660 Ti graphics card: which is best?
We test the Radeon HD 7870 against nVidia's GeForce 660 and 660 Ti graphics boards. Read on for 7870 v 660/660 comparative benchmarks and specs, as well as a list of the graphics cards commercially available. See also: Group test: what's the best graphics card?
Radeon HD 7870 or GeForce 660? That's a difficult choice, but we hope we'll be able to show you how the cards measure up, and where they fall down. We'll also explain why the 660 and the 660 Ti are actually two very different cards - albeit ones with extremely similar names. We'll be comparing the specs, and putting up some figures for you. So sit back and let us take you through the perilous landscape that marks the battleground between 7870 and 660. See also: Group test: what's the best budget graphics card?
7870 v 660: Ti or Not Ti
The 660 is not one but two different cards, we hear you say? Well, yes. The 660 and the 660 Ti, to be precise. And if you're thinking that means the 660 Ti is simply an overclocked 660, then you'd be wrong - very very wrong, as it turns out. As ever with graphics cards, the naming of the 660 and 660 Ti isn't as simple as it should be.
The addition of 'Ti' (short for 'Titanium', but really just a way of saying that the Ti version is enhanced) seems to suggest that the 660 Ti is a slightly improved version of the 660. In fact, while both the 660 and 660 Ti are members of nVidia's Kepler family (the successor to Fermi), they both use different chips to one another.
The 660 Ti is built around the same GK104 that forms the foundation of a number of other products, from the 670 through to the 680 and 690. The standard 660, on the other hand, employs the GK106. The main difference between the GK104 and GK106 is that the GK104 is a much bigger chip. The 660 Ti, for example, is formed from 3.54 billion transistors, rather than the 2.54 billion of the standard 660.
This was done so that nVidia could use the same chip across the 660, 670 and 680 ranges, but turn off certain parts in order to create the lower-specified and cheaper poducts like the 660 Ti. However, while the 660 Ti can't access its extra capabilities, it still has to carry much of the bulk, making it a slightly wasteful design. So nVidia decided to cut costs by creating a new mid-range GK106. The 660, effectively, is a fully-featured version of the GK106, with no parts turned off. And because there's no excess on it, its transistor count is accordingly much lower than on the GK104-powered 660 Ti.
Because the GK106 has a smaller and more efficient design, the standard 660 can run hotter and faster, and can be pushed up to higher speeds. The standard 660 offers a default core clock of 980MHz, with a potential Boost to 1,033MHz. You might expect the more expensive and powerful 660 Ti to boast higher clock speeds, but its larger design means it has to make do with the relatively meagre default figure of 915MHz, with a potential boost to 980MHz.
So, in a bizarre twist, the unboosted figure on the 660 is just as high as the boosted figure of the higher-end 660 Ti! The manufacturers of graphics cards mostly improve on these 'factory settings', but the 660's advantage remains. So even fast low-priced versions of the 660 Ti have a boost up to 1,019MHz, while the standard 660 sees as much as 1,111MHz. When you look for the 660s and 660 Tis with the very highest clock speeds of all, the 660 is still 26MHz higher than the fastest 660 Ti. See also: all components buying advice.
7870 v 660: The 660 Ti Fights Back
That, though, is very far from the end of the matter. That's because of the ways in which the 660 and 660 Ti otherwise vary. Despite the standard 660 being (technically) a fully-featured version of the GK106, those facilities are still more limited than on the 660 Ti. So while the standard 660 comes with five Streaming Multiprocessors (SMXes), the 660 Ti uses a far healthier seven of them. This allows the Ti to boast 40% extra when it comes to Stream Processors (1,344 rather than 960 on the standard 660) and Texture Units (112 rather than 80), for instance.
And those texture units become pretty important when it comes to raw power. Even at the default factory settings, the standard 660 offers a texture fill-rate of just 82.6GTps. The 660 Ti, though, uses its 112 texture units to increase this figure to 109.8GTps. That's despite the fact that its Boost core clock speed of 980MHz is some 53MHz behind that of the standard 660. Indeed, if we take the most overclocked 660 of all (the 1,137MHz EVGA 660 FTW), it produces a texture fill-rate of just under 91GTps - a long way short of the 109.8GTps turned out by the 660 Ti at its worst. And even a fairly low-priced version of the 660 Ti, such as the MSI, can push the Boost clock speed up to 1,019MHz - producing a fill-rate of 114.1GTps.
In many other respects, the 660 and 660 Ti are very similar. They both have standard memory clocks of 1,502MHz - the GDDR RAM increases this to an effective 6,008MHz. The memory interfaces are 192bit - as we'll see, that's inferior to the HD 7870's - and produce identical memory bandwidth figures of 144GBps.
7870 v 660: Enter the Radeon HD 7870
So far, the competition has been very friendly. But let's up the stakes a bit, and open the competition to nVidia's big rival, AMD. Since taking over the graphics chip specialists ATI, AMD has found itself in regular battle against nVidia. Many of the recent fights have gone the way of nVidia, with AMD falling a little behind on the development cycle.
In truth, the graphics card market hasn't made any great leaps for a significant amount of time, but nVidia's technology seems to have had a slight advantage for the last two years. That general overview, though, does nothing to explain whether the 660 and 660 Ti are superior to their natural competitor, the Radeon HD 7870.
The 7870, or 'Pitcairn', as it's called, was always designed as a mid-range product that would sit between the tooth-crunchingly powerful 7900 cards, and the more modest 7700 range. As such, it's a perfect competitor to the 660 and 660 Ti. Like the two nVidias, the 7870 is built using the efficient 28nm manufacturing process. It consists of 2.8 billion transistors, placing it some way in the middle of the 2.54 billion 660, and the 3.54 billion 660 Ti - albeit a little closer to the 660 than the Ti.
The AMD's architecture isn't directly comparable to that of the 660s, so it's dangerous to take differences in figures at face value. The 7870 does, though, have 22 of its Compute Units turned on, giving it 1,280 stream processors and 80 texture units.
This means that it falls only slightly behind the 660 Ti on stream processors - while remaining ahead of the 660 by 320. It fares rather less well on those texture units, though, only matching the lowly 660. The raw texture fill-rates of the 660s suggested that the number of texture units is more important than clock speeds, so will the 7870 struggle here?
And so on to clock speeds. As we have seen already, a smaller amount of transistors makes it easier to run at higher figures. That truth is proven again, with the 7870 offering a standard core clock speed of 1GHz. There's no Boost feature, so that 1GHz is essentially your figure. Even taking into account Boost, though, the Radeon still manages to beat the default clock setting of the 660 Ti by 20MHz, while falling some 33MHz behind the standard 660. Throw the texture units into the mix, though, and the 7870 registers a fill-rate of just 80GTps, falling behind even the 82.6GTps of the standard 660. It gets decimated by the 660 Ti's figure of 109.8GTps.
There is better news in terms of memory capabilities. That's not, though, down to the memory clock rates themselves. The 1.2GHz turns into an effective figure of 4.8GHz (taking into account the quadruping effect of the GDDR5 RAM), leaving it trailing both of the 660s by over 1.2GHz. (AMD may have done much to bring GDDR5 RAM to market, but it's nVidia that has ultimately squeezed the highest performance from the memory.) However, while both of the 660s are saddled by a 192bit memory interface, the 7870 has a much wider 256bit version, giving it an awful lot more elbow room.
Even with the lower memory clock rates, the 7870 turns out a very impressive 153.6GBps on memory bandwidth - beating both of the 660s by 9.6GBps. Use a version like the heavily overclocked Sapphire HD 7870, and you can get this up to 160GBps.
The 7870 hits home harder in other areas as well, delivering 32 Raster Operations to the 24 of the 660 and 660 Ti. It's also a better card for double precision, with its FP64 performance measuring just 1/16 of its FP32 performance - as opposed to 1/24 in the case of the 660s. None of those cards have good Compute facilities, although, in truth, such facilities are unimportant when it comes to home gaming. The 7870 is capable of driving more displays, with Eyefinity allowing for up to six displays, as opposed to the four of the 660s.
Having said that, a wide range of nVidia cards will support the maximum of four, whereas running running six displays from the ATI will require considerable research and a very careful choice of card. PhysX is another plus for the 660s, although very few games have good support for PhysX, making it something of a niche feature - if you don't know whether or not you'll miss it, then you almost certainly don't need it.
The 660 is the lowest on power, coming with a TDP of 140 watts. The 660 Ti, however, is surprisingly efficient, and has a TDP of just 150 watts, some way from the 7870's comparatively wasteful 175 watts. The 660 Ti is the outlier here, although that's perhaps not surprising given the Ti's fairly small clock speeds. In fact, there's plenty of potential to push the Ti further, making it a card that, with its default settings, is very economical.
7870 v 660: Benchmark Tests
No matter what the specifications say, the only thing that really matters is how well the graphics cards fare on real-world games. We've compared the 660, 660 Ti and 7870 at default settings, and across five different titles and three resolutions. We've also included figures for a 2GB version of the 760 (around £200), so that you have some sort of reference point.
7870 v 660: Crysis 3
The 660 Ti starts off with a pleasing victory at the lowest resolution of 1680x1,050, marching 4.1fps ahead of the standard 660. The 7870 falls some distance behind, finishhing a further 2.2fps back. The gap gets smaller over the larger resolutions, and by 2,560x1,600, has become just 2.8fps between the three cards. In the most typical resolution for these cards, though, the Ti wins handsomely.
7870 v 660: Bioshock Infinite Rage
The 660 Ti is even more decisive here, leading the 660 and 7870 by 6.7fps and 11.3fps respectively. Again the gap closes up at higher resolutions, although it's still almost 7fps at the top resolution.
7870 v 660: Stalker: Call of Pripyat
A rare case of a game that doesn't show the 7870 behind. Indeed, it finishes ahead of the standard 660 in all three resolutions, and at 2,560x1,536, the gap is as small as 0.8fps. Even at the more realistic resolution of 1,680x1,050, it's a mere 3.1fps.
7870 v 660: Battlefield 3
Normal service is resumed. Indeed, the 660 Ti opens up a very sizeable lead of 12.9fps over the 7870, and 6.4fps over the 660. The gap once again closes significantly at higher resolutions. Nonetheless, mainstream games will show good results with the 660 Ti.
7870 v 660: Assassin's Creed 3
Another very solid victory for the 660 Ti, beating the 7870 by 8.2fps, and the 660 by 3.4fps. Again, the gap has dropped to 4.1fps by the time we get to the top resolution.
|Crysis 3||Crysis 3||Crysis 3|
|Bioshock Infinite Rage||Bioshock Infinite Rage||Bioshock Infinite Rage|
|Stalker - Pripyat||Stalker||Stalker|
|Battlefield 3||Battlefield 3||Battlefield 3|
|Assassin's Creed 3||Assassin's Creed 3||Assassin's Creed 3|
7870 v 660: Verdict
The games tests are clear - the 660 Ti is the best performer here. The 660 is also very much superior to the 7870, and roughly halfway between that card and the 660 Ti. Of course, it would be unfair not to point out that the 7870 can be bought for less. We found some very good MSI and XFX versions available for £150-£161. The 660 Ti, on the other hand, costs between £175-£180. Nonetheless, this card does offer very much better frame rates. 8-12fps may not sound very much, but it often amounts to around 15% extra performance. That'll make for a much smoother experience, and allow you to raise resolutions or detail levels. For the extra £30 or so, we believe the 660 Ti to be worth the cash. It's also a better deal than the standard 660, offering a significant bump in return for an extra £10-£20. Of course, it comes down partly to how important money is to you. For our cash, though, the 660 Ti is an excellent buy.
Next page: the best 7870 and 660 cards you can buy >>