In the lead-up to Computex, NVIDIA is heading the pack, releasing their new GEFORCE GTX 780. This is their flagship for the GTX 700-series. While TITAN should remain at the top of NVIDIA’s single-GPU lineup, the GTX 780 – based off of the same GK110 GPU – shouldn’t be any slacker. Read on and we’ll show you just how close the GTX 780 can get!
Specifications & Features
Right off the bat, in their presentation, NVIDIA was out for blood. The GTX 780 has 50% more cores (up from 1,536) and 50% more memory (up from 2 GB) than the reference GTX 680 that came before it. The TITAN had 2,688 cores, so they didn’t cut the GK110 down too far for the GTX 780.
NVIDIA is claiming big strides in performance too, with a 70% improvement over the GTX 580 and over 40% improvement over the GTX 680.
The GTX 780 comes with GPU Boost 2.0, which you first learned about in our GTX TITAN introduction. If you missed that or need a refresher, go back and check it out.
Using the same cooler as the GTX TITAN, the GTX 780 can also claim very solid gains (or should I say losses) in the sound department. While this cooler is very quiet, it’s also tuned to allow the GPU to get up to a certain temperature (80 °C) and then hold it, which helps them obtain such quiet operation. That said, as we’ll see, you might want to make some adjustments to keep the GPU farther away from the 80 °C level because of how GPU Boost 2.0 works, but we’ll get into that later.
Speaking of cooling, NVIDIA has taken an additional step in the right direction. During our conference call, they pointed out that our ears pick up on changes in pitch and volume, even more so than slightly higher -but consistent- volume. NVIDIA has addressed their fan continuously ramping up and down by introducing their new Adaptive Temperature Controller. This will maintain the same temperatures as before, but do so while keeping the fan RPM ‘curve’ flat.
Here is where the rubber meets the road – NVIDIA’s claims of improvement over the competition’s HD 7970 GHz edition. Long in the tooth, the HD 7970 is still doing well for itself, but as you can see, in every metric NVIDIA has displayed here, you get a large increase over the HD 7970 GHz Edition’s performance.
They are also claiming solid SLI scaling. As measured with FCAT, greater than 175% SLI scaling appears pretty common. Speaking of SLI, like the TITAN before it, official support will extend only to tri-SLI. That’s ok because quad GPUs from either camp are known to scale very poorly, but so you know, there won’t be official support for quad-SLI.
Finally we have the specifications. This is a very conservative base and boost clock, as you’ll see in our overclocking results, but you have to start somewhere and NVIDIA starts the GTX 780 at 863 MHz base and 900 MHz boost, which is actually slightly higher than TITAN’s 837 MHz & 876 MHz, respectively.
As mentioned, the GTX 780 comes with a 3 GB frame buffer and it benefits from the GK110’s 384-bit memory bus. Clocking in at 6000 MHz GDDR5 (1,500 MHz quad-pumped), you’ll benefit from the same wide bus and fast speed as TITAN, just with half the buffer size.
What isn’t mentioned in their slides is price, so let’s get that out of the way. The reference GTX 780 will come with an MSRP of $649.
NVIDIA GEFORCE EXPERIENCE
Launching at the same time as their GTX 780, NVIDIA’s GEFORCE EXPERIENCE is coming out of beta today.
NVIDIA actually framed the GEFORCE Experience in terms many people can understand. They always attempt to get “game ready drivers” out the day a new game releases. They asked an important question: “What’s the first thing you do when you get a new game?” Do you go check your GPU manufacturer’s site for a new driver, or do you install it and play? You install it and play of course!
GEFORCE Experience takes control of your drivers and should automatically let you know that there is a new driver available and download it for you on the day that your shiny new game arrives. They’re trying to make PC gaming more accessible to the layperson. While many of the people reading this review would indeed check for the latest driver, using this new software will take that worry off of your shoulders.
They are touting their game ready drivers in this slide, which compare launch-day drivers with the ones available previously, and with solid gains it’s worth it to have the newest driver when your next big game comes in.
Drivers aren’t the only thing NVIDIA is striving to have ready on day one. They also will make every effort to have optimization profiles pushed out on the day a game launches. These profiles will check your system for which GPU & CPU you have and automatically set your graphics quality to match your system. A lot of you might want to do that yourselves (I know I would), but if you want one click set it and forget it optimization, NVIDIA has the ability to do that. That they plan on doing so on release date is icing on the cake.
Last, but not least, NVIDIA isn’t yet done with GEFORCE Experience. Later this year (late summer ETA), they plan on introducing ShadowPlay, a new gameplay recording feature. This isn’t your typical FRAPS recording, where you set it before you play, hoping to get something cool worth saving. ShadowPlay can be set to auto-record everything you do in real time, with a buffer of up to 20 minutes.
For instance, say you pull off the kill of your life. I’ve seen a video similar to this – a guy in the middle of a dogfight in BF3 jumps out of a plane, snipes the person they’re dog fighting, gets back IN the plane, and goes on about their business. That person happened to be recording his video at the same time, which is great. However, they had to sift through every bit of video they had to find the right place.
ShadowPlay takes out the guess work. It will automatically and continuously record your gameplay, deleting from the back-end as it goes. If you pull off that insane kill, you hit a hotkey and it instantly saves the last X minutes of gameplay to your drive for you to go back and look at later.
In addition, ShadowPlay doesn’t have near the overhead that FRAPS would have. There is a good chunk of FPS lost when recording with FRAPS. ShadowPlay will work on any Kepler card and will use the built-in H.264 encoder to handle the heavy lifting. NVIDIA expects only a 5% or less overhead when using ShadowPlay, so it will barely be noticeable.
NVIDIA has done good things with GEFORCE Experience. If you have a GEFORCE card, it would be worth looking into this program. If you have (or get) a Kepler card, when they add ShadowPlay it’ll be worth the download just to check that out.
Meet the NVIDIA GTX 780
With that out of the way, let’s meet the GTX 780. In what is perhaps the most poorly kept secret of the year, the GTX 780 does indeed look just like its big brother TITAN.
Thanks to our handy-dandy gallery, you don’t have to slog through a bunch of thumbnails any more either.
Like the TITAN before it, the GTX 780 will require two (one 8-pin and one 6-pin) PCIe power connectors. The GTX 780 comes with a TDP of 250 W.
The standard reference NVIDIA video output configuration should look familiar to everyone by now. The GTX 780 is no exception and comes with four video outputs – two dual-link DVI (one DVI-D, one DVI-I), one HDMI and one DisplayPort.
Sporting the same cooler, the GTX 780’s “GEFORCE GTX” text also glows a pleasant green when powered up. No word yet on whether EVGA’s software can control this one, but it should be able to.
So far so good. We loved the look of the TITAN, so you shouldn’t be surprised we’re fans of the GTX 780 as well. There will be plenty of partner-designed cards on the market soon, complete with custom coolers; but if you want a blower-style cooler, the reference model is a solid offering.
Under the Hood
The GTX 780 cooler is the same as that on the TITAN. It’s a very premium looking (and performing) cooler.
The cooler uses a vapor chamber connected to a nice long aluminum fin stack.
The PCB should look familiar, as it’s the same as the TITAN. The only significant differences are the vRAM (there is none on the back of the PCB this time) and the GPU itself, which is reduced from 2,688 CUDA cores on the TITAN to 2,304 on the GTX 780.
All 3 GB of vRAM resides on the GPU-side of the PCB.
The power section appears to have minor changes, but still has the same six-phase power design. It’s plenty for every day operation and moderate overclocking, but don’t go volt-modding one of these on the stock power plane. They’re not designed for it and you will blow something up (probably a MOSFET).
Finally, we have the GK110 GPU itself.
One more gratuitous PCB photo and we’re done here (thanks to NVIDIA for supplying the card photos).
Like all our GPU reviews, our test bed consists of an Ivy Bridge based system with an i7 3770K and RAM running at a reasonable DDR3-1866.
|CPU||i7 3770K @ 4.0 GHz|
|MB||ASUS Maximus V Extreme|
|RAM||Kingston HyperX Predator DDR3-2666 @ 1866MHz 9-9-9-24|
|GPUs||ASUS GTX 670 DirectCU II TOP
MSI GTX 680 Lightning
HIS HD 7970 X Turbo *
ASUS ARES II
NVIDIA GTX TITAN
NVIDIA GTX 780
|OS||Windows 7 Professional x64|
*Note the HD 7970 X Turbo was the only GPU tested with AMD’s new driver with our GPU test setup when I graphed these results. It is a HD 7970 with strong stock clocks, operating at 1180 MHz Boost and 1500 MHz on the memory. Thus, you’re seeing some of the best AMD can come up with vs. the GTX 780. The results may surprise you.
Installed, the GTX 780 looks just as nice as it did on its own.
NVIDIA GPU Boost 2.0
To reiterate, you should go back and check our our original description of GPU Boost 2.0 from our TITAN introduction. Once you’ve done that, come on back.
Back? Ok, good. Now, this graph bears a little bit of explanation. As you just saw, GPU Boost 2.0 bases its boost not only off of TDP constraints (which you see mostly when overclocking), but also temperature constraints. Once the GTX 780’s GPU reaches 80 °C – which is its default temperature target – it will ramp up the fan and reduce the GPU frequency to make the card stay at its temperature target.
This is where NVIDIA’s silence starts to work against them. As you can see below, the max stock boost clock is a strong 1006 MHz. Looking at the stock Heaven Xtreme run though, you can see precisely where the GPU hits 80 °C and the frequency tanks. It then goes through continuous fluctuations while the card wavers between 79 °C and 80 °C.
However, if you use PrecisionX (or the GPU software of your choice, depending on whose GTX 780 you buy) to set a more aggressive fan profile to sacrifice a little bit of that silence – while keeping the GPU away from 80 °C, you can see the frequency goes to 1006 MHz and sticks throughout the entire benchmark run.
Thus, you have a choice to make. You can let NVIDIA maintain its near complete silence (and it really is quiet!), or you can take control of your card’s fan profile and let it boost to the level it’s meant to boost. It doesn’t take a rocket scientist to figure out we prefer performance to silence at this site, but you have the choice.
This card really surprised me with its overclocking capability. Coming in with a 24/7 overclock of +180 MHz on the GPU and +350 MHz on the memory (in Precision X; actual frequency was +170 MHz per GPUz), the GTX 780 was very impressive, especially with the meager +38 mV NVIDIA allows us to toy with.
You haven’t seen any performance yet, but this solid overclock managed to score 13353 in 3DMark11 Performance, a very strong showing.
So, what does that overclock get us in actuality? Why, it looks to be a very nice 1201.9 MHz. With the aggressive fan profile set, the GPU is allowed to stay up there most of the run too.
Overclocking the GTX 780 was very impressive. If you could give it more voltage, it would be even more impressive, but with the less than strong enough power section, +38 mV is as much as you’re going to get.
What is very disappointing is that this +38 mV limit is going to stay with us, from this reference card all the way up to whatever card partners produce (think MSI Lightning, ASUS TOP or EVGA Classified models), even if they have a more robust power section, NVIDIA is only going to allow that same +38 mV limit.
They say this is for safety reasons – they want to give a little flexibility (38 mV) and will let users potentially impact the long-term well being of their cards, but they refuse to allow higher voltage increases that may jeopardize the short-term life of their GPUs.
While this is understandable, I’d love it if they would let the partners make that decision. With a stronger power section (such as EVGA’s ePower), we’ve seen great things come out of TITAN when you throw a bunch of voltage to it. Without an external power section, TITAN dies faster than you can say burnt MOSFET. So it doesn’t seem like the GPUs are the problem, but NVIDIA’s weak reference power designs.
We are hoping against hope that board partners figure out a workaround to this limitation. If nothing else, design your cards with very robust power sections and give us solder points to easily and quickly attach volt mods. That wouldn’t violate NVIDIA’s restriction and would give us the choice of doing what we please.
All of that said, the GTX 780 has very strong overclocking results even without a large amount of voltage control.
Temperature and Power Consumption
When viewing the temperature graph, keep in mind this is not a referendum on the cooler’s capability. The GTX 780, like the TITAN before it, is made specifically to warm up to 80 °C and stick, using a combination of increased fan speed and frequency throttling. Seeing these go up to 80 °C is normal and intentional.
The cooler is capable of much better temperatures, but you have to compromise on silence. It’s still very quiet – quieter than any other blower-style heatsink I’ve heard, but it does make noise. If you’re willing to make that compromise, as an example, in the Heaven Xtreme run above where you see it keeping a consistent boost frequency throughout, temperatures stayed in the low-to-mid 60 °C range, far from the 80 °C you see here.
Power consumption is right where it should be for a slightly cut-down TITAN, coming in between 40 and 50 W less, topping out at 353 W in 3DMark11.
Cooling is right where it’s designed to be and power consumption isn’t bad for a powerful card with a 250 W TDP. Let’s start testing this thing!
All video cards we test are tested per our Video Card Testing Procedure. Long story short – benchmarks are run at their default settings and games are tested at 1080p with all settings turned to max.
Good old 3DMark03 hasn’t ever been Kepler’s bag and we don’t see it getting better here. We continue to include it because it’s valuable to benchmarkers, but there is really no correlation to modern gaming, so don’t read too much into this result.
3DMark Vantage, on the other hand, is a good place to start meaningful results. The GTX 780 starts off strong, with an overclocked result better than the TITAN and beating the 7970 X Turbo.
3DMark11 shows much the same result. If the GTX 780 keeps this up, we might not only have an HD 7970 GHz-beater, but a strong alternative to TITAN.
Wow, I’m not sure they cut TITAN down enough, what do you think?
Another bench, another win. Things are looking good for this card!
Starting off with Aliens vs. Predator, the GTX 780 continues its trend of beating the HD 7970 and its overclocked result taking out the stock TITAN. This is turning out to be a very strong card.
More of the same with Batman: Arkham City.
Here the HD 7970 X Turbo actually squeezes out a win with the GTX 780 at stock, barely. When it comes to Battlefield 3, the GTX 780 has some competition.
The stock GTX 780 actually beats its big brother TITAN in Civilization V.
Dirt 3 shows the TITAN coming out on top of even the overclocked GTX 780, but the two are very close. The HD 7970 lags behind by a little bit, then a lot when you overclock the GTX 780.
Metro 2033 continues to be the number one toughest game in our repertoire. Only Crysis 3 is worse and I’ve started recording those results as well. Here the TITAN is the clear winner, with the GTX 780 an easy second and the HD 7970 trailing behind.
Well, it’s pretty clear – at 1080p, the GTX 780 is top of the heap. Only the TITAN comes out ahead – and even then the stock TITAN often loses to the overclocked GTX 780. I’m very glad they didn’t cut it down more than they did, because that’s great news for consumers who want a more affordable option.
NVIDIA Surround Testing
Surround testing takes all the games (except Civilization V, which isn’t cooperating) and runs them in a triple-monitor setup in NVIDIA Surround / AMD Eyefinity configuration. This is where horsepower meets the road. If your GPU isn’t strong enough, it will crash and burn here.
|ASUS GTX 670 DirectCU II TOP
ASUS Matrix HD 7970 Platinum
ASUS ARES II
NVIDIA GTX TITAN
NVIDIA GTX 780
The Surround results do a great job of mirroring all of the games above – the GTX 780 spends most of its time out-performing the HD 7970 and, when overclocked, beating out the stock TITAN. Nothing to complain about here.
When considering such high resolutions, this is as good of a time as any to go over the GTX 780’s frame buffer. As mentioned, the GTX 780 comes with a 384-bit wide, 3 GB frame buffer operating at 1502 MHz. There have been people since the beginning of time complaining that 2 GB this or 3 GB that just isn’t enough. In the case of the GTX 680’s 2 GB, they may well have been right; however, NVIDIA thinks they have hit the sweet spot at 3 GB.
To those that say 3 GB isn’t enough, NVIDIA says, in their testing, it is enough for every game on the market today and the ones they know about that are coming. If you are adamant that 3 GB just isn’t enough for you, we’ve got some bad news – there will be no GTX 780s on the market – aftermarket or reference – that will come with a frame buffer larger than 3 GB, period. If that is extremely important to you (which it really shouldn’t be, based on the results above), NVIDIA would point you to TITAN.
Pushing the Limits
There isn’t much to this section – because of a certain other review, the water loop was in use elsewhere. The cooler used was an Intel cooler, and you know how great those are for overclocking. Thus, I picked the most GPU dependent benchmark (3DMark Fire Strike) and went to see how far the GTX 780 would go…and go it did, up to +190 MHz CPU offset and a whopping +650 MHz (in PrecisionX, +325 MHz actual) on the vMEM, giving us a score of very close to 10,000 for 3DMark Fire Strike Performance.
I bet it would have topped 10,000 if I could have overclocked the CPU a bit. This is a very strong showing. We can expect good things from those that are willing to hard-mod their cards (either with external power or hard-modding partner cards that have beefed up power sections).
Final Thoughts & Conclusion
Based on the venerable TITAN’s GK110 GPU, the GTX 780 is a very solid offering to lead the way into the new GTX 700 series. The performance is just great, absolutely nothing to complain about and plenty to celebrate. The GTX 780 is an absolute beast.
Frankly, I also like the reference cooler. They had a winning design and stuck with it, no complaints there either. For better or worse though, we probably won’t see a great many of those on the market. Partners are rushing to get their custom designs out the door and the majority of them will have different cooling. At least one partner I’ve spoken with said they will make reference cards, but they will be a very limited run; the majority of the production effort will be put into their custom designs, which is understandable. They need to be able to separate themselves from the competition and everyone is going to be trying to do that. You can expect the first of these very shortly and more will roll out as the weeks go on.
On the software side, GEFORCE Experience is a neat tool and should help do what it professes – make PC gaming more accessible. Keeping people that may not be keen on watching for driver updates up to date with the latest drivers is no small thing. For NVIDIA, it will at least keep them happy with the performance of their cards and not complaining about poor performance on the day a new game releases. The profiles will be perfect for those same people too. What has me excited -and probably a good number of our audience- is the upcoming ShadowPlay feature. That will be a …wait for it… game changer. Har har.
Back to the GTX 780, let’s talk a little bit about the price – $649 MSRP. It’s a little steep for me, but that’s what you can do when you’ve got the best. AMD did it with the HD 7970 when it launched and it’s tough to blame NVIDIA for doing it now. Until AMD has an answer for GK110, I fear we’re stuck with a high premium for both this card and TITAN.
In the TITAN review, I said that was an $800 card with a $200 price premium for being the best GPU on the planet. The GTX 780 is similar, but on a lesser scale. This is a $600 card with a $50 premium for being very near the best. It’s regrettable but not unexpected; that’s the way the market works.
Taking TITAN’s current price into consideration, they’re putting the GTX 780 where it should be, offering significant savings ($350 is nothing to sneeze at) for a card that can overclock far enough to get to TITAN’s performance level. So, is $649 a fair price? It’s very close, if not quite where I’d like to see it. $599 would look better when looking at purchasing one of these.
With that, I think I’ll put us down for today. What has the GTX 780 got to offer? Stellar performance? Check. Quiet, effective cooling? Yep, it has that too. It’s even power-efficient for its performance level. Yes folks, NVIDIA has taken the top single GPU spot and now offers a more affordable option with very near the same performance. AMD has their work cut out for them.