nVidia GTX 760 v Radeon HD 7950: different technologies differing concepts
When the 7950 was launched, back in 2012, it was at the vanguard of a new rethinking of AMD’s strategy. The company had achieved considerable success with its VLIW (Very Long Instruction Word) vector architecture. This was a fairly blunt weapon designed to roll out large quantities of 3D graphics. However, the range of applications which a GPU typically has to handle, has widened considerably in recent years, with the high-end high-intensity Compute instructions only the visible tip of a broad mountain. VLIW wasn’t best-equipped to handle the changing-universe, as it tended to be highly efficient when left to tackle something other than its beloved 3D graphics. This wasn’t a problem for nVidia, as the company’s architecture was already scalar rather than vector in approach, and so could quickly turn its pins to a far wider range of areas. But with the 7970 and 7950, AMD brought in a new architecture, GCN (or Graphics Core Next), which was far more in line with nVidia’s scalar thinking. The 7950 is the older of the two products (the 7950 came out in early 2012, rather than the mid 2013 of its rival), but whereas the 7950 brought completely new thinking to AMD’s table, the 760 is really a retooling of older technology, an attempt to bring together the greatest hits from nVidia’s GK104 (Kepler) series. You’ll have spotted the GK104 in GPUs up and down nVidia’s recent ranges, from the GTX 660 and 680s, through to several of the members of the 700 series, including the GTX 750 and 760. Given its relative vintage, it’s unsurprising to see the GTX 760 lag behind the 7950 in terms of technology. The Radeon has the greater number of transistors, for instance, with 4.31 billion – the 760 has just 3.5 billion. The 760 is also seen mostly in a 2GB version, whereas the 7950 plumps squarely for a healthy 3GB. Both GPUs are, in their way, cut down models, though. The 760 takes that GK104 design, and enables just six of the eight SMXes (Streaming Multiprocessors) – the higher-end GK104 alternative, the GTX 770, has all eight switched on, giving it 33% more in terms of texture units and stream processors. Similarly, the 7950 can be compared to its bigger brother, the 7970, which has all of the 32 CUs (computing units) turned on, whereas the 7950 makes do with just 28 of them being enabled. Theoretically, the 7950 is closer to the max – its 28 CUs being just a 12.5% drop from the max, as opposed to the 25% fall in the case of the GTX 760. That holds up in terms of raw figures, too, with the 7950’s 1,792 stream processors and 112 texture units placing it significantly ahead of the 1,152 stream processors and 96 texture units of the GTX 760. The figures are slightly deceptive, as AMD’s designs have always tended to slant more towards weight of stream processors than the nVidia scalar architecture. Even with GCN, that bias remains in place, but it doesn’t necessarily mean that the AMD framework will provide more realworld performance. The AMD architecture has distinct advantages in some areas. Theoretically, the more versatile nVidia framework should be very well suited to areas like Compute. However, the GTX 760’s older GK104 underpinnings let it down here. When it comes to handling FP64 ( 64bit double-precision) calculations, the 760 struggles, slipping to a sixteenth of the performance shown with FP32. The Radeon HD 7950, however, chugs along quite happily at a quarter of the speed. Admittedly, not many purchasers of the 7950 or 760 will be looking for good FP64 performance, but the new Radeons are clearly fairly flexible. (See also: Best graphics card)
nVidia GTX 760 v Radeon HD 7950: core clock speeds
And there are still areas where the GTX 760 has its own sizeable advantage. The 760 may be a simpler and less sophisticated chip, but that leads to a highly efficient design that consumes less power. Consequently, clock speeds can be turned up. At standard factory settings, the GTX 760 begins with a core clock rate of 980MHz, but a quick Boost means this can be propelled to 1,033MHz when an extra burst of performance is needed. Board manufacturers have improved on the original settings, turning out overclocked variants that can push up to 1,150MHz. In contrast, the standard 7950 doesn’t have a Boost option, so at basic factory settings, the core clock rate is marooned at a mere 800MHz. New versions of the 7950 have since been unveiled, and you can in fact get a model with Boost, but even Sapphire’s excellent overclocked version can only muster a Boost of 925MHz – that’s a good 225MHz down on the best of the 760s. The 7950 does have that advantage in terms of texture units (112 to the GTX 760’s complement of 96), but it’s that dominant core clock speed that gives the GTX 760 an unassailable lead here. Even at its factory settings, the 760 can stretch to a texture fill rate of 99.2GT/sec. The standard 7950, stuck with a low normal speed, and absolutely no Boost, finds itself drowning on a figure of 56GT/sec. This is a dire figure, and AMD does much to fix this in the current generation of cards – the R9 280X, for instance, offers 128GT/sec. There’s no denying the poor core clocks of the standard 7950, though. The Sapphire Boost-enabled version does improve this significantly, though, and its texture fill rate of 103.6GT/sec is only a whisker behind the best of the 760s – the Gigabyte and MSI models, with rates of 110.4GT/sec.
nVidia GTX 760 v Radeon HD 7950: memory clock speeds
Turn our minds to the memory, and the 7950 ought to have the advantage again. Not that everything goes the AMD’s way. Once more, the relative simplicity and efficiency of the nVidia design allows it to push out an advantage on raw clock speeds. The GTX 760 comes with a standard memory clock rate of 1.5GHz, and the quadrupling capabilities of DDR RAM means that this figure is effectively transformed into 6GHz. That’s a nice little bump up on the 7950, which uses a default 1.25GHz memory clock to create an effective rate of 5GHz. That’s 1GHz behind the GTX 760, and a similar amount behind most of the other AMDs and nVidias. However, like the GTX 770, the 760 is hampered by its narrow memory interface. This is a 256bit model which struggles to match up to the relatively epic 384bit interface of the 7950. That 384bit interface is quite an achievement in a relatively low-end card, and even the newer R9 280X can’t better it. Bandwidth calculations show that the 7950’s memory subsystem is a smooth performer that cooks up a bandwidth figure of 240GB/sec. That’s better even than the GTX 770, and it’s a long way ahead of the 192GB/sec of the GTX 760, despite the latter’s superior clock rates. Once more, the 7950 looks the better GPU on paper, smacking the 760 around the head with some chunky specifications. The 7950’s 3GB of RAM is better than the 760’s default complement of 2GB. Extra memory is only likely to be crucial if you want to run multiple-screen games, and go beyond a 2560×1440/1600 resolution. And, should that be your aim, you probably won’t want these cards. 2GB, then, will suffice, although it’s well worth buying the 4GB version of the 760 if you think you’ll be interested in running two 760s in tandem.
nVidia GTX 760 v Radeon HD 7950: big-screen gaming and power
Both GPUs offer a similar quantity of Raster Operations – 32. The 7950 is capable of running up to six displays at a time, as opposed to the 760’s four. However, we wouldn’t recommend you do that in either case. Indeed, the GPUs aren’t powerful enough to make a really good job of triple-screen gaming – for that you’d need to be looking to at least the GTX 770 or R9 280X. The 760 does have a few extra features, such as support for PhysX gaming. However, the latter hasn’t really been exploited to the full by programmers, and while it’s a nice extra, it’s unlikely to be a deal-clincher. The 760 does have an advantage in terms of power. While the 7950 is itself relatively modest, the 760’s efficient design means that it pumps out at the pixels with surprisingly low power output. The TDP of 170 watts is a good 30 watts below the figure of the 7950. However, the 7950 isn’t the quietest of cards, at least not in its default form, and we found it to be around three decibels louder than the highly discreet 760. If it’s a quiet power-friendly card that you want, the 760 certainly has the advantage.
nVidia GTX 760 v Radeon HD 7950: benchmark tests
On-paper specifications can say whatever they want, but the only thing that really matters is that the card soars through real-world gaming. We’ve compared the GTX 760 and Radeon 7950 frame for frame at default settings, and across four different titles and three resolutions. These have also been put up against a Radeon 7970 (at round about £250), so that you have a point of comparison.
nVidia GTX 760 v Radeon HD 7950: Crysis 3
The 760 begins with a tidy win at the lowest resolution of 1680×1,050, crafting a clear 6.3fps lead over the 7950. By the time we get to the top resolution of 2560×1600, the lead has fallen to a mere 1.2fps. Nonetheless, the 760 does retain an advantage all the way through – and at the lower and more relevant resolutions, it’s quite impressive. It also fares very well against the 7950’s bigger brother, the 7970, and only at the highest resolution does the 7970 get itself a lead – and here a relatively modest 0.2fps.
nVidia GTX 760 v Radeon HD 7950: Bioshock Infinite Rage
If anything, the GTX 760 reinforces its lead here, perhaps showing that on-paper specifications aren’t to be trusted. At the 1600×900 resolution, it builds up a lead of 8.3fps, and this lead remains at 3.8fps even over the top resolution of 2560×1600. The 7970 starts to reveal its superiority here, though, opening up a 4.5fps lead at the lower resolution. Curiously, though, by the time we get to the highest resolution, the 760 has overhauled the 7970 again.
nVidia GTX 760 v Radeon HD 7950: Battlefield 3
The GTX 760 is emphatic in this game, demolishing the 7950 by 16fps at the lowest resolution. The gap has fallen to 3.9fps by the time we reach 2,560×1,600. But at the lower resolutions, where it really counts, the GTX 760 is dominant. The 7970 is beaten in every resolution, albeit by just 0.4-0.6fps.
nVidia GTX 760 v Radeon HD 7950: Assassin’s Creed 3
The 760 ends with one more victory, finishing 7.4fps ahead of the 7950 at 1600×900. Even at the top resolution, the lead remains a clear 5.6fps. The 7970 strikes back marginally, though, finishing 1.3fps ahead at the lowest resolution. By 2560×1600, the lead has fallen to a mere 0.2fps.
nVidia GTX 760 v Radeon HD 7950: verdict
The 7950 isn’t a very easy card to find at the moment. That’s probably intentional on AMD’s part, as the company knows that the 7950 really can’t compete with the 760. Even if the 7950 is available for around £200, it’s an inferior product when it comes to gaming. The 760 turns an advantage in most tests. It’s not usually a very big one, but 10 – 15% is typical, especially at lower resolutions. That means the 7950 would have to be available for around the £160/£170 mark in order to be competitive. The 760 gives you a very solid leap forwards. If you’re looking to upgrade, we would recommend spending a little more on the 770, but the 760 will do the job. The 760 is also pretty quiet, and very low on power, making it a very good choice for modest systems. We would, though, warn against the 4GB version. There’s simply no need for the extra memory. You’re not going to use the 760 for the sort of demanding game setups that require 4GB. By 2016, it might be that graphics cards need the extra memory to properly handle the latest games. But the 760 will be looking distinctly jaded by then, so you’re better off trying to spend a sum just above £200 on a 760, and then looking to upgrade it in a couple of years time. Read more: GTX 760 vs GTX 770 graphics card comparison review.
nVidia GTX 760 v Radeon HD 7950: available graphics cards
GTX 760
£195 – KFA GeForce GTX 760 2048MB – 2GB, Core Clock 980MHz (1,033MHz Boost), Memory Clock DDR Effective 6,008MHz, 2yr Warranty£200 – Gigabyte GeForce GTX 760 WindForce OC – 2GB, Core Clock 1,085MHz (1,150MHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty£210 – Asus GeForce GTX 760 DirectCUII – 2GB, Core Clock 1,006MHz (1,072MHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty£210 – EVGA GeForce GTX 760 SC – 2GB, Core Clock 1,072MHz (1,137MHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty£230 – Inno3D GeForce GTX 760 iChill – 2GB, Core Clock 1,060MHz (1,124MHz Boost), Memory Clock DDR Effective 6.2GHz, 2yr Warranty£250 – Gainward GeForce GTX 760 Phantom 4096MB – 4GB, Core Clock 1,072MHz (1,137MHz Boost), Memory Clock DDR Effective 6.2GHz, 2yr Warranty£260 – MSI GeForce GTX 760 OC Gaming Edition 4096MB – 4GB, Core Clock 1,085MHz (1,150MHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty
Verdict – The GTX 760 is all about the value pricing, so you don’t want to push the costs up by seeking out high clock rates. Even so, where there’s speed available at little extra cost, you don’t want to spurn it. The Gigabyte version is a spectacular deal, giving you a 1,150MHz Boost clock for around the £200 mark. The Asus and EVGA are also decent choices, and offer some value at £210. In our opinion, extra memory would be a waste on a product like this one. If you do want to go down that route, though, the Inno3D is an alluring card, and offers searing specs along with beautifully-conceived cooling facilities.
Radeon HD 7950
£265 – Sapphire Radeon HD 7950 Flex OC – 3GB, Core Clock 860MHz (925MHz Boost), Memory Clock DDR Effective 5GHz, 3yr Warranty Verdict – It’s now very hard to find the 7950, so you don’t really have a great deal of choice. In all honesty, £265 is a bit too much to pay for a card that’s a relatively modest performer. Hunt around (particularly on Ebay), and you can pick up models for under £200, though. Once you can get the tag down to £170 or lower, the 7950 starts to make some sense. We wouldn’t want to be paying in excess of £200 though.
