Original Link: http://www.anandtech.com/show/7103/nvidia-geforce-gtx-760-review
NVIDIA GeForce GTX 760 Review: The New Enthusiast Keplerby Ryan Smith on June 25, 2013 9:00 AM EST
A little less than a month has transpired since the launch of the GeForce GTX 770, and we’re back again with another new NVIDIA card. As is usually the case with NVIDIA video card family launches, NVIDIA has started at the top and has been working their way through the list, refreshing consecutively lower-end video cards. GTX 770 saw the $400 price point refreshed with a new high-end GK104 part, and now the sub-$300 price point is due for a refresh with a new lower-end GK104 part.
Launching today, that part will be GeForce GTX 760, the direct successor to the GTX 660 Ti. Whereas the GTX 770 fully supplanted the GTX 680 and put the GTX 670 in a tough spot, GTX 760 will be doing the same thing from the other direction, replacing GTX 660 Ti while also assuming the rest of the market segment formerly inhabited by the GTX 670. As a result NVIDIA is gunning for performance between a GTX 660 Ti and GTX 670, while bringing the price down to $249, nearly $50 below GTX 660 Ti. Consequently this is a lot like the GTX 770 launch – a GK104 refresh that will see NVIDIA delivering more for less.
|GTX 770||GTX 760||GTX 660 Ti||GTX 560 Ti|
|Memory Clock||7GHz GDDR5||6GHz GDDR5||6GHz GDDR5||4GHz GDDR5|
|Memory Bus Width||256-bit||256-bit||192-bit||256-bit|
|FP64||1/24 FP32||1/24 FP32||1/24 FP32||1/12 FP32|
|Manufacturing Process||TSMC 28nm||TSMC 28nm||TSMC 28nm||TSMC 40nm|
The GTX 760 is essentially composed of the only GK104 configuration we didn’t see with the 600 series: 6 SMXes attached to a full 32 ROP, 256-bit memory bus. This is the fewest number of enabled SMXes on any desktop GK104 part – GTX 670 and GTX 660 Ti both had 7 SMXes – but it’s 32 ROPs versus the 24 ROPs on GTX 660 Ti. As a result we’re looking at a configuration that technically has less shading and texturing throughput than the retiring GTX 660 Ti, but more ROP throughput and memory bandwidth than that same part, on-par with the equally retiring GTX 670.
Note that this is also the only desktop GK104 part that potentially ships with 3 GPCs; with 2 SMXes disabled, a GPC goes with it if those SMXes are part of the same GPC. Though as with the other NVIDIA parts where there’s a variable number of GPCs, we haven’t seen a measurable impact on performance, so it’s more of a technicality than a meaningful difference.
Like the GTX 770, the GTX 760 is fairly highly clocked for a GK104 part. Its base clockspeed of 980MHz is 65MHz (7%) higher than both the GTX 670 and GTX 660 Ti, which helps to offset the lost SMX. At the same time this gives the GTX 760 more ROP throughput than even the GTX 670. The end result is a product whose performance bounces between the GTX 660 Ti and GTX 670 depending on the workload, and typically is biased towards the GTX 670. Altogether we’re looking at about 92% of the GTX 660 Ti’s shading/texturing performance and 43% more ROP throughput. Memory clocks on the other hand are unchanged at 6GHz, but thanks to its full 256-bit memory bus GTX 760 will have 33% more memory bandwidth than GTX 660 Ti.
On the performance front GTX 760’s other trick compared to the GTX 600 series parts it replaces is that like the GTX 770 it’s a GPU Boost 2.0 part, meaning its boost performance is further min-maxed than GPU Boost 1.0 parts. Thermals willing, our sample boosts to 1149MHz out of the box – the official boost clock is over 100MHz lower at 1033MHz – a bigger and more frequent boosting outcome than what we saw with the GTX 660 Ti. So even though GTX 760 still theoretically has lower shader performance than GTX 660 Ti, there’s only a single benchmark where the GTX 760 doesn’t beat its predecessor, and it’s a non-gaming benchmark at that.
Meanwhile GTX 760’s TDP will stand at 170W, the same TDP as the GTX 670. Accounting for throttling differences from GPU Boost 2.0 that will have GTX 760 reaching its TDP more often, GTX 760 will essentially draw as much or more power than GTX 670, never mind GTX 660 Ti. Like GTX 770 NVIDIA has little choice but to raise their TDP here at any given tier in order to unlock more performance. 170W in turn is a rather intentional number that will allow NVIDIA’s partners to recycle their GTX 670 designs since the maximum power draw (and maximum heat generation) remain unchanged. Idle power on the other hand is going to be marginally lower, as like the GTX 770 the GTX 760 implements a lower idle clock state that shaves off a couple of watts.
Moving on, with this launch we’ll see NVIDIA shrink their desktop product stack a bit. With GK110 based cards at the high-end of NVIDIA’s stack and price compression below $500 pushing prices down, the desktop GK104 product stack is being reduced from 3 cards – GTX 660 Ti, GTX 670, GTX 680 – to just two cards. NVIDIA’s configurations leave them an out should they ever need it, but as it stands these will be the only two GK104 cards in NVIDIA’s 700 series lineup, with GTX 770 and GTX 760 splitting up the old market segments.
With a launch price of $249, GTX 760 is another aggressive launch for NVIDIA compared to both their previous generation products and AMD’s current lineup, with NVIDIA now offering performance notably better than what have been more expensive products. Other than its lower 150W TDP, GTX 660 Ti is fully redundant at this point; GTX 760 is faster and cheaper. GTX 670 is still going to be a hair faster, but with GTX 670 prices still above $300 it’s not being placed at a value sweet-spot, and like the rest of the GK104 based 600 series lineup is slowly being phased out entirely.
The more dynamic matchup will be GTX 760 versus AMD’s product lineup. $249 cuts right down the middle of AMD’s current product lineup, putting GTX 760 between the Radeon HD 7870 and 7950. At the same time NVIDIA’s goal here is to beat the 7950 in performance – keep in mind the GTX 660 Ti tied the original (non-boost) 7950 – which is something the GTX 760 doesn’t have too much trouble with. AMD and their partners have already prepared for this somewhat, as after-rebate prices on 7950 cards have come down from about $300 to about $280 in the last month, but that means GTX 760 still carries a $30 lower price tag. Once more there are no bundles in play on the NVIDIA side, so the wildcard on the AMD side will once more be their 4 game Level Up with Never Settle Reloaded bundle.
Finally, this is another hard launch. GTX 760 cards will be available from the usual retailers and e-tailers. Meanwhile since GTX 760 is designed to be a drop in replacement for GTX 670 on board designs, NVIDIA’s partners will be launching with fully custom cards right away, most of them based on their previous customized GTX 670 designs. Reference style cards will likely be few in number, but unlike the GTX 770 launch (with its relatively expensive Titan cooler) the GTX 670/760 reference cooler does not price itself out of the market.
|June 2013 GPU Pricing Comparison|
|$400||GeForce GTX 770|
|Radeon HD 7970||$360|
|$350||GeForce GTX 670|
|Radeon HD 7950||$280||GeForce GTX 660 Ti|
|$250||GeForce GTX 760|
|Radeon HD 7870||$200||GeForce GTX 660|
Meet the GeForce GTX 760
As we mentioned earlier, due to its 170W TDP the GTX 760 is essentially a drop-in replacement for the GTX 670 in board designs. Even NVIDIA’s reference design reflects this, with the reference GTX 760 being based on the reference GTX 670’s cooler and PCB. Consequently there’s not much to say about the GTX 760’s design that we haven’t said before. If you’ve seen a GTX 670 then you’ve seen a GTX 760, down to the last screw. For consistency we’ll quickly go over the GTX 760’s design, but this is generally a rehash of what we’ve already said about the GTX 670.
Like GTX 670 before it, GTX 760 is another “small” GK104 design. The complete card is 9.5” long, however the actual PCB is far shorter at only 6.75” long, 3.75” shorter than the GTX 770’s PCB. In fact it would be fair to say that rather than strapping a cooler onto a card, NVIDIA strapped a card onto a cooler. Meanwhile we see that once again NVIDIA is using Hynix R0C (6GHz) GDDR5 RAM here.
The cooler itself is fairly simple, utilizing an aluminum heatsink melded with a copper base plate. A separate heatsink covers the VRM circuitry – once again mounted at the front of the card – while the fan hangs off of the rear of the card. This unfortunately is the same fan as on the GTX 670, which means it makes a faint grinding sound at low RPMs. This design has proven more than adequate for the GTX 670s, though it does leave some room for NVIDIA’s partners to improve on overall cooling performance, especially with open air coolers.
Elsewhere at the top of the card we’ll find the PCIe power sockets and SLI connectors. Two 6-pin PCIe sockets provide the necessary external power, while a pair of SLI bridge connectors at the other end allow for up to 3-way SLI. NVIDIA’s standard I/O configuration also makes an appearance here, offering 2x DL-DVI, 1x HDMI, 1x DisplayPort 1.2.
Looking at clockspeed bins for a moment, it comes as no great surprise that like the GTX 770, the GTX 760 operates at 1.2v at its highest stock clockspeed bin. The higher voltage versus the 1.76v limit on the 600 series causes power consumption to jump up more rapidly at the highest bins, but is necessary to unlock the 1100MHz+ clockspeeds the card is capable of.
|GeForce Clockspeed Bins|
|Clockspeed||GTX 770||GTX 760|
Finaly, for overclockers there is one further bin (1162MHz) available via overvolting to 1.212v. The maximum power target meanwhile is 115%, for a final power limit of 195W.
The Last of the 700 Series & The Test
In something of an unusual move, along with announcing the GTX 760, NVIDIA also laid out their desktop video card plans for the next several months, leading through the fall. Typically NVIDIA doesn’t announce the non-existence of something, but then again since they apparently have nothing left to hide, there’s little reason not to.
In any case, the GeForce GTX 760 will be the last desktop 700 series card for the next several months. NVIDIA will not be introducing any further desktop cards according to the schedule they’ve provided us, so their lineup will be stable from here on. The 700 series allowed NVIDIA to introduce GK110 based cards and refresh their GK104 based cards, while GK107 and GK106 based products will not be changing. This means that rather than introducing a GTX 750 for example for GK106, NVIDIA will simply keep the top GK106 as the GTX 660.
This move is admittedly a bit weird for how NVIDIA normally does things, as with Fermi they updated their lineups top-to-bottom. Whether this means NVIDIA is planning a late update based on new chips – ala the GeForce GT 200 series, NVIDIA’s 40nm pipe cleaner – or if they simply don’t see a need to roll out new product numbers remains to be seen. But since NVIDIA has added GK106 parts as recently as March, and their top GK106 part doesn’t leave them much room for growth, there’s also a lack of technical opportunity to refresh the rest of their lineup like there was for their GK104 parts. Then again, AMD hasn’t bumped up the series number of their competing retail parts, so there’s little incentive (for once) to play number games in retail.
In any case the current lineup is most likely what we’ll be looking at through the rest of the year, until Maxwell sometime in 2014. This will leave the GTX 760 as NVIDIA’s top 1080p card, while the GK106 based GTX 660 will remain as NVIDIA’s more budget oriented 1080p card.
The press drivers for the GTX 760 are 320.39, a further bug fix of the existing R319 series drivers that also add support for the GTX 760. On the AMD side we’re using a mix of Catalyst 13.5 (7970) and Catalyst 13.6 (7950, 7870).
For comparison purposes we’ve also dug up a few older cards. Naming aside, NVIDIA’s GTX 560 Ti was their last $250 card and the class of card most 2 year cycle buyers will be coming from. Meanwhile we’re also including AMD’s Radeon HD 6870 and NVIDIA’s GTX 460 1GB. Finally, we’re going to include both the Radeon HD 7950 and 7950 Boost in our charts. The Boost edition has largely supplanted the original in retail, but frustratingly there are still some non-Boost (or otherwise sub-850MHz) cards on the market, so this covers both scenarios.
|CPU:||Intel Core i7-3960X @ 4.3GHz|
|Motherboard:||EVGA X79 SLI|
|Power Supply:||Antec True Power Quattro 1200|
|Hard Disk:||Samsung 470 (256GB)|
|Memory:||G.Skill Ripjaws DDR3-1867 4 x 4GB (8-10-9-26)|
|Case:||Thermaltake Spedo Advance|
AMD Radeon HD 7970
AMD Radeon HD 7950 Boost
AMD Radeon HD 7950
AMD Radeon HD 7870
AMD Radeon HD 6870
NVIDIA GeForce GTX 770
NVIDIA GeForce GTX 760
NVIDIA GeForce GTX 670
NVIDIA GeForce GTX 660 Ti
NVIDIA GeForce GTX 660
NVIIDA GeForce GTX 560 Ti
NVIDIA GeForce GTX 460 1GB
NVIDIA ForceWare 320.18
NVIDIA ForceWare 320.39
AMD Catalyst 13.5 Beta 2
AMD Catalyst 13.6 Beta 2
|OS:||Windows 8 Pro|
As always, starting off our benchmark collection is our racing benchmark, DiRT: Showdown. DiRT: Showdown is based on the latest iteration of Codemasters’ EGO engine, which has continually evolved over the years to add more advanced rendering features. It was one of the first games to implement tessellation, and also one of the first games to implement a DirectCompute based forward-rendering compatible lighting system. At the same time as Codemasters is by far the most prevalent PC racing developers, it’s also a good proxy for some of the other racing games on the market like F1 and GRID.
Though NVIDIA’s goal is to surpass the 7950 – something they generally accomplish – in some titles AMD’s parts still hold such an innate advantage as to hold off the GTX 760, and this is one such case. Performance of the GTX 760 trails the 7950 by 18% here, with the GTX 760 falling just short of averaging 60fps at 1920 with advanced lighting. But as we’ve seen with other reviews, this is the only game where AMD holds an advantage quite this great.
It’s interesting to note though that this is also one of the only games the GTX 760 struggles to beat the GTX 660 Ti, owing to the heavy shader load placed on these cards by the advanced lighting system. The two are essentially tied with that lighting turned off, while the GTX 760 takes a small lead with it off.
Total War 2: Shogun
Our next benchmark is Shogun 2, which is a continuing favorite to our benchmark suite. Total War: Shogun 2 is the latest installment of the long-running Total War series of turn based strategy games, and alongside Civilization V is notable for just how many units it can put on a screen at once. Even 2 years after its release it’s still a very punishing game at its highest settings due to the amount of shading and memory those units require.
Moving on to Shogun, we have our first title where the GTX 760 can beat the 7950B. The difference here is about 7%, which is going to be about average for the difference between the two cards. Meanwhile this is also a good example of where the GTX 760 will perform relative to the GTX 660 Ti and GTX 670 in most cases. The GTX 760 ends up coming very close to the GTX 670, leaving the GTX 660 Ti behind by 6%.
Meanwhile looking at our past cards, this is also a fairly typical case for the performance of the GTX 760 over that of the GTX 560 Ti and GTX 460. GTX 760 isn’t going to double GTX 560 Ti’s performance here, but it’s otherwise large enough that it reaches the magnitude we’d expect for a generational leap. In the case of Shogun that’s a 77% performance advantage, while the GTX 760 eclipses the GTX 460 1GB by 138%. In NVIDIA’s product cycle the GTX 760 is the closest replacement for these two cards for buyers in 2-3 year replacement cycles, and it shows.
The third game in our lineup is Hitman: Absolution. The latest game in Square Enix’s stealth-action series, Hitman: Absolution is a DirectX 11 based title that though a bit heavy on the CPU, can give most GPUs a run for their money. Furthermore it has a built-in benchmark, which gives it a level of standardization that fewer and fewer benchmarks possess.
With Hitman the GTX 760 once again lags the 7950B. At 1920 with 4x MSAA this is a 13% gap, though with framerates in the 40s neither card does exceptionally well for an action game here. Even as NVIDIA’s fastest 1080p card, 1080p at High settings is a more realistic option, which pushes the GTX 760 to 70.3fps, while also closing the gap to the 7950B to just 7%.
The minimum framerates in HItman are fairly steady, so although the GTX 760 can only hit 40fps for an average, it bottoms out above 30fps. Still, it trails the 7970B by 10% at our higher 1080p settings.
Another Square Enix game, Sleeping Dogs is one of the few open world games to be released with any kind of benchmark, giving us a unique opportunity to benchmark an open world game. Like most console ports, Sleeping Dogs’ base assets are not extremely demanding, but it makes up for it with its interesting anti-aliasing implementation, a mix of FXAA and SSAA that at its highest settings does an impeccable job of removing jaggies. However by effectively rendering the game world multiple times over, it can also require a very powerful video card to drive these high AA modes.
With Sleeping Dogs the GTX 760 is once again back on top, although this time it’s a close fight between it and the 7950B. In this case the gap at our higher 1080p settings is just 3%, nearly a tie. Meanwhile it’s interesting to see the GTX 760 doing so well compared to the GTX 670 here, even beating it just slightly. The use of SSAA hits the ROPs and shaders pretty hard, so while we’d typically expect the GTX 760 to fall behind the GTX 670 here, this appears to be a case where the higher core clockspeed and resulting higher ROP performance works in the GTX 760’s favor.
With our minimum framerates however the GTX 760 falls behind the 7950B. Here AMD’s competitor performs about 7% better at our highest 1080p settings, keeping AMD’s card above 50fps.
Up next is our legacy title for 2013, Crysis: Warhead. The stand-alone expansion to 2007’s Crysis, at over 4 years old Crysis: Warhead can still beat most systems down. Crysis was intended to be future-looking as far as performance and visual quality goes, and it has clearly achieved that. We’ve only finally reached the point where single-GPU cards have come out that can hit 60fps at 1920 with 4xAA.
With Crysis: Warhead, 60fps at 1080p with everything turned up remains the realm of the GTX 770 and other higher-end cards. The GTX 760 can only muster 49.8fps, which means it would be necessary to at least drop MSAA here to get the GTX 760 above 60fps. This also puts the GTX 760 about 7% behind the 7950B, which enjoys the usual AMD architecture advantage in this game.
On a side note, this is one of those titles that really benefits from ROPs, driving a very large wedge between the GTX 760 and the outgoing GTX 660 Ti. The 20% performance improvement at our highest 1080p settings is definitely atypical, but it’s a good example of where going from 7 SMX + 24 ROPs to 6 SMX + 32 ROPs can pay off handsomely.
Minimum framerates continue the trend we saw with our averages. The GTX 760 continues to trail the 7950B here while significantly outperforming the GTX 660 Ti.
Far Cry 3
The next game in our benchmark suite is Far Cry 3, Ubisoft’s island-jungle action game. A lot like our other jungle game Crysis, Far Cry 3 can be quite tough on GPUs, especially with MSAA and improved alpha-to-coverage checking thrown into the mix. On the other hand it’s still a bit of a pig on the CPU side, and seemingly inexplicably we’ve found that it doesn’t play well with HyperThreading on our testbed, making this the only game we’ve ever had to disable HT for to maximize our framerates.
With Far Cry 3 we’re back to a game where the GTX 760 has a clear lead, surpassing the 7950B by an unexpected 28% at our highest 1080p settings. However 41.5fps is less than idea for an action game like Far Cry 3, so we have to drop the MSAA to take a crack at 60fps. Even then, the GTX 760 leads by 14%.
This also happens to be another great example of the ROP difference between GTX 760 and GTX 660 Ti. With MSAA enabled the GTX 760 takes a clear lead, but without MSAA the two cards are practically tied.
Meanwhile Far Cry 3 is also another strong example of the performance gap between the GTX 760 and cards 2-3 years old. GTX 760 leaves those older cards in the dust by anywhere between 77% for the GTX 560 Ti and 6870, up to 138% for the GTX 460 1GB. CPU performance may have plateaued over the last half-decade, but thanks to the embarrassingly parallel nature of graphics, we continue to see massive gains over a whole generation.
Our major multiplayer action game of our benchmark suite is Battlefield 3, DICE’s 2011 multiplayer military shooter. Its ability to pose a significant challenge to GPUs has been dulled some by time and drivers, but it’s still a challenge if you want to hit the highest settings at the highest resolutions at the highest anti-aliasing levels. Furthermore while we can crack 60fps in single player mode, our rule of thumb here is that multiplayer framerates will dip to half our single player framerates, so hitting high framerates here may not be high enough.
Battlefield 3 has continued to favor NVIDIA parts and the GTX 760 is no exception. The gap at 1080p with MSAA is 20%, one of the largest leads for the GTX 760 out of all of our games. Even the full-fledged 7970 is still slower than the GTX 760 here by a few frames per second.
Meanwhile it’s interesting to note that this is another title that really favors the ROP performance advantage of the GTX 760, with the GTX 760 once more shooting well ahead of the GTX 660 Ti, coming within a few frames per second of the GTX 670. How close these cards depends on the game – as we’ve seen it’s anywhere between equal to a GTX 660 Ti to equal to a GTX 670 – but this is a fairly typical example of the GTX 760 giving the much more expensive GTX 670 a run for its money.
Finally, looking at our last generation cards the GTX 760 once again cements its position as a solid generational upgrade. At 1080p without MSAA we’re looking at performance 80% better than a GTX 560 Ti, and more than doubling the 6870 and GTX 460 1GB.
Our other strategy game, Civilization V, gives us an interesting look at things that other RTSes cannot match, with a much weaker focus on shading in the game world and a much greater focus on creating the geometry needed to bring such a world to life. In doing so it uses a slew of DirectX 11 technologies, including tessellation for said geometry, driver command lists for reducing CPU overhead, and compute shaders for on-the-fly texture decompression.
Civilization V is another title that sees NVIDIA typically do well, with the GTX 760 surpassing the 7950B by 8% regardless of the resolution. This also happens to be one of the few games where even the GTX 760 can hit 60fps at 2560.
Bioshock Infinite is Irrational Games’ latest entry in the Bioshock franchise. Though it’s based on Unreal Engine 3 – making it our obligatory UE3 game – Irrational had added a number of effects that make the game rather GPU-intensive on its highest settings. As an added bonus it includes a built-in benchmark composed of several scenes, a rarity for UE3 engine games, so we can easily get a good representation of what Bioshock’s performance is like.
The 9th benchmark in our suite, our performance data is becoming increasingly consistent. Once more the GTX 760 surpasses the 7950B, this time by 7% at our highest 1080p settings, with both cards offering enough punch to pass 60fps. Meanwhile it’s not quite as close to the GTX 670 as it is in some other games, but nevertheless it’s another case of where the closest 600 series card is the GTX 670 and not the GTX 660 Ti.
Our final benchmark in our suite needs no introduction. With Crysis 3, Crytek has gone back to trying to kill computers, taking back the “most punishing game” title in our benchmark suite. Only in a handful of setups can we even run Crysis 3 at its highest (Very High) settings, and that’s still without AA. Crysis 1 was an excellent template for the kind of performance required to drive games for the next few years, and Crysis 3 looks to be much the same for 2013.
Our final benchmark goes out with another solid victory for the GTX 760. 1080p at Very High quality isn’t going to be a great experience – this is Crysis, after all – but 1080p at High quality will see the GTX 760 average just shy of 60fps, 14% ahead of the 7950B. Also once again the GTX 760 ends up being very close to the GTX 670, trailing it by just 2%.
Meanwhile taking one last look at our older cards, the GTX 760 is decidedly average, outperforming the GTX 560 Ti by 68% here. The gap over the GTX 460 is larger yet, at 113%.
As always we’ll also take a quick look at synthetic performance, though as GTX 760 is just another GK104 card, there shouldn't be any surprises here. These tests are mostly for comparing cards from within a manufacturer, as opposed to directly comparing AMD and NVIDIA cards. We’ll start with 3DMark Vantage’s Pixel Fill test.
The 3DMark Pixel Fill test confirms what we’ve seen in or earlier results, with the full 32 ROP/256-bit configuration of the GTX 760 giving the card a notable advantage in this ROP and memory bandwidth bound scenario. It also validates why the GTX 670 and GTX 760 are so close in these scenarios.
Moving on, we have our 3DMark Vantage texture fillrate test, which does for texels and texture mapping units what the previous test does for ROPs.
The texel test on the other hand correctly points out the worst case scenario. In purely shader and texture bound scenarios, even with its higher clockspeeds the GTX 760 still can’t make up for the lack of an SMX relative to the GTX 660 Ti and GTX 670. Thankfully as we’ve seen in our game benchmarks performance is being dictated more often by the ROP improvement than the loss of the SMX.
Finally we’ll take a quick look at tessellation performance with TessMark.
Tessmark is surprisingly consistent, with the GTX 670, GTX 660 Ti, and GTX 760 all in a virtual tie. GTX 760 lacks the Polymorph Engine that comes with the 7th SMX, but in this test the lack of tessellation performance from that loss is getting completely offset by the higher clockspeeds of the GTX 760.
Jumping into compute, we aren’t expecting too much here. Outside of DirectCompute GK104 is generally a poor compute GPU, and the loss of an SMX relative to the GTX 660 Ti isn’t doing the GTX 760 any favors here. By all appearances the GTX 760 is even more of a pure gaming card than the GTX 660 Ti was.
As always we'll start with our DirectCompute game example, Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. While DirectCompute is used in many games, this is one of the only games with a benchmark that can isolate the use of DirectCompute and its resulting performance.
Civilization V once more validates that NVIDIA’s DirectCompute performance is generally up to snuff in this case. The fact that the GTX 760 is ahead of the GTX 660 Ti by any degree took us by surprise at first, but we’re likely looking at a scenario where the wider memory bus and/or larger L2 cache of GTX 760 offset some of the general compute gap.
Our next benchmark is LuxMark2.0, the official benchmark of SmallLuxGPU 2.0. SmallLuxGPU is an OpenCL accelerated ray tracer that is part of the larger LuxRender suite. Ray tracing has become a stronghold for GPUs in recent years as ray tracing maps well to GPU pipelines, allowing artists to render scenes much more quickly than with CPUs alone.
Luxmark is entirely about compute performance, and as a result this is an exceptionally poor showing for the GTX 760, with the GTX 660 Ti having no trouble besting it.
Our 3rd benchmark set comes from CLBenchmark 1.1. CLBenchmark contains a number of subtests; we’re focusing on the most practical of them, the computer vision test and the fluid simulation test. The former being a useful proxy for computer imaging tasks where systems are required to parse images and identify features (e.g. humans), while fluid simulations are common in professional graphics work and games alike.
Breaking down our CLBenchmark results, the computer vision test has frequently favored raw clockspeed over total shader throughput, which gives the GTX 760 an interesting advantage here. It’s capable of easily leaving the GTX 660 Ti in the dust and even edge out the GTX 670. Of course this is still less than 2/3rds the performance of even the slowest AMD GCN card, reflecting AMD’s superior computer performance.
The fluid simulation is especially brutal in that regard. Once again shifting back to an almost complete reliance on shader throughput, GTX 760 slightly trails GTX 660 Ti, never mind the nearly three-fold difference between it and the 7950B.
Moving on, our 4th compute benchmark is FAHBench, the official Folding @ Home benchmark. Folding @ Home is the popular Stanford-backed research and distributed computing initiative that has work distributed to millions of volunteer computers over the internet, each of which is responsible for a tiny slice of a protein folding simulation. FAHBench can test both single precision and double precision floating point performance, with single precision being the most useful metric for most consumer cards due to their low double precision performance. Each precision has two modes, explicit and implicit, the difference being whether water atoms are included in the simulation, which adds quite a bit of work and overhead. This is another OpenCL test, as Folding @ Home has moved exclusively to OpenCL this year with FAHCore 17.
Unlike some of our other compute benchmarks, the GTX 760 doesn’t fare too poorly here when it comes to single precision. However it’s still notably behind the 7950B in this case. And with double precision it’s no contest.
Wrapping things up, our final compute benchmark is an in-house project developed by our very own Dr. Ian Cutress. SystemCompute is our first C++ AMP benchmark, utilizing Microsoft’s simple C++ extensions to allow the easy use of GPU computing in C++ programs. SystemCompute in turn is a collection of benchmarks for several different fundamental compute algorithms, as described in this previous article, with the final score represented in points. DirectCompute is the compute backend for C++ AMP on Windows, so this forms our other DirectCompute test.
As another compute throughput bound benchmark, the GTX 760 is essentially tied with the GTX 660 Ti. This benchmark is somewhat memory bandwidth sensitive, which is why the GTX 760 doesn’t outright lose to the GTX 660 Ti here.
Power, Temperature, & Noise
As always, last but not least is our look at power, temperature, and noise. Next to price and performance of course, these are some of the most important aspects of a GPU, due in large part to the impact of noise. All things considered, a loud card is undesirable unless there’s a sufficiently good reason – or sufficiently good performance – to ignore the noise.
With both the 600 series GK104 parts and the GK104 based GTX 770 as a backdrop here, we’re not expecting any great surprises here. Power consumption is going to approach the GTX 670 thanks to GPU Boost 2.0 and the identical TDPs. And since the cooler is also the same, acoustics should also be similar. Only idle power should show any real variation.
|GeForce GTX 760 Voltages|
|GTX 770 Max Boost||GTX 760 Max Boost||GTX 770 Idle|
Like the GTX 770, GTX 760 is spec’d to a stock top voltage of 1.2v. This buys NVIDIA the 1100MHz+ clockspeeds we’re seeing, but it drives up power consumption.
|GeForce GTX 760 Average Clockspeeds|
|Max Boost Clock||1149MHz|
|Far Cry 3||
Looking at the average clockspeeds among our games, despite the official 1033MHz boost clock for GTX 760 always boosts to an average over 1100MHz, with most games boosting to around 1130MHz. The limiting factor here is typically temperatures, which quickly hit 80C at the highest boost bin and its 1.2v operating voltage. A card with a better cooler would likely have no trouble sustaining 1149MHz here.
As we saw with the GTX 770 a few weeks ago, NVIDIA’s idle power consumption as measured at the wall has come down a couple of watts compared to the older GK104 cards. With a lower idle clockspeed compared to the GTX 660 Ti and its ilk, NVIDIA has been able to shave off that wattage without any more substaintial changes. However I continue to be surprised that we’re seeing any difference between the GTX 760 and the 7950B, as AMD is no slouch here either.
Moving on to load power under Battlefield 3, our results validate our earlier theories on power consumption based on the spec sheets. Wall power with a GTX 760 is 6W higher than GTX 670 in this test, and 18W higher than with a GTX 660 Ti. There’s no getting around the fact that GTX 670-like performance requires GTX 670-like power consumption.
At the same time the power consumption of the GTX 760 is almost identical to the 7950B at the wall. Though given the 7950B’s lower performance in this benchmark, what we’re actually seeing is lower CPU power consumption offsetting the 7950B’s higher power consumption.
With Furmark the story is fairly similar for the NVIDIA cards, all of which come close together due to their similar TDPs. Meanwhile the 7950B pulls about 30W more at the wall, in-line with its 200W TDP. NVIDIA edged out AMD on power efficiency with the 600 series, and we’re going to see the same thing on the 700 series.
Based on the same cooler and GPU as the GTX 670, the GTX 760 offers no grand revelations here. 31C is standard for this type of blower cooler.
Even though the GTX 760 and GTX 670 have identical TDPs, the presence of GPU Boost 2.0 means that the GTX 760 operates nearer its TDP more often, and in this case the 80C throttle imposed by GPU Boost makes itself felt. 81C does end up being the worst among the cards in this collection, but it’s a reasonable temperature for a GK104 GPU, to the point where in temperature limited scenarios this is exactly where it’s designed to top out at.
With FurMark our results are much the same as with BF3. The slower response time of the fans on the GTX 770 and GTX 760 mean that both cards top out at 82C before temperatures level out, a few degrees warmer than with the GTX 670 and GTX 660 Ti. Alternatively the 7950B and its more powerful fan keep temperatures at or below 75C for any workload we throw at it.
Since NVIDIA used the same cooler/fan assembly as on the GTX 670, the GTX 760 is going to give us similar acoustics. Unfortunately that includes the characteristic faint grinding noise that we’re previously mentioned. Idle noises as a result are respectable, but GTX 770 this is not.
Moving to fan noise under load, here we can see NVIDIA’s default fan curve and 80C throttle once again pay off. The GTX 760 hits 50dB under BF3, which for this specific cooler is actually doing a bit better than average, as evidenced by the slight advantage the GTX 760 has over the GTX 670. The GTX 770 and Fermi cards are something of spoilers here though, a reminder of what the Titan cooler and open air cooling can do. We will have some custom GTX 760s next week, so it will be interesting to see how much better these cards are when it comes to noise. 50dB is by no means bad, especially for the performance, but at this point we’ve seen lower noise levels out of cards with similar price tags and TDPs.
Finally, FurMark once again paints a picture very similar to what we saw with Battlefield 3. For NVIDIA cards this is middle of the road, while at 5dB quieter than the reference 7950B it’s a significant improvement.
Overclocking GTX 760
Like the GTX 770 last month, the GTX 760 is clocked relatively high. Its 980MHz base clock is 65MHz higher than either the GTX 670 or GTX 660 Ti. As a result some of the overclocking headroom that we would expect has been eaten into by the overclock. Yet on the other hand the higher voltage for the highest boost bin – 1.2v versus 1.175v – gives us some additional overclocking headroom as long as our cooling can keep up.
At the same time the GTX 760 gets a slightly larger TDP adjustment window than most other GK104 parts. Here we can push the TDP up by up to 15%, from 170W to 195W. This is actually more than the GTX 770, so in TDP limited scenarios we should have a slightly easier time hitting the higher boost bins.
|GeForce GTX 760 Overclocking|
|Max Boost Clock||1149MHz||1306MHz|
In practice GTX 760 ended up exceeding our expectations. In fact it even ended up exceeding GTX 770 with regard to the size of the overclock and the highest boost bin we reached. Altogether we were able to increase the core clock by 150MHz on our GTX 760 sample, going from a base clock of 980MHz to 1130MHz. At the same time our max boost clock went from 1149MHz to 1306MHz, making this the first Kepler we’ve seen to surpass 1300MHz. The end result is that core overclocking ends up being more potent than we were expecting.
Meanwhile the memory overclock is fairly standard for a GK104 part. It’s something of a roll of the dice, and in this case we were able to hit 6.5GHz before memory performance started regressing. Thus we’re altogether looking at a 15% core overclock coupled with a milder 8% memory overclock.
Given GTX 760’s base shader/texture performance deficit due to only having 6 SMXes, overclocking proves to be very effective on our final results. The performance gains in all 5 of our games were on the order of 12% to 14%, just a bit less than the core overclock itself. With the wide gap between the GTX 770 and GTX 760 in terms of specs this generally isn’t enough to completely catch up to NVIDIA’s top-tier GK104 card, but it can erase a large portion of the gap.
The end result of this overclock won’t dramatically change the GTX 760’s performance profile, but it should help to push it over 60fps in those situations where performance at 1080p at ultra quality settings was marginal. Or from a fixed performance standpoint, we’re approaching Radeon HD 7970 performance with this overclock.
Moving on to power consumption, increasing our TDP has the expected hit. GTX 760 of course has a larger adjustment window than GTX 770, so the increase in power consumption ends up being noticeably larger. Overclocking in this manner pushes the GTX 760 off of the power/performance efficiency curve to some degree, so the GTX 760 loses its edge in that respect.
The end result is that power consumption at the wall jumps by 40W under BF3, and 50W under FurMark. This is tied with and better than the stock GTX 770 respectively, though as we’ve already seen it doesn’t match the GTX 770’s rendering performance. Furthermore as we’ll see, cooling is going to play a big factor here, with the reference GTX 760 cooler not always being the best choice for overclocking.
With the temperature limit raised to 95C, the reference GTX 760 cooler reaches equilibrium at 85C under BF3 and 87C under FurMark. These temperatures are still easily within spec for GK104, but they’re definitely starting to crawl up here. Meanwhile, though it’s not directly measurable the impact of leakage is certainly making itself felt here. If we had cooler temperatures power consumption would at least be marginally lower.
Finally we noise we can see that those new equilibriums also come at new, much higher noise levels. The reference cooler can keep up with the higher heat load created by overclocking, but it’s having to work hard to do so. These noise levels aren’t terrible – in fact they’re about even with the reference 7950 – but they do represent a real cost in terms of noise. Consequently, unlike NVIDIA’s Titan cooler that comes with the reference Titan, GTX 780, and GTX 770, when it comes to GTX 760 there’s very clearly room for improvement with custom coolers.
Bringing things to a close, in the last month NVIDIA has launched three different video cards, carving out the GeForce GTX 700 series. As the final and cheapest card in that launch window, GTX 760 is going to be the most affordable and highest volume card, and also the card that that will face the most competition from AMD. By launching a refresh card at a time when AMD is going to be sitting it out, NVIDIA essentially gets to dictate in what environment their products will launch and what their competition will be. NVIDIA doesn’t get to rewrite the laws of physics and is ultimately beholden to GPU clockspeeds, power consumption, and yields like anyone else, but they can still exercise a great deal of control through the clockspeeds and prices they set.
To that end this launch is a great deal like the GTX 770 launch last month, with NVIDIA improving performance, lowering prices, and putting AMD on the defensive all at the same time. Thanks to these performance improvements and price cuts, the GTX 760 ends up coming within 3% of the soon to be retired GTX 670 and easily surpasses the GTX 660 Ti, all the while coming in at a price well below both at $249. Like most mid-cycle upgrades this is more about bringing existing performance levels down to new prices, and to that end NVIDIA has delivered on those goals. Ultimately it’s not a new level of performance, but it’s a new price for what a few months ago would cost $350 or more.
With that said, like any good refresh the presence of the 700 series and the retirement of the 600 series looks to shake up the market, and once more AMD is going to be on the receiving end here. Rather unlike the GTX 770 versus the 7970 GHz Edition, the GTX 760 is not tied with any AMD product. At 1080p it is clearly ahead of both the stock and boost versions of the 7950, by 13% and 8% respectively. This is by no means a commanding lead and AMD still offers better performance in some cases, but on average the GTX 760 is faster, quieter, and $30-$50 cheaper than AMD’s closest competitor.
As a result the competitive landscape is clearly in NVIDIA’s favor for the time being. AMD has their Never Settle Reloaded bundle to boost the value of the 7950, and if this was a repeat of the GTX 660 Ti launch – where the two cards were tied – then that strategy would be solid. Ultimately with such a large game bundle only the individual buyer can truly assign a value to AMD’s bundle, but in this case we believe AMD can’t afford to be slower and more expensive at the same time. At current prices NVIDIA’s GTX 760 has AMD beat, in essence repeating the GTX 670 launch by once more undercutting the 7950.
Wrapping things up, having established the GTX 760’s current control of the $250 price point let’s talk about the wider market for the GTX 760. As a mid-cycle refresh the performance gains over the 600 series won’t knock anyone’s socks off, but then like most mid-cycle refreshes this isn’t a product targeted at existing 600 series owners. Rather this is targeted at buyers looking to upgrade their older 55nm/40nm generation video cards, or with the recent launch of Haswell, putting together a new system outright.
With a $249 price tag the GTX 760 is most straightforward successor to enthusiast cards like the GTX 560 Ti and GTX 460 1GB. In the case of the former, now one full cycle old, the performance gains are solid, with GTX 760 improving on the GTX 560 Ti by about 67%. This isn’t exceptional by any means (the GTX 570 to GTX 770 was 75%) but it’s about average for a 2 year (generational) improvement. Otherwise for a true doubling we’ll have to wait for one more year, as evidenced by the better than 100% performance gains over the 3 year old GTX 460 1GB.