. It does worry me a little bit in the launch reviews though, because of the following.

The second number is what NVIDIA calls the boost clock, and this one is far more nebulous, as it relates to the operation of GPU Boost itself. With GPU Boost NVIDIA does not have an explicit top clock; they’re letting chip quality play a significant role in GPU Boost. Because GPU Boost is based around power consumption and temperatures, higher quality GPUs that operate with lower power consumption can boost higher than lower quality GPUs with higher power consumption. In essence the quality of the chip determines its boost limit under normal circumstances.

Everyone knows that overclocking is YMMV. You can look at many review sites and try to get a feel for the limits, but it's still hit and miss. What boost for the 680 means though is that now STOCK performance is variable. If NVIDIA binned cards to get the absolute best ones out to reviewers, even with all other components being the same the card someone buys off the shelf might perform worse than a stock review card because the review card boosted to 1102MHz and the user's card only reached 1030 or 1050.

Ultimately it's a good thing for consumers, but review sites will have to be more diligent and give more information on what the clocks actually are, much like they would in an OC review.

 

[JTsyo]

Well, according to the NV team, this WAS the 660Ti We can probably assume the real 660Ti would fall somewhere in the 7800 series performance range.

I might pick one of these up, would like to see a head to head 1200+ core review and go from there.

Well whatever it's being called. The price is what matters to me. Whatever they have for around $200 will be what I get, might move up a bit if the performance is there.

 

[RussianSensation]

Gonna see where prices settle down because price/perf obviously has to change, especially on the AMD side where they overpriced everything.

Look at HD7950 vs. GTX680. $50 difference, staggering performance delta with 3x 1080P screens. It's making HD7950 look like a last generation videocard.

BitTech Review

I know you run 5760x1080 Eyefinity so you'll like this:

Looks like the 2GB VRAM has once again been overblown.

In Battlefield 3, GTX680 beats HD7950 by 38% in 1080P and by 29% at 1600P.

Looks like HD7950 and HD7970 each need at $100 price cut.

 

[frostedflakes]

Their 7970 overclocked unusually bad, wonder how it would looked like vs tahiti at 1250.

The OC comparisons are at stock volts I think, 7970s don't do 1.25GHz with stock voltage. According to TPU the 680 driver limits voltage to 1.175V for the Core Boost. That appears to be about the same as the stock 3D voltage of the 7970 TPU reviewed, and with that voltage they were able to hit 1075MHz core (16% OC).

http://www.techpowerup.com/reviews/AMD/HD_7970/31.html

What voltage is needed to hit 1.2GHz+ with a 7970? 1.25V? Would be interesting to see how far the 680 could OC if you feed it more voltage. I expected it to have similar OCing headroom to the 7970 (meaning if you OC'd both to their max stable speed, 1.2GHz or so, performance gap between them would decrease since 680 achieves the performance it does by shipping with a higher stock clock). The TPU review suggests 680 tends to OC a bit better at the same voltage, though, so 680 might have a bit more headroom with increased voltage.

 

[-Slacker-]

The GPU boost seems like a great alternative to manual overclocking, reducing the average power draw of the card, heat, noise and potentially increasing the life span of the chip.

I just think it's a bit weird that nobody is willing to tackle this issue form a benching perspective - even though it's a more efficient form of overclocking, it is still overclocking, and the card is being pitted against a competitor with actual stock clocks. The guru3d article shows how they recorded the gtx680 going as high as 1124 Mhz, up form it's 1006 Mhz base clock.

That's a 12% overclock, even though it's not there in every game. I think that would translate into a 8~10% increase in performance.

 

[Olikan]

Looks like the 2GB VRAM has once again been overblown.

finnaly!!! ...this talk about GB is over

 

[blackened23]

Hmm. From that same Bit-tech review, the overclock scaling (beyond stock gpu boosts) doesn't seem very good compared to tahiti. Hopefully its just a glitch with new overclocking software or drivers.

 

[blastingcap]

The GPU boost seems like a great alternative to manual overclocking, reducing the average power draw of the card, heat, noise and potentially increasing the life span of the chip.

I just think it's a bit weird that nobody is willing to tackle this issue form a benching perspective - even though it's a more efficient form of overclocking, it is still overclocking, and the card is being pitted against a competitor with actual stock clocks. The guru3d article shows how they recorded the gtx680 going as high as 1124 Mhz, up form it's 1006 Mhz base clock.

That's a 12% overclock, even though it's not there in every game. I think that would translate into a 8~10% increase in performance.

Yeah I am trying to wrap my head around it but it does appear to be the case that GPU Boost, which you can't turn off, is basically overclocking the GTX 680 so that comparing it at "GPU-boosted stock" to a stock 7970 is not entirely fair. I would like to see a lot of OC vs OC tests to determine what the actual difference is between the cards. I still expect 680 to win, but not by as much as the "GPU boosted stock" vs. real stock differentials.

 

[railven]

Sweet!

TigerDirect.com's Price Protection Policy:

If TigerDirect.com reduces it’s regularly advertised price, or if you find a lower advertised price from a Qualifying Competitor, during the advertised return period offered on an item you purchased from TigerDirect.com; simply contact us and TigerDirect.com will issue you a TigerDirect.com credit in the amount of the price difference good toward the future purchase of products sold by TigerDirect.com. This price protection policy only applies to regularly advertised products.

Got a quick response from a rep, if AMD reduces the price of HD 7970 within the return period (ala TD's own policy - which is 30 days) you may qualify for a price adjustment!

Woots woots! Now if AMD responses soon, I can cash in on whatever the difference is and maybe upgrade to an SSD sooner than later!!!

Competition - I love it!

 

[Zebo]

GPU boost is a great idea, just like Turbo Boost from Intel. It does worry me a little bit in the launch reviews though, because of the following.


Everyone knows that overclocking is YMMV. You can look at many review sites and try to get a feel for the limits, but it's still hit and miss. What boost for the 680 means though is that now STOCK performance is variable. If NVIDIA binned cards to get the absolute best ones out to reviewers, even with all other components being the same the card someone buys off the shelf might perform worse than a stock review card because the review card boosted to 1102MHz and the user's card only reached 1030 or 1050.

Ultimately it's a good thing for consumers, but review sites will have to be more diligent and give more information on what the clocks actually are, much like they would in an OC review.

It's not good for OCers like Turbo boost is no good due to instabilities that result. There are already guys running 1800Mhz with these cards on LN2 and from reading TPU that wouldn't be possible without disabling power limiter and temp limiter nvidia has in dynamic overclocking therefore disabling has been done - I don't read Chinese because that's where i saw these high speeds but fix will come to xtreme soon enough. This is a non issue really. For joe six pack he's safe OCing now and that saves vendors money. For ppl that dont mind voiding warranty bios moddng will have to be done for xtreme performance

 

[RussianSensation]

I just think it's a bit weird that nobody is willing to tackle this issue form a benching perspective - even though it's a more efficient form of overclocking, it is still overclocking, and the card is being pitted against a competitor with actual stock clocks.

That's how the consumer gets the product out of the box. The reviewers are doing exactly what should be done: Test a GPU the way most consumers are going to be using them. Then some are adding a section at the back with manual overclocking which allows HD7970 to catch up. That's fair. Stock GPU Boost is at most 1110Mhz. Three sliders allow you to increase the card’s power draw above its stated 195W by up to 30% for a total of 253W.

It looks like a 1200-1250Mhz HD7970 will actually beat GTX680. However, that automatically rules out all reference HD7970s though if you actually want good noise levels. Most non-reference HD7970s have a price premium right now. Also, we've seen some HD7970's fail to reach 1200mhz despite having premium components. Either way, you aren't going to be gaining much here since since even an 1125mhz HD7970 couldn't take out GTX680 in BF3. That pretty much means you'd need 1175-1200mhz on HD7970.

Hmm. From that same Bit-tech review, the overclock scaling (beyond stock gpu boosts) doesn't seem very good compared to tahiti. Hopefully its just a glitch with new overclocking software or drivers.

The scaling is small because the overclock is only 100MHz! The stock card already runs up to 1110MHz and with their overclock at most it could do was 1,210MHz. Basically if you have an HD7970 that can hit 1200MHz with ease, it's going to be pretty much similar performance on average, faster in some games, slower in others compared to an 1200MHz GTX680 it seems. For people who have overclocked their HD7970s by 30-35%, obviously this isn't an upgrade.

However, compared to HD7950 this is insane. $50 more expensive and easily 30-40% more performance out of the box. HD7970 needs to be overclocked to just match a GTX680. I mean it's like taking an FX8120 and overclocking it to match a stock 2600K and then you and up with a card consuming 100W more power with same performance.

 

[blastingcap]

This is just one sample vs one sample one one game and doubtlessly there are variations in hardware so results vary, but:

http://www.hardwarecanucks.com/foru...s/49646-amd-radeon-hd-7970-3gb-review-24.html

http://www.hardwarecanucks.com/foru...616-nvidia-geforce-gtx-680-2gb-review-27.html

GPU-boosted-680 vs stock 7970: +18.4% average

GPU-boosted-680 vs OC'd 7970: +16.7% average

(Edit: I am looking only at BF3 results as I need to get going soon, others can crunch the math further.)

The minimums are what matter more but they are inconsistent even within the same reviewer (25 vs 28 min. on the stock 7970 for instance) so I'm not sure what to make of it.

Need to see more results, maybe we can start a thread to compare OC vs OC?

 

[Silverforce11]

Actually the overall average at 1600p is 3% and 9% at 1080p vs a STOCK 7970. We all know how well they OC.

That's a "stock" gtx680 with Turbos to 1.2ghz in BF3, so in effective its already OC vs stock 7970 numbers.

Doesn't look that amazing in that light actually.

 

[Grooveriding]

Actually the overall average at 1600p is 3% and 9% at 1080p vs a STOCK 7970. We all know how well they OC.

That's a "stock" gtx680 with Turbos to 1.2ghz in BF3, so in effective its already OC vs stock 7970 numbers.

Doesn't look that amazing in that light actually.

It's a wash at 1600P between the two if you look at a large segment of games averaged:

Not to mention nvidia is taking us to the cleaners with such a small perf. improvement over the 580.

Came down to games for me, BF3 I play more than anything else and it is doing well on nv, and just the right moment I guess, cheaper too.

They better take at least 6 months to release GK110, or slap in the face, rageeeee.

 

[Ajay]

Yeah I am trying to wrap my head around it but it does appear to be the case that GPU Boost, which you can't turn off, is basically overclocking the GTX 680 so that comparing it at "GPU-boosted stock" to a stock 7970 is not entirely fair. I would like to see a lot of OC vs OC tests to determine what the actual difference is between the cards. I still expect 680 to win, but not by as much as the "GPU boosted stock" vs. real stock differentials.

I don't see that as unfair at all. Turbo boost is the stock performance mode of the GTX 680. Just like when Intel came out with Turbo boost for their processor line.

As an overclocker, the GTX 680 is a let down, since it looks like it might take a hardware hack to get around it.

 

[badb0y]

So in some cases the gpu auto OCs upto 1200mhz?

 

[tviceman]

I'm stuck at work and it's really slow right now, so I took 10 minutes to performance increase going from gtx580 to gtx680 at 1920 and 2560 on anand's benchmarks only. I know techpowerup does the same thing, but meh. Here are the improvements over gtx580

Improvement from gtx580 to gtx680 at 1920x1200
Crysis Warhead - 17.2%
Metro2033 - 32.6%
Dirt 3 - 36.7%
Shogun 2 - 62.9%
Batman AC - 35.2%
Portal 2 - 44.5%
BF3 - 40.4%
Starcraft II - 40.2%
Skyrim - 7.4%
Civ V - 8.8%

Average improvement across 10 games at 1920x1200 = 32.6%

Improvement from gtx580 to gtx680 at 2560x1600
Crysis Warhead - 18.5%
Metro2033 - 34.5%
Dirt 3 - 38%
Shogun 2 - 51.4%
Batman AC - 34.1%
Portal 2 - 47%
BF3 - 48.9%
Starcraft II - 39%
Skyrim - 37.8%
Civ V - 20.5%


Average improvement across 10 games at 1920x1200 = 37%


So 37% at 2560 is just shy of 40%, but I think overclocked gtx680's vs. overclocked gtx580's can hit rise above 40%. All in all, very good release in my opinion. It performed a little bit better across the board than I thought it would (I was think 30% average). I'm still going to wait for nonreference versions, and for prices to drop to around $450, but I think this is my next GPU purchase.

EDIT: Looks like HardwareCanuck's results are almost identical to Anandtech's.

 

[bryanW1995]

See this is the fallacy of your argument and just shows that you're here to cheer nvidia, regardless of what's in your signature (I'm preemptively denying that appeal). Each company offers products that are diverse in features and performance, yet you're trying to change goal posts or set up pseudo-restraints in order to say "THERE NVIDIA'S BETTER I WIN," which is ridiculous. I've already shut this down in another thread: http://forums.anandtech.com/showthread.php?t=2234044 , and I'll do it again here. Either participate in the discussion or don't post, I'm not derailing this thread further.

You are way off base criticizing him. For one thing, you're wrong. Gtx 680 is 10% faster and uses less power, it's a better card. Well, that kind of ends the discussion, right?

 

[Grooveriding]

So in some cases the gpu auto OCs upto 1200mhz?

TPU has the best explanation of OCing I have seen for the card

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_680/31.html

He explains the auto OC on the prior page as well. Certainly is not a straight up push your sliders and go for OCing card.

 

[RussianSensation]

Came down to games for me, BF3 I play more than anything else and it is doing well on nv, and just the right moment I guess, cheaper too.

They better take at least 6 months to release GK110, or slap in the face, rageeeee.

Ya but it's still just 30-35% faster than GTX580 in BF3 at 1600P. For you this card is still a pass. $1000 on 2 of these will barely be worth it over what you have. Long live the Tri-GTX480s.

GTX685 can't get here fast enough.

 

[blastingcap]

More OC vs OC comparisons: http://vr-zone.com/articles/asus-gtx-680-2gb-overclocking-review-win-some-lose-some/15322-5.html

This is gonna vary from card to card as chip quality differs of course. But at least they compare more than one game. 7970 doesn't look as bad when both cards are going all-out OC. But it's still clearly losing in BF3. Check out the AvP differential though, as that finally pushes the GtX 680 beyond its 2GB VRAM.

TPU also found that 7970 catches up a bit when OC vs OC:

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_680/31.html

 

[-Slacker-]

Yeah I am trying to wrap my head around it but it does appear to be the case that GPU Boost, which you can't turn off, is basically overclocking the GTX 680 so that comparing it at "GPU-boosted stock" to a stock 7970 is not entirely fair. I would like to see a lot of OC vs OC tests to determine what the actual difference is between the cards. I still expect 680 to win, but not by as much as the "GPU boosted stock" vs. real stock differentials.

Yeah, that and the potential problem that MrTeal pointed at - that not all 680s would dynamically overclock the same; On one hand that would seem like a non issue, since the boosts in clock speeds up form their base value (1006--->1120) aren't that high compared to more "enthusiastic" overclocks done on traditional video cards, so it may seem like virtually all 680s would reach that quota.

On the other hand, that true if you're thinking about manual overclocks; It may be that not every gtx680 could reach those highs due to the fact that a lot of variables are at play when the dedicated chip on the card determines how high the overclock will be, like power draw, temperature etc.




edit

That's how the consumer gets the product out of the box. The reviewers are doing exactly what should be done: Test a GPU the way most consumers are going to be using them. Then some are adding a section at the back with manual overclocking which allows HD7970 to catch up. That's fair. Stock GPU Boost is at most 1110Mhz. Three sliders allow you to increase the card’s power draw above its stated 195W by up to 30% for a total of 253W.

It looks like a 1200-1250Mhz HD7970 will actually beat GTX680. However, that automatically rules out all reference HD7970s though if you actually want good noise levels. Most non-reference HD7970s have a price premium right now. Also, we've seen some HD7970's fail to reach 1200mhz despite having premium components. Either way, you aren't going to be gaining much here since since even an 1125mhz HD7970 couldn't take out GTX680 in BF3. That pretty much means you'd need 1175-1200mhz on HD7970.

Sounds fair when you consider that many people don't overclock their cards themselves, but there are also many who do, for whom getting a hd7970 near the clocks that you're talking about would be a must.

I'm going to have to disagree with your reliance on one article to show that an overclocked hd7970 can't beat a "stock" gtx 680. We don't know the frequency that the gtx680 is running at, and I think we've established that it can boost it's speed to at least 1124Mhz form it's 1006mhz base clock.

 

[RussianSensation]

Actually the overall average at 1600p is 3% and 9% at 1080p vs a STOCK 7970. We all know how well they OC.

That's a "stock" gtx680 with Turbos to 1.2ghz in BF3, so in effective its already OC vs stock 7970 numbers.

Doesn't look that amazing in that light actually.

A stock GTX680 turbos to 1110mhz, not 1200mhz. You can create a new +50 to 100mhz offsets and the boost can hit 1200mhz.

Even with manual overclocking on the HD7970 it couldn't catch up to an overclocked GTX680. Also look at how consistent the performance of GTX680 is. If you play Metro 2033 most of the time, for sure get the HD7970. Nothing wrong with buying a card if you play 2-3 games most of the time in which HD7970 @ 1150mhz+ is actually faster. People who want an overall great performing card will want the GTX680 though, especially with a lower price and features.

Also, all this talk for 4 generations about power consumption a lot. Well here HD7970 with 1150-1200mhz GPU clocks is actually better than a stock GTX680, but it'll for sure consume at least 100W more power, if not worse.

If we can actually buy non-reference HD7970s for $475-500, then a case can be made for an overclocked HD7970. Problem is most HD7970s with non-reference coolers are pushing $570-600. And how many of them have EVGA's warranty? I think a reference HD7970 needs to drop to $450 (at least).