AMD Radeon R9 280X vs nVidia GeForce GTX 770: out with VLIW and in with GCN
When AMD disbanded its old VLIW (Very Long Instruction Word) architecture in 2012, and replaced it with an all-new GCN (Graphics Core Next) platform, this marked a seachange in the company’s thinking. The older VLIW technology had been very good at simply hurling pixels and texels at the screen, but lacked finesse when it came to applications other than straightforward 3D graphics. With big rival nVidia have long adopted ‘scalar’ architecture that could think on the spot more easily, and try its hand at a number of different application types (including the increasingly important Compute code), it was important that AMD brought its main GPU framework into line with modern developments. Having done that with GCN (Tahiti), the company now feels it can spend a bit of time developing the platform rather than coming up with another dramatic overhaul. The Radeon R9 280X, then, is less a radical new product, but more a culmination of everything that’s good about the GCN architecture. To a certain extent, the same could be said of the GTX 770. This GPU is built around the GK104 (Kepler), a splendid chip which has seen service in a range of nVidia products, from the older GTX 660 and 680s, to the GTX 750 and 760. The GTX 770 refresh came in 2013, but the GK104 is considerably older than that. As such, it’s perhaps unsurprising to notice the GTX 770 looking relatively unsophisticated in contrast with even the R9 280X. Both the GTX 770 and R9 280X are created using the 0.28nm manufacturing process, but the AMD is stuffed with considerably more transistors – 4.31 billion rather than the 3.5 billion of the GTX 770. And while the R9 280X has a healthy 3GB of memory as its default configuration (in line with many previous GCN GPUs), the GTX 770 starts with a mere 2GB. You can buy a 4GB version of the 770, but this will cost you around an extra £50. We’ll look later at whether this is necessary or not. In many respects, the R9 280X is rather like the older HD 7970. That 7970 pushed the Tahiti chip to its limits, enabling all 32 of its CUs (computing units), and achieving 2048 stream processors and 128 texture units. The R9 280X follows in almost identical footsteps, offering the same complement of stream processors and texture units. The GTX 770 is much of the same mind, enabling all eight of the GK104’s SMXes (Streaming Multiprocessors), and delivering the same 128 texture units. The streaming processors are a little down in number, though, and even this maximum version of the GK104 can offer no more than 1536 stream processors. It’s important not to get too hung up on that difference, though. Directly comparing the AMD and nVidia architectures has frequently been a rather misleading activity, and AMD’s solutions have frequently housed larger numbers of streaming processors, but not always with apparent benefit. We’ll see presently whether the AMD architecture delivers on its promise.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: core clock speeds
It’s not as though AMD hasn’t pulled back the throttle elsewhere. Strangely, despite the R9 280X seeming, in many respects, to be like the older HD 7970, the R9 falls back in terms of clock speeds. The standard factory settings allow for a core clock rate of just 850MHz – the 7970 offered 9.25GHz. However, unlike the standard 7970, the 280X does at least have a Boost facility, allowing it to hit 1GHz. Not that this allows it to match up to its nVidia rival, and the GTX 770 starts off with a simple 1046MHz clock rate, before building on it with a Boost that propels the card to 1085MHz. Board manufacturers have tried to squeeze the extra potential from both cards. Typical 280X boards top out at about 1,070MHz, although it’s certainly possible to find 1,100MHz, and even 1,150MHz models. Such a figure is fairly average for the GTX 770, where speeds in the region of 1,189MHz to 1,202MHz are available in large numbers. In short, the GTX 770 definitely seems to have a sizeable advantage in terms of core clock speeds. The 770 and the R9 280X both have the same number of texture units (128), so that rise in clock speeds becomes significant when we look at texture fill rates. Using the factory settings of both GPUs, the GTX 770 is noticeably faster, hitting a figure of 138.9GT/sec in reply to the 280X’s less impressive 128GT/sec. The lead increases if you look at typical third-party boards that use these GPUs. A good number of GTX 770 cards reach texture fill rates of between 152.2GT/sec to 153.9GT/sec, whereas the majority of the 280X boards are stuck in the region of 134.4GT/sec to 140.8GT/sec. The GTX 770 seems to have a healthy lead here, then.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: memory clock speeds
The relative lack of complexity in the GTX 770 means that it can run higher clock speeds across the board. It repeats the trick with the memory, and its 1.75GHz rate works out as an effective 7GHz when the quadrupling effect of the RAM is taken into account. That 7GHz is right at the edge of what’s achievable with the current memory chips, and so comprises an impressive achievement. The R9 280X makes no attempt to match this, plumping instead for a 1,502MHz (6,008MHz DDR effective) memory clock. This is by no means poor, and many other cards (the GTX 760 amongst them) offer similar figures. However, it is still advantage to the GTX 770 in terms of raw clock rates. That advantage doesn’t last, though. That’s primarily down to a disparity in memory interfaces. While the GTX 770 is saddled with a middling 256bit interface, the 280X’s extends to a sizeable 384 bits. Bandwidth calculations show that this gives the Radeon a very great boost in terms of its ability to juggle graphics data, and its rate of 288.4GB/sec destroys the GTX 770’s comparatively meagre figure of 224GB/sec. The very best third-party 280X cards stretch the memory clock to 6.4GHz, giving the Radeon an unassailable 307.2GB/sec – even the most powerfully overclocked GTX 770 can only reach 230.4GB/sec. This section, then, hands a victory to the 280X.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: is 2GB sufficient?
While the Radeon has a default figure of 3GB, the GTX 770 starts at just 2GB – although you can buy a 4GB version of the GTX as well. For standard single-screen gaming, 2GB will be more than sufficient, and you’re unlikely to see a difference of more than a frame or two unless the game is incredibly complex. However, if you’re going up to a dual-screen setup, and pushing the resolution beyond 2560×1440/1600, the 2GB complement could feel inadequate. The question really is how ambitious are you as a gamer? The 2GB GTX 770 is well-priced for a card that should give you powerful frame rates for the next two years. Whether it’ll retains its freshness in 2016/2017, though, is hard to say. Certainly 4GB will make it more future-proof. For standard gaming, though, 2GB will be more than sufficient.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: other features
Both GPUs offer an identical number of Raster Operations – 32. Both are also capable of impeccable three-display gaming, although the GTX 770 adds the potential of PhysX support. This feature is unlikely to be a ncher though, as we have yet to see gamers programmers produce the support that makes it a must-have. The GTX 770 is superior in terms of power output, and works on a more efficient and less hungry design. The 770 is also the quieter card, so those who like to watch their decibels may be swayed by the nVidia.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: higher-end for higher-compute?
The difference between gaming systems and higher-end machines – which traditionally have used Quadro-Tesla chips – has been falling in recent years. The new generation of GPUs are the first ‘home’ versions that can be expected to produce strong performance in high-end calculations, such as those practised by Compute. Or, at least, that’s the idea. Although the GTX 770 could, theoretically, offer strong Compute capabilities, such features have been heavily limited by nVidia. That’s not so in the case of the R9 280X, where AMD has allowed it to offer searing FP64 (64bit double-precision) calculations. In this area, the 280X manages to beat the 7970 (itself no slouch at Compute). It still lags behind the nVidia Titan, but then, that card costs twice as much, and is very much tailored towards higher-end users. Should gamers care too much about the 280X’s Compute capabilities? Well, it shouldn’t be a deal-clincher, certainly, but it does hint at the huge versatility within the R9 280X. If you’re looking to the future, the Radeon looks better-placed to ride the trends.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: the might of mantle?
A potential powder keg comes courtesy of AMD – in the shape of Mantle. This is an API (Application Programming Interface) that’s designed to squeeze extra performance and capabilities from AMD’s GCN hardware that existing generic APIs (such as OpenGL and Direct3D) simply can’t deliver. The problem is that games programmers will need to support Mantle directly, raising the prospect of games being released that require AMD hardware to work – or to work to their fullest. This would potentially fragment the market, taking us back to the PC’s earlier days, when picking a brand of graphics card meant far more than a trimmer price tag or a bit of extra speed. Whether AMDs’s attempt to grab a stranglehold succeeds will depend on gaming companies. AMD should get an extra boost there thanks to Mantle’s strong crossover appeal with the Xbox One games console. In truth, it’s unlikely to be significant for the next year or so, but it’ll be interesting (and will perhaps pose something of a dilemma) to see whether, one day, a GPU-specific API like Mantle looks like genuinely splitting the graphics card market. (See also: nVidia GTX 760 v Radeon HD 7950 GPU comparison: which is the best mid-range graphics card?)
AMD Radeon R9 280X vs nVidia GeForce GTX 770: benchmark tests
While the specifications tend to be fairly important, the best technology in the world can’t help the card if it doesn’t hit the heights in real-world gaming. We’ve witnessed the clash of these two powerful cards, setting them against one another at default settings, and across four different titles and three resolutions. So that you have some sort of benchmark to compare them to, we’ve also dropped in a GTX 760 (at round about £200), so that you have a yardstick.
Crysis 3
The 770 squeezes out a noticeable but not massive lead at the lowest resolution of 1680×1,050, scooping up an extra 4.3fps. This lead falls only slightly as the detail levels climb, resulting in a 3.8fps deficit at 2,560×1,600 – this is actually a larger lead in terms of percentages. The 760 is substantially behind both the 770 and 280X, finishing 7.6fps behind the latter at the lowest resolution, and closing that marginally (to 5.9fps) over the highest mode.
Bioshock Infinite Rage
The GTX 770 wins again here, but not by much. The 280X is within 3.3fps of it at 1,600×900, and just 1.7fps away at 2,560×1,600. The GTX 760 is a long way behind both, falling back some 17.4fps from the 280X at the lowest resolution. It does partially bridge that gap as the detail levels are piled on though, and strangely, the lead is just 7.6fps at 2,560×1,600.
Battlefield 3
Battlefield 3 is a very straightforward benchmark, and the GTX 770 obliges with a monster three-figure score at 1,600×900, placing it some 8.2fps ahead of the 280X. At the highest resolution, that gap is closed very significantly, with the 280X just 3.4fps behind as the detail reaches its stickiest levels. The GTX 760 is a further 7.5fps back from the 280X at the lowest resolution, but a more respectable 2.5fps behind at 2,560×1,600.
Assassin’s Creed 3
The GTX 770 finishes with another capable but not gigantic lead, its 1,600×900 resolution yielding a lead of 4.6fps, while the top resolution sees it just 2.7fps ahead – noticeable, but only just. The GTX 760 is a rather gaping 11.8fps behind the 280X at the lowest resolution, but closes to within 6.4fps at the top 2,560×1,600 mode.
AMD Radeon R9 280X vs nVidia GeForce GTX 770: verdict
The decision over which GPU to choose will come down largely to money. Neither GPU offers much in the way of new technology (although the Mantle support on the 280X might prove a significant innovation in years to come), but the GTX 770 is less power-intensive, and operates more quietly. Crucially, it’s also a faster GPU, even if there isn’t usually more than a few frames in it. So price becomes the crucial issue. The MSI GTX 770 ends up being the cheapest of all cards, making this the best buy. In truth, though, if there’s a difference in price of less than £20, the GTX 770 is the product to have. There would be a slight concern about the memory, and the 3GB of the 280X is likely to look better two years from now. But assuming you’re not looking to run games with multi-monitors, the memory is unlikely to be a killer. The GTX 770, then, looks the slightly better deal. Both, though, are very impressive pieces of silicon that offer superb frame rates for the money. (See also: GTX 760 vs GTX 770 GPU comparison review: Which GK104 Kepler graphics card is best?)
AMD Radeon R9 280X vs nVidia GeForce GTX 770: available graphics cards
GTX 770
£252 – MSI GTX 770 Twin Frozr – 2GB, Core Clock 1,098MHz (1,150MHz Boost), Memory Clock DDR Effective 7GHz, 2yr Warranty£270 – Asus GeForce GTX 770 DirectCU OC – 2GB, Core Clock 1,058MHz (1,110MHz Boost), Memory Clock DDR Effective 7GHz, 3yr Warranty£276 – Gigabyte GeForce GTX 770 WindForce 3x OC – 2GB, Core Clock 1,137MHz (1,189MHz Boost), Memory Clock DDR Effective 7GHz, 3yr Warranty£300 – KFA2 GeForce GTX 770 LTD OC – 2GB, Core Clock 1,150MHz (1,202MHz Boost), Memory Clock DDR Effective 7GHz, 2yr Warranty£320 – Gainward GeForce GTX 770 Phantom – 4GB, Core Clock 1,046MHz (1,085MHz Boost), Memory Clock DDR Effective 7GHz, 2yr Warranty£330 – Inno3D GeForce GTX 770 iChill Herculez 4096MB – 4GB, Core Clock 1,150MHz (1,196MHz Boost), Memory Clock DDR Effective 7.2GHz, 3yr Warranty£360 – EVGA GeForce GTX 770 Classified ACX 4096MB – 4GB, Core Clock 1,150MHz (1,202MHz Boost), Memory Clock DDR Effective 7GHz, 3yr Warranty Verdict – For the very best of value, look to those 2GB cards, as you won’t generally see more than a frame-rate or two extra when lashing for the more expensive 4GB versions.The Gigabyte and KFA2 offer some very juicy stats, especially when you’re focusing on the Boost clock speed. But you don’t need to eke out every last drop of pace. Purely as a value buy, the MSI is fantastic, and its highly-capable 1,150MHz Boost clock means you’ll hardly notice the difference between this and the KFA2. If you’re looking to the longer term, you may find the extra RAM worth it though. The Inno3D is probably the best choice here, and the 1,196MHz Boost clock means that the £330 price tag isn’t too bitter a pill to swallow.
Radeon R9 280X
£260 – Powercolor Radeon R9 280X TurboDUO – 3GB, Core Clock 880MHz (1,030MHz Boost), Memory Clock DDR Effective 6,008MHz, 2yr Warranty£270 – Sapphire Radeon R9 280X Dual-X OC – 3GB, Core Clock 870MHz (1,020MHz Boost), Memory Clock DDR Effective 6,008MHz, 2yr Warranty£290 – Gigabyte Radeon R9 280X WindForce 3X OC – 3GB, Core Clock 1GHz (1.1GHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty£300 – MSI Radeon R9 280X Gaming Edition OC – 3GB, Core Clock 1GHz (1,050MHz Boost), Memory Clock DDR Effective 6,008MHz, 3yr Warranty£300 – Asus Radeon R9 280X DirectCUII – 3GB, Core Clock 970MHz (1,070MHz Boost), Memory Clock DDR Effective 6.4GHz, 3yr Warranty£320 – Sapphire Radeon R9 280X Toxic OC – 3GB, Core Clock 1.1GHz (1,150MHz Boost), Memory Clock DDR Effective 6.4GHz, 2yr Warranty Verdict – The 280X cards all come with 3GB of RAM as standard, so you won’t need to make any tricky decisions about memory capacities. However, there’s still a decent range of prices, going from the £260 Powercolor to the highly expensive £320 Sapphire Toxic. In fairness to Sapphire, the Toxic does come with some fantastic specifications, and a 1,150MHz Boost clock and 6.4GHz memory are peerless. At the other end of the scale, the Powercolor comes with very meagre clock speeds, but will still offer the majority of the power, but for £60 less. A good compromise would be the Gigabyte. This offers a 1GHz core clock, with a fantastic Boost to 1.1GHz. Yet it costs just £30 more than the Powercolor.
