Print
On Tuesday we posted our preview of the GK110-based Geforce GTX Titan from Nvidia, and like all of yall we were eager to stuff the Titan into a test system to see what it could do in both single-card and dual-card configurations. Now that the dust has settled and our initial testing is complete, we have to say we think we misunderstood what Nvidia was said to us when we asked them how the Titan compares to the GeForce GTX 690. The Titan is one hellishly fast single GPU, but it's not the fastest single-card solution for gaming. That title still rests comfortably with the dual-GPU GTX 690.
According to notes we took at the initial briefing, when we asked them how the Titan compares to the GTX 690, the rep answered by saying the two cards were "eye-to-eye." At the time we thought he was talking about performance, but now that we’ve tested the cards and learned the two cards share the same $999 price tag, we’re beginning to think he meant the two cards both basically compete for the same customer, not that their performance is the same. Looking at our notes he said there are times when the 690 is faster, and times when the Titan is faster, but that's not what our benchmarks show. Oh no, they show the the GTX 690 is still the fastest single-card GPU available.
Double-trouble: Two Titans in SLI is total overkill for single-display gaming. Not that there’s anything wrong with that.
On the one hand, this is a bit of a letdown since everyone (including us) always wants to see a new champion crowned. And with all the excitement over the Titan launch, a lot of people wanted to see the Titan arrive on a shiny horse and lay down the law in Benchmark-ville. And it certainly has for single-GPU comparisons, but as the benchmark chart clearly shows the GTX 690 is still faster in every test, as it should be since it has dual GPUs. The Titan is clearly the fastest single-GPU card available though, and with AMD announcing its HD 7000 series cards will remain "stable throughout 2013" it doesn't look like there will be a challenge from AMD to the Titan's supremacy any time soon, or at least not until Q3 at which time AMD has spoke of new products coming, but we have no idea if a replacement for the HD 7970 is one of them. That does not mean the Titan is worth more than double the cost of a Radeon HD 7970 though.
So let's examine the numbers. First we tested the Titan against all of the other single-card flagships available, including the GTX 680 and 690, and the HD 7970 (non-GHz) and the malevolent HD 7990 from PowerColor. We began testing with an Asus Radeon HD 7970 TOP GHz card but it had some issues with a few benchmarks, so it was scrapped in favor of a non-overclocked card from Asus, which was unfortunate. We also added a few new tests this time around and removed several aging tests such as Stalker, Far Cry 2, and Just Cause 2. We used the latest "official" drivers for both Nvidia and AMD, and this gave a clear advantage to Nvidia as it had just released version 314.07 as we began testing, while AMD is still using its 13.1 drivers that were released in January. Nvidia made a Titan specific driver available to us for testing that was deemed "final" that was numbered 314.09. We should also mention that AMD pointed out that its Catalyst 13.2 Beta driver offered improvements in Far Cry 3, Crysis 3 and other titles, but we didn't use it due to its Beta status. We really wished AMD had made that driver final, as we would have loved having a new driver for our tests.
In single card tests the GTX 690 remains the card to beat, beating the Titan in every test by a decent margin. In some tests such as Far Cry 3 and Crysis it was relatively close, but the GTX 690 was always faster. It traded blows in testing with the dual HD 7970 card from PowerColor, which is about to go into the rig and hands of the winner of our Devil 13 Giveaway (stay tuned for the official announcement soon).
So what we see here is a continuum of performance when it comes to Nvidia GPUs that starts at the GTX 680 and then goes up quite a bit to the GTX Titan. Anyone upgrading from the GTX 680 to the Titan will see a performance boost, but it's not quite the doubling of performance we expected from the Titan's almost doubling of specs from the GTX 680. Performance improves even more going up to the GTX 690, which is almost double the horsepower of a GTX 680 both literally and in terms of the benchmark performance. So the old guard remains — the GTX 690 is still the card to beat if you just need one card to get the job done.
That said, the GTX 690 is too big and too expensive for most users. The same can also be said about the Titan, at least in terms of expense, but it's smallish 10.5-inch form factor will allow it to slide into most ATX cases and even some SFF cases with fewer issues than the GTX 690, making it a very compelling option for folks with smaller cases. There's also the SLI issue, as we all know that SLI or CrossFireX isn't always a walk in the park, as demonstrated in our testing. You have to wait for drivers to fully take advantage of it with the latest games and benchmarks, and that is highly annoying. At least with a single card none of that monkey business ever occurs.
Best scores are bolded. Our test bed is a 3.33GHz Core i7 3960X Extreme Edition in an Asus P9X79 motherboard with 16GB of DDR3/1600 and a Thermaltake ToughPower 1,050w PSU. The OS is 64-bit Windows Ultimate. All games are run at 2560x1600 with 4XAA except for the 3DMark tests.
*This score is indicative of only one GPU being recognized due to drivers not being compatible with this particular benchmark.
Hit the next page for dual-card benchmarks, Titan versus SLI/Crossfire Configs, and our closing thoughts.
Comments are closed on this article
Ninjawithagun
March 04, 2013 at 7:07am
I too have three EVGA GTX680 FTW+ 4GB cards in 3-way SLI and agree that they perform very well whether I'm running 3 x 24" 120Hz monitors at 6000 x 1080 (2 x 120 for bezel correction), or running a single 2560 x 1440 120Hz monitor (Overlord Tempest) the GTX680s do not disappoint in performance. Based upon the fact that only one game, Metro 2033, gains from any kind of performance improvement due to the GK110 GPU in the Titans, it's just not worth the upgrade. I really do hope Nvidia has something better up their sleeve come later this year. The Titan definitely feels like a gap filler and nothing more. Definitely not revolutionary, but definitely wallet-thinning to be sure :)
nickdanger013
February 24, 2013 at 2:34pm
what's up with the fire strike results? are these the graphics scores only or the total score? i have a sytem with an i7-3770 @4.2ghz, 8GB DDR3 @2000mhz and i pull between 6000 and 6100 with my gtx 670, and i pull over 7000 with my reference 7970, over 10000 when running crossfire. why is your score for 7970 so low??? i've seen 7970 pull number that low when you have a driver problem, like switching from Nvidia to AMD without using driver sweeper, but that is not looking like a correct result there.
nickdanger013
February 24, 2013 at 2:38pm
*correction 6644 for single 7970
http://www.3dmark.com/3dm/131824
crossfired 7970:
http://www.3dmark.com/3dm/256803
Galaxy GC GTX 670:
http://www.3dmark.com/fs/91620
Fyrwulf
February 24, 2013 at 11:00am
So let's get this right, rather than buy a 7970 Ghz from another manufacturer or even asking them to send you one, you asked Asus to send you a neutered 7970? You also didn't flash the non-Ghz with the Ghz BIOS, which is an AMD-sourced way to turn a non-Ghz card into a Ghz card.
There's a serious integrity fail here in that your WORDS made it sound like a podium sweep for the Green camp, yet from actually looking at the benchmarks I see the 680 about even with the non-Ghz 7970, which basically means that a Ghz 7970 would consistently beat the 680. Also, the driver excuse is really weak, since most every Radeon user happily installs the newest beta drivers and rolls back if there's a problem.
joshnorem
February 25, 2013 at 6:15pm
Hey Frywulf - we had a 7970 TOP from Asus that worked fine for most of the testing but then started to freak out the day before deadline. We also had to test the SLI setups and all of the other cards and there was simply no time for us to swap the card out and start all over. We've also neve been able to get two GHz cards from AMD and we've asked, believe me. We asked again today. I'll let you know if we ever see that dream fulfilled.
Why would we have a problem publishing benchmarks showing the HD 7970 being faster than the GTX 680? The last time we compared to two we stated that the HD 7970 had the advantage, at least in terms of frame rates. We know the cards are eye-to-eye for the most part.
Regarding the drivers we're considering changing our policy but the "no beta" rule has been in effect for at least as long as this magazine has been around. We are considering changing it though because it is a major bummer to not be able to run dual GPUs on newer games and we hate having situations like this.
Thanks for your feedback on all of this. I appreciate it.
nickdanger013
February 26, 2013 at 6:08pm
i would like to see the no beta rule go away. i saw a significant bump in performance on my 7970's running catalyst 13.2 beta. maybe not go away, but include a section with an asterisk where you compare Nvidia's beta and non-beta drivers to AMD's beta and non-beta drivers.
joshnorem
February 26, 2013 at 8:02pm
We discussed it today and I think you're right. We're going to do away with it and just say what driver we used.
kixofmyg0t
February 25, 2013 at 10:50am
Maximum PC is heavily biased towards nVidia. Remember the laughable AMD build they put together a few months ago? It wasn't a attempt to please any AMD fans. They put the AMD build against clearly more expensive(way WAY more expensive and in turn faster) hardware just to get the numbers to look as bad as possible.
I'm a bit late reading this because when I saw it I figured MaximumPC would skew the numbers to make the Titan blow everything else out of the water. Only after seeing benchmarks on other sites did I come back here and actually read their results.
When the Matla based 7990 drops, if MaximumPC even bothers to mention it you can bet they'll throw it into a 4 core Bulldozer (or i3) rig to test against Zero point.
joshnorem
February 25, 2013 at 10:29pm
Hi Kixofmyg0t - testing the AMD system against the Zero point sparked a debate in our office because it was too lopsided, I agree. We certainly didn't build the system just to trash it in the end. We are not invested in the results of these tests and get no money from either manufactur, nor do we get any special treatment from Nvidia either. We are looking for a better way to compare systems that we've built and it will not involve the zero point because you are right, it's not always a fair comparison.
And if you think we skewed the results to make AMD look bad I don't know what to say to that. However, we will be doing some more testing so be sure to watch for an update shortly that will include the Ares II :)
josh
JCCIII
February 23, 2013 at 5:16pm
NVidia has not finished working on the GTX Titan; it’s just in a new phase, the con’sumer phase, and what a scam it is at $1000. It is bested by the GTX 690 in every test; ‘as the benchmark chart clearly shows, the GTX 690 is still faster in every test (MPC),’ and ‘Two GTX 680s in SLI still beat out two GTX Titans in SLI in several tests including Batman, Crysis 3, and Heaven 4.0 (MPC).’
This is relevant because whenever NVidia’s people release their single-fastest GPU solution, it is lesser than their previous fastest GPU in SLI and dual GPU card, here two GTX 680s or one GTX 690. Just understand, with their GTX Titan, the company’s people continue to push for new higher prices, passing the $500-$600 mark, really—to $1000—with hype and artificial shortages to create false demand, working the market with their greed.
An example of the way NVidia operates, the GTX 480 was more than $500 at the time when I bought three. After that, I discovered all GTX 480s were so lame that they were disabled at the factory; but, a few months later the GTX 580 arrived, and it was a real Fermi. NVidia’s people tricked me with their GTX 480.
Do not let NVidia rip you off as I have let them do it to me. I trusted them. NVidia’s people will scam you if you are ill-advised, and they will make it sound fun.
Sincerely,
Joseph C. Carbone III; 23 February 2013MPC @ http://www.maximumpc.com/article/features/nvidia_geforce_titan_%E2%80%93_benchmarks2013
aferrara50
February 26, 2013 at 11:37am
you're obviously not the target market for the card. The titans are gold when it comes to tri or quad sli. I sold my 690s and picked up 4 EVGA Titan SCs b/c the 690s didn't have the vram I needed for a 7680x1440 resolution. The vram limitation was so great max payne 3 couldn't even run 8xmsaa. Considering that they're the fastest single gpu card ever made the price isn't that bad.
It performs ~20% less than a 690, but it only single gpu, allowing for you to run 3 or 4 of them, the reason the card is so great.
TBH there's no reason to have 1 of them. 2 is understandable if you're running surround and want to crank up MSAA, but other than that 3 or 4 are the sweet spot.
Baer
February 22, 2013 at 5:11pm
Really a great review and useful data, thank you.
I am specking out a new high end rig and I have been back and forth between the 690 or a pair of 680's.
I run three 1920 X 1200 (5760 X 1200) in surround, not only for gaming but also for productivity programs. In reading this it looks so far that a pair of 680's may be my best bet. (I welcome comments on this)
I am looking forward to the updated report after the additional testing is done.
Renegade Knight
February 21, 2013 at 6:14pm
I'd like to see a triple monitor set up review. I was sold on the GTX680 as my next upgrade but checked it's performance against the 690 and learned that with a 35fps goal on the GPU intensive games it was going to have to be the 690. If this could pull it off I can save a little on the power bill (though not yet the wallet).
ripple
February 21, 2013 at 3:58pm
Thanks for the review, nice job on this one. According to your results, I should be buying the GTX 690 but I am leaning toward the new technology in the GTX Titan. I just hope that I can get my hands on one when it is available!
vistageek
February 21, 2013 at 3:07pm
Something is wrong here. The Titan SLI should beat the 680 SLI without any trouble. A single titan should lose to a gtx 690 by %15ish.
joshnorem
February 21, 2013 at 7:35pm
These benchmarks were completed in time for the NDA to lift but we had to cram all night to get them finished. There were some anomolies though as you pointed out, and we will investigate. In the end we felt confident enough to publish them as they were straight from our test machine but further testing is needed. We also need to try tri-sli and some other tests as well.
John Pombrio
February 21, 2013 at 1:52pm
The rumors pretty much nailed the specs beforehand so really no surprises (good job guys!). Its just that the card is too expensive for the benchmarks it kicks out. When it get to $800, I will buy it...
jstein076
February 21, 2013 at 10:03am
I'll keep my GTX 680 ftw+ 4gb in tri-sli. They were way cheaper and seems like they will have the same if not better performance.
ugzz
February 21, 2013 at 8:44am
with the 6GB of ram on the Titan and the 2GB (usable) by the 690 you guys *need* to test a triple head setup. Even at 2560x1600 with just over 4 million pixels your not really pushing the limits of the memory, but with the 690s 2gb i wouldn't be surprised if you saw some very interesting numbers at say 5760x1080, adding another 1.5 million pixels. Or.. dare i say it.. 7680x1600, for a whopping 12+ million pixels. (or the easier to come by 7680x1440)
Honestly i'm super surprised i'm not seeing more of these benches now that Nvidia and AMD both support it natively with 1 card now.
Bundle that with falling monitor prices and the availability of 2560x1440 korean monitors in the 300-400 range.
Silverbullet
February 21, 2013 at 7:56am
Great review, although a bit of a let down from Nvidia. This card only makes sense if you were looking for pure compute HP and NOT for gaming. If gaming, stick with the proven cards already out there for less than half the price. Would have been nice to find out how she overclocks...early reports are that the core can be cranked (1150+) which should make a pretty significant difference. Still not enough to justify the price if you're only looking for a gaming GPU, IMO.
acidic
February 21, 2013 at 8:09am
guru3d.com has a nice article with 3 way SLI and 3 monitor setups. hilbert got his up to 1176 in the oc guide
HiGHRoLLeR038
February 21, 2013 at 7:39am
nice, thanks guys! So i only need to drop 2 grand on gfx cards to play Crysis 3 on max? yikes...
iplayoneasy
February 21, 2013 at 7:29am
You want the fastest, low temp, overclockable single card? You buy and Asus Ares II
trajan2448
March 04, 2013 at 4:35pm
Until the runt frame problem in Crossfire is addressed Crossfire is effectively useless. The fps numbers are highly inflated with slivers of frames which are unviewable. This has been reported in Tech Report, PC Perspective and Hardocp. This has sparked many investigations into FRAPS, latencies, and how to ACCURATELY judge performance. Buyer beware.
Airheadq
February 21, 2013 at 7:14am
How about compute performance metrics? That's the other side of the coin to Titan.
Peanut Fox
February 21, 2013 at 9:03am
This is what I'm looking forward to seeing. Some Blender render tests with Cycles please! As it stands the 3GB 580 is still the card to beat.
Peanut Fox
February 21, 2013 at 5:38pm
Not in Cycles render times. They're worlds apart. OpenGL just doesn't do it in that space.
AFDozerman
February 22, 2013 at 7:18am
Just reading online, it looks like OpenCL is only supported on Nvidia cards for some reason. So actually, we were both wrong. You are actually comparing GPU compute to CPU instead of Nvidia to AMD. I knew something was fishy about a card from 2009 beating a few month old card, especially when AMD has been touting OpenCL compute for years...
