Nvidia GeForce GTX 980 4GB Review

Manufacturer: NVIDIA
UK price (as reviewed): MSRP £429 (inc VAT)
US price (as reviewed): MSRP TBC

When Maxwell first launched, it marked a shift in Nvidia’s launch strategy. The GTX 750 Ti was and still is a good card, but starting with such a low-end product was unheard of for a new Nvidia architectural launch until then. The reason was that efficiency and performance per watt had overtaken pure performance as the company’s primary focus. This makes sense in an increasingly mobile market, and indeed it wasn’t long at all before the same GM107 GPU as used in the GTX 750 Ti found itself in more power-constrained environments like notebooks. After all, GM107 was billed as the most efficient GPU ever built, and it had the numbers to prove it, with the GTX 750 Ti regularly marching the GTX 480 in performance but at a fraction of the power consumption.

At the time, many people, including ourselves, stated that they couldn’t wait to see what Maxwell could do when applied at the high end. Well, now the time is here. Today, Nvidia is launching not one but two new graphics cards; the new flagship GTX 980 and the GTX 970, both of which are based on a brand new GPU, GM204, which Nvidia claims is not only the world’s new most efficient GPU, but also the world’s fastest. Bold words indeed.

Click to enlarge
This review is focussed solely on the flagship GTX 980, with Nvidia having sampled us with a reference board. The GTX 970, meanwhile, has reference specifications but is not launching with a true stock card or PCB. Coolers and clock speeds are thus liable to vary significantly between Nvidia’s partners. You can read our three card launch day roundup of GTX 970 reviews right here, featuring ASUS, EVGA and MSI.

A new GPU launch wouldn’t be complete without some naming confusion to clear up. If you’re wondering what happened to the GTX 800 desktop series, the simple answer is that it’s been skipped. The Maxwell GTX 750 Ti launching into the product stack between Kepler GTX 600 products was confusing enough, but Nvidia is using this opportunity to skip a series in the desktop parts to bring them into closer alignment with the mobile parts, which have fallen out of sync due to previous rebranding strategies. It’s a bit of a mess, but them’s the breaks.

Maxwell’s focus is still efficiency above anything else, but it should still see a small performance increase over the GTX 780 and even the GTX 780 Ti. After all, GPUs are incredibly power limited, so anything that improves efficiency also increases the headroom within a given power limit for more performance. The specifications of the new cards are outlined below, and discussed in greater detail over the page. The main thing to note, however, is the 165W TDP for the GTX 980. This is very low indeed, almost unbelievably so for a flagship card, and is testament again to Maxwell’s efficiency. Despite a 33 percent higher core count and a 12 percent higher clock speed, this TDP is 30W lower than the GX 680, which also gives the GTX 980 approximately twice the GFLOPS per watt of that card – pretty amazing given that both are made on the same 28nm process.

  Nvidia GeForce GTX 980 4GB Nvidia GeForce GTX 970 4GB Nvidia GeForce GTX 780 Ti 3GB Nvidia GeForce GTX 780 3GB Nvidia GeForce GTX 680 2GB GPU Architecture Maxwell Maxwell Kepler Kepler Kepler Codename GM204 GM204 GK110 GK110 GK104 Base Clock 1,126MHz 1,050MHz 876MHz 836MHz 1,006MHz Boost Clock 1,216MHz 1,178MHz 928MHz 900MHz 1,058MHz Stream Processors 2,048 1,664 2,880 2,304 1,536 Layout 4 GPCs, 16 SMMs 4 GPCs, 13 SMMs 5 GPCs, 15 SMXs 4 GPCs, 12 SMXs 4 GPCs, 8 SMXs Rasterisers 4 4 5 4 4 Tesselation Units 16 13 15 12 8 Texture Units 128 104 240 194 128 ROPs 64 64 48 48 32 Transistors 5.2 billion 5.2 billion 7.1 billion 7.1 billion 3.54 billion Die Size 398mm² 398mm² 533mm² 551mm² 294mm² Process 28nm 28nm 28nm 28nm 28nm Memory Amount 4GB GDDR5 4GB GDDR5 3GB GDDR5 3GB GDDR5 2GB GDDR5 Frequency 1.75GHz (7GHz Effective) 1.75GHz (7GHz Effective) 1.75GHz (7GHz Effective) 1.5GHz (6GHz Effective) 1.5GHz (6GHz Effective) Interface 256-bit 256-bit 384-bit 384-bit 256-bit Bandwidth 224GB/sec 224GB/sec 336GB/sec 288GB/sec 192GB/sec Card Specifications Power Connectors 2 x 6-pin PCI-E 2 x 6-pin PCI-E 1 x 6-pin, 1 x 8-pin PCI-E 1 x 6-pin, 1 x 8-pin PCI-E 2 x 6-pin PCI-E Stock Card Length 267mm 267mm 267mm 267mm 252mm TDP 165W 145W 250W 250W 195W

The GTX 980 is a DirectX 12-compatible card. In fact, all of Nvidia’s Fermi, Kepler and Maxwell GPUs will support the upcoming API, although only GM2xx GPUs will fully support the newer rendering features like Conservative Rasters and Raster Ordered Views. Developers can access such features through the DirectX 11 API, hence the support for DX12 in Fermi and above, but as we understand it native support for them and therefore, presumably, better performance when using them will only be available in the latest GPUs, such as GM204.

The new flagship card again supports up to four screens concurrently, but the display controllers have been upgraded, as it’s now possible to run four 4K MST displays, the previous limit being two. Further, 5K resolutions are now supported – the maximum supported resolution is 5,120 x 3,200 at 60Hz. DisplayPort is still version 1.2, but HDMI has been upgraded to 2.0. GM204 is the world’s first GPU with support for this standard, which has enough bandwidth to run 4K displays at 60Hz.

Nvidia’s fixed function encode unit, NVENC, carries with it the upgrades first found in the GTX 750 Ti, including a local decode cache and support for the low power GC5 state. However, it also adds native support for H.265 (HEVC) encoding, and H.264 encoding is around 2.5 times faster than with Kepler. This speed boost enables 4K video to be encoded at 60fps, which Nvidia is able to take advantage of with ShadowPlay.

Finally, Nvidia’s G-Sync technology is of course supported. In fact, you can run three G-Sync displays at once in a Surround configuration using the card’s DisplayPort capabilities, should you so desire.

GM204: Maxwell Fully Realised

The GM107 chip inside the GTX 750 Ti was referred to at the time as a first generation Maxwell part. GM204, meanwhile, is the start of the second generation and is essentially the full realisation of the architecture. 5.2 billion transistors have been crammed into the 398mm² die. As mentioned, it’s manufactured using the same (albeit now very mature) TSMC 28nm process as that which Kepler was first built on. Delays in the development of the 20nm process are responsible, but even on this same process Nvidia has been able to make significant efficiency gains.

Click to enlarge – The GM204 die and block diagram
The main building blocks of GM204 are Nvidia’s redesigned streaming multiprocessors, now referred to as SMMs, which are arranged into Graphics Processing Clusters, or GPCs. The SMMs have the same design as those in GM107, and you can read more about them in our coverage of the GTX 750 Ti. Essentially, however, Nvidia cut the CUDA core count from 192 in Kepler to 128, and split these into four blocks of 32, each with its own warp and scheduling hardware. Reorganising the data path into a simpler, power-of-two structure like this allowed the cores to be fully utilised more often and improved scheduling efficiency by reducing the need for the complex resource sharing required in Kepler, where the warp and scheduling hardware had access to all 192 cores. Ultimately, Nvidia improved performance per core by around 35-40 percent in shader limited workloads, effectively double performance per watt over Kepler and maintained around 90 percent of performance with just two thirds of the original core count.

Click to enlarge – a Maxwell SMMWhere GM107 has a single GPC with five SMMs, GM204 has four GPCs, each of which has its own raster engine and four SMMs. As well as the 128 CUDA cores, each SMM has its own geometry unit in the form of the PolyMorph engine, and eight texture units. There are thus 2,048 cores, 16 geometry units and 128 texture units in total. The doubling up of SMs and thus geometry units compared to GK104 (the GTX 680’s GPU) will help in geometry heavy workloads, such as those with high levels of tesselation.

Clock speeds have also been given a bump. In fact, the GTX 980 ships with Nvidia’s highest clock speeds yet. The GPU has a base clock of 1,126MHz, and a rated boost clock of 1,216MHz. This boost clock can be considered an average of sorts; the actual boost frequency you see depends on a number of factors including power, temperature and workload requirements. In our own testing, we found it would boost to a maximum of 1,240MHz, but stayed on average between 1,177MHz and 1,200MHz.

Those of you with an eye for detail may have noticed that GM204 has the same number of texture units as GK104 – there are twice as many SMs in the new GPU, but Maxwell SMs have half as many texture units (eight) as their Kepler counterparts. The reason for this is, as you might have expected, efficiency. Nvidia found that eight texture units offered the most efficient balance of math and texture processing for today’s games. Also, the 12 percent higher clock speed of GM204 means that texture fill rate is up by the same amount.

Four 64-bit memory controllers grant the GTX 980 a 256-bit memory interface, the same as the GTX 680 and actually less than the GK110 parts (GTX 780, GTX 780 Ti and the GTX Titan family). However, Nvidia has a few ways of preventing this relatively narrow interface from becoming a bottleneck. First is simply the use of fast memory; all GTX 980 parts will ship with its GDDR5 clocked at 1.75GHz (7GHz effective), the joint highest within Nvidia’s line-up. As such, memory bandwidth stands at 224GB/sec, a respectable but not outstanding figure.

Next is the size of the onboard L2 cache, which is a whopping 2MB (512KB aligned with each memory controller). This is the same as that of the GTX 750 Ti, but it’s also bigger than any other card on the market, including the GTX Titan range. The result of so large a cache is less reliance on the comparatively slow and power-hungry off-chip memory and less demand on the memory bus.

Click to enlarge
Finally, for when GM204 does need to access the off-chip GDDR5, Nvidia has introduced third generation delta colour compression, similar to what we saw in AMD’s Tonga GPU with the R9 285. It is essentially a lossless method of compressing frame data to reduce the memory bandwidth requirements when writing to (and later when reading from) memory, and this third generation of the process gives the compressor more choices when calculating delta values. Nvidia found that in most cases, the majority of frame data can be compressed, and there is no performance penalty for doing so. There is also no image quality reduction since it must always be lossless – if this isn’t possible, it will be written uncompressed. According to Nvidia, the result of the technique is a reduction of bytes needed from memory per frame of around 25 percent on average.

Like L2 cache segments, ROPs are again aligned with the memory controllers. In GM107, we saw eight ROPs for each of the two controllers, but in GM204 there are 16 per controller for a total of 64 ROPs. This is double the count of the GTX 680, and with the higher clock speeds pixel fill rate is more than doubled. This, along with the doubling of the frame buffer from 2GB to 4GB, equips the GTX 980 for 4K gaming.

New Graphics Processing Techniques

As well as the new card, Nvidia was keen to share with us details of three new graphics processing techniques, each with their own catchy acronyms: Voxel Global Illumination (VXGI), Dynamic Super Resolution (DSR) and Multi-Frame Sampled Anti-Aliasing (MFAA).

Voxel Global Illumination (VXGI)

VXGI is a real-time global illumination engine developed from a technique first revealed in 2011 by Nvidia engineer Cyril Crassin (see the paper on it here and a video on it here) that uses voxel grids and cone tracing to calculate an approximation of global illumination, which is essentially the tracing of all direct light (that from a light source) and indirect light (that which has previously bounced off or passed through one or more objects) in a scene at any given time. Nvidia considers it to be the next big leap forward for lighting, and a small step closer to real-time ray-tracing.

Click to enlarge
Modelling lighting is the most computationally difficult problem in graphics. Indirect light in particular is very costly and complex, as you have to consider the material properties of every object in the scene to render it accurately, and this becomes even harder to manage with geometry that is dynamic rather than fixed. Current global illumination techniques typically rely on pre-calculated lighting and carefully placed artwork, neither of which are particularly well suited to animated or moving objects. VXGI aims to make the process a real-time one that is compatible with and suitable for dynamic geometry and lights for more realistic lighting.

Click to enlarge
First is the process of voxelisation, which essentially turns a 2D scene into a series of 3D cubes called voxels. It is then determined for each voxel whether it is empty or filled with an object, and if it’s the latter then the percentage of coverage is also calculated.

Click to enlarge
With the coverage stage complete, it’s time to add lighting information. Each object in the scene will have information about how it responds to light – its colour, opacity, reflectiveness and so on, and this information is stored in any voxel with which an object interacts. The scene is then rendered multiple times from the perspective of the light source(s) in the scene, which tells you the amount of light that hits each voxel.

Using the stored information about where objects are and how the objects respond to light, it’s then possible to make calculations about the colour, direction and intensity of any light that is reflected from any object. Essentially, you’re left with a voxel data structure that represents the lighting information in a scene at a given time.

Click to enlarge
The final step is of course to rasterise the scene for display on a 2D screen, with voxel data structure being one that’s very powerful for accurate lighting calculations. Indirect lighting is done through cone tracing, which is an approximation of ray tracing (whereby individual rays are traced through a scene). As you can imagine, ray tracing is hugely expensive and far out of the realm of real-time calculations for now, but cone tracing reduces thousands of individual rays into voxel cones which are then traced through the voxel grid to produce indirect lighting effects. Cones can have their properties altered to produce different lighting effects too, and it can all be done in real-time, bringing a heightened level of realism to the light of moving objects.

Click to enlarge
VXGI will be available in Unreal Engine 4, as well as other major engines, from Q4 of this year. Its software algorithm will run on all GPUs, and it is also scalable across less powerful processors such as those in consoles (e.g. by lowering the density of the voxel grid) but it will be accelerated on GM2xx Maxwell parts thanks to their native support for Viewport Multicast and Conservative Raster, both of which speed up the voxelisation process.

Dynamic Super Resolution (DSR)

Next up is Dynamic Super Resolution, or DSR. This is a feature being introduced with this generation of Maxwell products, but not one that will remain Maxwell-exclusive. It is essentially the process of rendering a game internally at a higher resolution than your display’s native one, and then downscaling it to match your display resolution. The benefit is a better looking final image that has less aliasing artifacts, though it of course comes with a performance hit – it is for less demanding games where even at full settings you have available graphics headroom but lack a display with a high enough resolution to take full advantage of it.

Click to enlarge
Similar results can be obtained using methods that are already available, but they are nearly always fiddling to set-up and may very well come with a host of incompatibility problems in certain games. With DSR, it’s fully integrated into the driver; for novice users, it can be quickly enabled or disabled in GeForce Experience. To ensure a final image that is as smooth and jaggy-free as possible, Nvidia applies a 13-tap Gaussian blur when downscaling.

Click to enlarge

Multi-Frame Sampled Anti-Aliasing (MFAA)

The final technique is MFAA, Multi-Frame Sampled Anti-Aliasing, designed for situations opposite to DSR, i.e. where you need more frames per second but don’t want to sacrifice image quality. This technique is exclusive to GM2xx GPUs, as it relies on support for a technique called multi-pixel programmable sampling. Standard MSAA uses fixed sample patterns – 4x MSAA, for example, will sample a pixel through which an edge passes at four fixed locations to calculate its colour, and these locations remain fixed from frame to frame.

Click to enlarge
Multi-pixel programmable sampling, however, allows developers to program custom pixel sample positions which can vary from frame to frame or even within a single frame, and there are many positions to choose from: 256 in a standard 16×16 pixel grid. This technique can be used to generate new AA methods, and MFAA is one that Nvidia has made itself. Like DSR, it will be implemented at the driver level and enabled and disabled with a control panel setting. It is not available in the launch driver, but will be coming soon with support for a wide range of exisiting games.

Click to enlarge
The aim of MFAA is to produce an image that is almost identical to a certain level of MSAA, but at the performance hit of the MSAA level below (e.g. 4x MSAA image quality at a 2x MSAA performance level). Essentially, it uses a 2x MSAA sample pattern, but varies the position of the sample from frame to frame. It then combines the sampled data of the current frame with that of the previous frame to produce double the amount of sample positions than it is currently sampling, applying what Nvidia calls a Temporal Synthesis Filter to convincingly merge the data for the current frame. The result, Nvidia claims, is about 30 percent more performance with effectively the same image. Of course, such claims remain to be seen.

The Nvidia GeForce GTX 980 4GB Card

With our architectural and feature discussion behind us, we can now look at the physical incarnation of the GTX 980 itself. It’s a dual slot model that uses a design first introduced with the GTX 690 and made all the more familiar with the GK110 GPU-based cards. In fact, the card is even the same length as the GTX 780 etc. at 267mm. It’s encased in an aluminium frame and incorporates a black magnesium alloy housing for the fan. Also included is the trademark LED backlit green GeForce GTX logo along the top of the card. The build quality is again truly fantastic and while we still think it looks great, we do wonder whether now would have been a good time for Nvidia to spruce the design up a little.

Click to enlarge
Examining the rear I/O we find that the reference set of display outputs has moved away from DVI and towards DisplayPort – something we’re fully in support of. The single dual link DVI-I connection is joined by one HDMI 2.0 port and three DisplayPort 1.2 connections. As said, any four of these can be used at once.

Along the top edge we find a pair of SLI connectors for multi-GPU enthusiasts, as well as two supplementary 6-pin PCI-E power connectors. This itself is a sign of the card’s low power credentials – high end launches are almost invariably fitted with a 6-pin/8-pin combination.

Click to enlarge
A new addition to the design is full cover backplate, providing users with aesthetic benefits as well as protection and additional heat dissipation for the PCB and the components within. It’s also partially removable; you can remove a section of the backplate as shown to increase airflow between cards that are densely packed together e.g. for 3-way SLI users or 2-way SLI users that don’t have a spare PCI slot between their cards. The small section may not look like much but Nvidia says it has been finely tuned to provide optimal airflow between cards, though without a second reference model we’re unable to test the impact on temperatures and/or fan speed.

Click to enlarge
The GPU is cooled by three copper heat pipes filled with high purity water that dissipate their heat through the main heatsink fins, which are anodised in black. There’s also a thick metal contact plate which, together with the numerous thermal pads, draws heat away from the all of the memory chips and power components (MOSFETs, capacitors and controllers) – it’s a very high quality and complete cooling solution.

Click to enlarge
A single radial fan is responsible for generating airflow. It draws air in both through itself and over the small set of aluminium fins at the card’s far side, as these are not sealed in. Air is then directed through the main fin stack before being expelled entirely through the rear I/O panel, which now has bigger gaps in so that more air can escape.

Click to enlarge
The PCB reveals a 4+1 phase power arrangement. Board partners are of course likely to develop their own power delivery systems for custom cards, but even this set-up is plenty for a 165W part. Nvidia claims that 1,400MHz is a very realistic target for these cards when overclocking, and we’ll of course be putting that to the test.

The GTX 980 includes a feature first introduced with the GTX 780 Ti, whereby it receives power evenly from all three sources (2 x 6-pin connectors and the PCI-Express bus). Overclocking can cause imbalance and less stable clock speeds, but onboard power balancing circuitry means that the card can start to draw more power from the other inputs should one become maxed out or imbalanced.

Click to enlarge
The Samsung memory has the product code K4G41325FC-HC28. These are 1.5V 4Gb parts, so eight are needed in total to make up the 4GB, and all eight are found on the front of the PCB. They are rated for data rates up to 7,000Mbps, which is handy since that is the speed they are clocked too. Hopefully they respond well to overclocking.

How We Tested

Our GPU test rig is equipped with the multi-GPU powerhouse that is the Asus Maximus V Extreme, allowing us to run up to 4-card SLI and CrossFire with 8x PCI-E 3.0 lanes per card. The CPU is an Intel Core i5 3570K running at 4.2GHz to raise the CPU headroom, and it’s paired with 8GB of 2,400MHz Corsair Dominator DDR3. Our chassis of choice is NZXT’s Switch 810, a case big enough to house even a pair of Asus’ Ares 2 graphics cards. The Lepa G1600 1600W PSU offers more wattage, 12V rails and 8-pin PCI-E power connectors than we’d ever need.

Our benchmarks are a mix of custom in-game time demos and manually-played sections, using FRAPS to record the average and minimum frame rates. We strive to not only record real-world performance you will actually see, but also present the results in a manner that is easy to digest. We test at 1,920 x 1,080 (1080p) and 2,560 x 1,440, as well as at 3,840 x 2,160 (4K) with higher end cards.

Test System

• Intel Core i5 3570K (operating at 4.2GHz – 42 x 100MHz)
• Asus Maximus V Extreme motherboard
• 2 x 4GB Corsair 2,400MHz DDR3 memory
• Lepa G1600 1600W PSU
• Windows 7 Home Premium 64bit
• Samsung SSD 830 256GB SSD

AMD graphics cards

• AMD Radeon R9 295X2 8GB (1,018MHz GPU, 5GHz memory) (Launch driver – Catalyst 14.4 beta)
• AMD Radeon R9 290X 4GB (1,000MHz GPU, 5GHz memory) (Catalyst 13.12)
• AMD Radeon R9 290 4GB (947MHz GPU, 5GHz memory) (Catalyst 13.12)
• AMD Radeon R9 280X 3GB (1,000MHz GPU, 6GHz memory) (Catalyst 13.12)
• AMD Radeon R9 285 2GB (918MHz GPU, 5.5GHz memory) (Launch driver – Catalyst 14.30.1005-Beta2)
• AMD Radeon R9 280 3GB (933MHz GPU, 5GHz memory) (Catalyst 14.4)
• AMD Radeon R9 270X 2GB (1,050MHz GPU, 5.6GHz memory) (Catalyst 13.12)
• AMD Radeon R9 270 2GB (925MHz GPU, 5.6GHz memory) (Catalyst 13.12)
• AMD Radeon R7 265 2GB (925MHz GPU, 5.6GHz memory) (Catalyst 14.4)
• AMD Radeon R7 260X 2GB (1,100MHz GPU, 6.5GHz memory) (Catalyst 13.12)

Nvidia graphics cards

• Nvidia GeForce GTX 980 4GB (1,126MHz GPU, boosting to 1,216MHz, 7GHz memory) (Launch driver – GeForce 344.07)
• Nvidia GeForce GTX Titan Black 6GB (889MHz GPU, boosting to 980MHz, 7GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 780 Ti 3GB (876MHz GPU, boosting to 928MHz, 7GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 780 3GB (863Mhz GPU, boosting to 900MHz, 6GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 770 2GB (1,046Mhz GPU, boosting to 1,085MHz, 7GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 760 2GB (980Mhz GPU, boosting to 1,033MHz, 6GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 660 2GB (980MHz GPU boosting to 1,033MHz, 6GHz memory) (GeForce 334.89 WHQL)
• Nvidia GeForce GTX 750 Ti 2GB (1,020MHz GPU boosting to 1,085MHz, 5.4GHz memory) (GeForce 335.23 WHQL)

Battlefield 4

Publisher: EA

From our Battlefield 4 review:

‘From the start to the end of the campaign you literally have no idea who’s who, why they do or don’t like each other and even what part you play in the whole thing – the latter point not being helped by the fact you’re mute throughout the whole game. You just meander from mission to mission and get on with the task set before you.

Still: who cares, right? Battlefield has long been about its multiplayer, and sure enough here, again the game truly shines. Largely it’s a continuation of what came before but there are enough extras that it feels, if not totally new, different enough to learn all over again.’

With its demanding Frostbite 3 engine, Battlefield 4 is a tough challenge for any GPU. We run the game at its highest ‘Ultra’ settings with motion blur at 50 percent and the resolution scale at 100 percent. We also manually disable MSAA at 4K, as it’s unnecessary and too demanding at this resolution. We run a 60 second benchmark on the game’s sixth campaign level, Tashgar, during the on rails section at the level’s start, and begin the recording as soon as the subtitle for the first line of dialogue appears on screen.

Click to enlarge

Battlefield 4

1,920 x 1,080, DirectX 11, ‘Ultra’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 270X 2GB
• AMD Radeon R9 270 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 102

  • 125

• • 66

  • 87

• • 66

  • 82

• • 64

  • 81

• • 59

  • 72

• • 57

  • 69

• • 55

  • 67

• • 47

  • 57

• • 45

  • 55

• • 39

  • 47

• • 38

  • 47

• • 37

  • 45

• • 36

  • 42

• • 32

  • 38

• • 30

  • 37

• • 29

  • 34

• • 23

  • 30

• • 22

  • 27

0

25

50

75

100

125

Frames Per Second

• Minimum
• Average

Battlefield 4

2,560 x 1,440, DirectX 11, ‘Ultra’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 980 4GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 270X 2GB
• AMD Radeon R9 270 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 75

  • 91

• • 44

  • 54

• • 43

  • 54

• • 42

  • 57

• • 40

  • 48

• • 38

  • 46

• • 37

  • 45

• • 30

  • 37

• • 30

  • 36

• • 26

  • 31

• • 25

  • 30

• • 24

  • 29

• • 23

  • 27

• • 21

  • 25

• • 19

  • 23

• • 19

  • 22

• • 15

  • 19

• • 14

  • 17

0

25

50

75

100

Frames Per Second

• Minimum
• Average

Battlefield 4

3,840 x 2,160, DirectX 11, ‘Ultra’ Settings, 0x AA

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 280X 3GB
• Nvidia GeForce GTX 770 2GB

• • 51

  • 67

• • 28

  • 41

• • 28

  • 36

• • 27

  • 35

• • 25

  • 39

• • 25

  • 39

• • 21

  • 31

• • 21

  • 28

• • 19

  • 27

0

10

20

30

40

50

60

70

Frames Per Second

• Minimum
• Average

BioShock Infinite

Publisher: 2K

From our Bioshock Infinite review :

“Your character Booker’s trip to Columbia is in the interest of retrieving a girl, Elizabeth, from captivity. It’s imperative that you find her so that you can clear a pile of debts, but it’s increasingly clear as your time in columbia continues that this isn’t a simple hero’s tale. Elizabeth has the power to control “tears in reality” that are popping up all over the city.”

“It’s an adult, thoughtful and compelling work that shames many (if not most) other attempts in the medium. Bioshock Infinite is an incredibly good video game. It might be one of the best.

We use the handy in-built benchmarking tool to run a timedemo from two sections near the start of the game. However, we’ve found the game-generated results to be unreliable, so use FRAPS to record the frame rate over a 40 second sequence of gameplay during the second test. The results taken are an average of three repeated tests.

Click to enlarge

BioShock Infinite

1,920 x 1,080, Ultra Detail with DOF

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280X 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 280 3GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 265 2GB
• AMD Radeon R9 260X 2GB

• • 153

  • 182

• • 117

  • 137

• • 101

  • 122

• • 100

  • 120

• • 88

  • 105

• • 87

  • 98

• • 82

  • 92

• • 72

  • 90

• • 67

  • 77

• • 66

  • 76

• • 58

  • 72

• • 56

  • 65

• • 49

  • 55

• • 45

  • 57

• • 44

  • 50

• • 40

  • 48

• • 39

  • 45

• • 32

  • 36

0

50

100

150

200

Frames Per Second

• Minimum
• Average

BioShock Infinite

2,560 x 1,440, Ultra Detail with DOF

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 285 2GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 280 3GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 110

  • 120

• • 73

  • 87

• • 69

  • 81

• • 68

  • 79

• • 56

  • 66

• • 56

  • 62

• • 53

  • 58

• • 46

  • 56

• • 42

  • 47

• • 42

  • 47

• • 37

  • 45

• • 36

  • 40

• • 30

  • 33

• • 28

  • 35

• • 27

  • 30

• • 24

  • 27

• • 23

  • 29

• • 19

  • 21

0

25

50

75

100

125

Frames Per Second

• Minimum
• Average

BioShock Infinite

3,840 x 2,160, Ultra Detail with DOF, 0x AA

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB

• • 53

  • 59

• • 35

  • 43

• • 35

  • 41

• • 34

  • 40

• • 27

  • 33

• • 27

  • 30

• • 26

  • 29

• • 23

  • 28

• • 20

  • 23

0

10

20

30

40

50

60

Frames Per Second

• Minimum
• Average

Crysis 3

Publisher: EA

Mixing the open-world combat of Crysis with the more tightly scripted urban jungle of Crysis 2, Crysis 3 is a smorgasbord of visual effects and polygons galore. With DirectX 11 support, high resolution textures and incredibly detailed characters models, it’s laid down the gauntlet for the next generation of consoles and games alike when it comes to gorgeous graphics

We test using the Very High detail preset and with Very High texture resolution. lens flare and motion blur are both enabled, although due to its heavy performance impact, anti-aliasing is disabled.

As explained earlier, we use a custom macro-driven 60 seconds play-through from the single player mission Red Star Rising. The 60 seconds of gameplay takes place in a large open environment heavy on water and particle effects. Each test is repeated three times, with the average result taken.

Click to enlarge

Crysis 3

1,920 x 1,080, 0x AA, 16x AF, ‘Very High’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 285 2GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 280 3GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 89

  • 113

• • 64

  • 77

• • 61

  • 75

• • 61

  • 73

• • 54

  • 66

• • 52

  • 64

• • 50

  • 62

• • 46

  • 55

• • 40

  • 50

• • 37

  • 48

• • 36

  • 45

• • 35

  • 43

• • 31

  • 37

• • 29

  • 35

• • 28

  • 34

• • 25

  • 31

• • 23

  • 27

• • 21

  • 25

0

25

50

75

100

125

Frames Per Second

• Minimum
• Average

Crysis 3

2,560 x 1,440, 0x AA, 16x AF, ‘Very High’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 270X 2GB
• AMD Radeon R9 270 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 64

  • 80

• • 42

  • 49

• • 39

  • 47

• • 38

  • 46

• • 36

  • 43

• • 34

  • 41

• • 33

  • 41

• • 29

  • 35

• • 28

  • 33

• • 25

  • 31

• • 23

  • 28

• • 23

  • 28

• • 19

  • 24

• • 18

  • 22

• • 18

  • 21

• • 16

  • 20

• • 13

  • 17

• • 13

  • 16

0

10

20

30

40

50

60

70

80

Frames Per Second

• Minimum
• Average

Crysis 3

3,840 x 2,160, 0x AA, 16x AF, ‘Very High’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 780 3GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB

• • 34

  • 40

• • 19

  • 23

• • 18

  • 22

• • 18

  • 22

• • 17

  • 22

• • 17

  • 21

• • 16

  • 19

• • 14

  • 17

• • 13

  • 16

0

10

20

30

40

Frames Per Second

• Minimum
• Average

The Elder Scrolls V: Skyrim

Publisher: Bethesda

From our The Elder Scrolls V: Skyrim review:

‘It’s like watching Star Wars and genuinely thinking, ‘what about those poor Death Star construction workers?’ You’re missing the point: Skyrim is a huge and engaging world to explore and it treats you with great moments, from your first dragon encounter to finally being able to craft dwarven armour.’

We’ve updated our Skyrim benchmark to include the official high resolution texture pack, available as a free DLC. We set the game to its ‘Ultra’ setting and record a sixty second manual play through just outside the town of Whiterun during a thunderstorm. We use a section where we are able to run forward in a straight line for a minute without being attacked so the benchmark remains consistent, and use the third person camera view.

Click to enlarge

The Elder Scrolls V: Skyrim

1,920 x 1,080, 8x AA, 16x AF, ‘Ultra’ Settings, w/ High Res Texture Packs

• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 270X 2GB
• AMD Radeon R9 270 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 660 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB

• • 117

  • 170

• • 109

  • 160

• • 106

  • 141

• • 105

  • 145

• • 102

  • 133

• • 87

  • 120

• • 85

  • 103

• • 82

  • 102

• • 77

  • 94

• • 74

  • 99

• • 74

  • 90

• • 64

  • 82

• • 63

  • 77

• • 62

  • 79

• • 46

  • 59

• • 45

  • 58

0

25

50

75

100

125

150

175

Frames Per Second

• Minimum
• Average

The Elder Scrolls V: Skyrim

2,560 x 1,440, 8x AA, 16x AF, ‘Ultra’ Settings, w/ High Res Texture Packs

• Nvidia GeForce GTX 980 4GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R7 265 2GB
• Nvidia GeForce GTX 750 Ti 2GB

• • 97

  • 127

• • 61

  • 74

• • 58

  • 69

• • 55

  • 73

• • 46

  • 56

• • 32

  • 40

0

25

50

75

100

125

Frames Per Second

• Minimum
• Average

The Elder Scrolls V: Skyrim

3,840 x 2,160, 8x AA, 16x AF, ‘Ultra’ Settings, w/ High Res Texture Packs

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 780 3GB

• • 82

  • 119

• • 57

  • 72

• • 56

  • 72

• • 56

  • 71

• • 54

  • 69

• • 50

  • 65

• • 50

  • 64

0

25

50

75

100

125

Frames Per Second

• Minimum
• Average

Unigine Valley 1.0 Benchmark

Publisher: Unigine

We’ve also updated our testing to include Unigine’s free Valley 1.0 benchmarking tool. It works well as a graphics benchmark as it is GPU limited and is thus incredibly taxing on the GPU whilst placing the CPU under very little stress.

Until now, we’ve only run the full benchmark at 2,560 x 1,440. However, following feedback in our forums, we’ll now be including Unigine Valley results at 1,920 x 1,080 too, using the same settings (fullscreen, AA off, Ultra detail). This means many more users can directly compare their results to ours, as 1080p is by far the more popular resolution. Unigine’s scoring system is effectively linear: a card with 2,000 points is considered twice as fast as one with 1,000 points, and half as fast as one with 4,000 points. We’ll be retesting and adding more cards to the 1080p chart in the near future.

Unigine Valley 1.0

1,920 x 1,080, 0x AA, ‘Ultra’ Quality

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 265 2GB

• • 5528

• • 4890

• • 3919

• • 3914

• • 2667

• • 2497

• • 2444

• • 1647

• • 1647

0

1000

2000

3000

4000

5000

6000

Score (higher is beter)

• Score

Unigine Valley 1.0

2,560 x 1,440, 0x AA, ‘Ultra’ Quality

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 290 4GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 285 2GB
• AMD Radeon R9 280 3GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270 2GB
• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 265 2GB
• AMD Radeon R7 260X 2GB

• • 4313

• • 3047

• • 3000

• • 2947

• • 2487

• • 2378

• • 2210

• • 2135

• • 1760

• • 1657

• • 1508

• • 1508

• • 1284

• • 1263

• • 1148

• • 981

• • 967

• • 846

0

1000

2000

3000

4000

Score (higher is beter)

• Score

Power Consumption (Idle and Gaming)

In order to get an idea of a GPU’s power draw, we run our Unigine Valley 1.0 benchmark, which is our most GPU limited test. We use a watt meter to measure the maximum total system power draw during the test, and also take an idle reading at the Windows desktop (2,560 x 1,440).

Power Consumption

Idle (Windows 7 Aero Desktop) and Load (Unigine Valley 1.0)

• Nvidia GeForce GTX 750 Ti 2GB
• AMD Radeon R7 260X 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270 2GB
• AMD Radeon R7 265 2GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 760 2GB
• AMD Radeon R9 285 2GB
• Nvidia GeForce GTX 980 4GB
• AMD Radeon R9 280 3GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R9 280X 3GB
• Nvidia GeForce GTX 780 3GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX Titan Black 6GB
• AMD Radeon R9 290 4GB
• AMD Radeon R9 290X 4GB
• AMD Radeon R9 295X2 8GB

• • 100

  • 174

• • 103

  • 222

• • 103

  • 227

• • 107

  • 239

• • 114

  • 242

• • 107

  • 252

• • 110

  • 292

• • 107

  • 297

• • 104

  • 299

• • 108

  • 305

• • 106

  • 338

• • 111

  • 350

• • 108

  • 373

• • 111

  • 380

• • 106

  • 384

• • 114

  • 401

• • 114

  • 409

• • 128

  • 665

0

100

200

300

400

500

600

700

System Power Draw in Watts (lower is better)

• Idle
• Load

Thermal Performance (Idle and Gaming)

Thermal output is also measured using Unigine’s Valley 1.0 benchmark, as its DirectX 11 features will stress all parts of a modern GPU. We leave all GPU fan profiles and settings as they come and leave the benchmark running for ten minutes so that temperatures plateau. We record the peak GPU temperature using GPU-Z, and present the data as the delta T (the difference between the GPU temperature and the ambient temperature in our labs). We also take an idle reading at the Windows desktop (2,560 x 1,440).

Heat

Idle (Windows 7 Aero Desktop) and Load (Unigine Valley 1.0)

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 750 Ti 2GB
• Nvidia GeForce GTX 660 2GB
• AMD Radeon R9 270X 2GB
• Nvidia GeForce GTX 770 2GB
• AMD Radeon R7 260X 2GB
• Nvidia GeForce GTX 760 2GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX 780 3GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX Titan Black 6GB
• AMD Radeon R9 290 4GB
• AMD Radeon R9 290X 4GB

• • 10

  • 35

• • 5

  • 41

• • 9

  • 41

• • 8

  • 50

• • 10

  • 55

• • 8

  • 56

• • 7

  • 57

• • 13

  • 58

• • 10

  • 59

• • 14

  • 59

• • 9

  • 61

• • 19

  • 72

• • 19

  • 72

0

10

20

30

40

50

60

70

80

Delta T in °C (lower is better)

• Idle
• Load

*As we do not have stock samples of the AMD Radeon R9 280X, R9 285, R9 280, R9 270 or R7 265, they are not included in the thermal performance graphs.

Nvidia GeForce GTX 980 Overclocking Results

We used the latest version of MSI Afterburner to overclock the GTX 980, as we found it to have the best compatibility with the new chips and it’s very easy and clear to use.

We began by maxing out the power and temperature limits to give us the maximum amount of headroom possible. These values were set to 125 percent and 91°C respectively.

After our session of trial and error, we finished on a stable base clock of 1,327MHz, giving us a rated boost clock of 1,416MHz. In practice, the card was happily boosting to between 1,425MHz and 1,450MHz, which is an incredible feat for a reference GPU, and shows just how much headroom Maxwell really has. The 200MHz base clock increase is a tasty 18 percent boost. We fed the GPU with an extra 40mV in Afterburner to keep things stable at these speeds.

This healthy core overclock was joined by a solid memory overclock too. The eight Samsung memory chips were happy to run up to 7.8GHz, an excellent figure for chips only rated to 7GHz in the first place – we’ve never yet managed to break the 8GHz point, so we can’t ask for much more here.

Overclocking – Battlefield 4

2,560 x 1,440, DirectX 11, ‘Ultra’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB (OC)
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 980 4GB
• AMD Radeon R9 290X 4GB
• Nvidia GeForce GTX 780 3GB

• • 75

  • 91

• • 49

  • 66

• • 44

  • 54

• • 43

  • 54

• • 42

  • 57

• • 40

  • 48

• • 38

  • 46

0

25

50

75

100

Frames Per Second

• Minimum
• Average

Overclocking – Crysis 3

2,560 x 1,440, 0x AA, 16x AF, ‘Very High’ Settings

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB (OC)
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• AMD Radeon R9 290X 4GB

• • 64

  • 80

• • 49

  • 57

• • 42

  • 49

• • 39

  • 47

• • 38

  • 46

• • 36

  • 43

0

10

20

30

40

50

60

70

80

Frames Per Second

• Minimum
• Average

Overclocking – Unigine Valley 1.0

2,560 x 1,440, 0x AA, ‘Ultra’ Quality

• AMD Radeon R9 295X2 8GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX 980 4GB
• Nvidia GeForce GTX Titan Black 6GB
• Nvidia GeForce GTX 780 Ti 3GB
• Nvidia GeForce GTX 780 3GB
• AMD Radeon R9 290X 4GB

• • 4313

• • 3565

• • 3047

• • 3000

• • 2947

• • 2487

• • 2378

0

1000

2000

3000

4000

Score (higher is beter)

• Score

Performance Analysis

Overall, looking at the performance and power figures, the GTX 980 does indeed assert itself as both the fastest and most efficient GPU currently available.

In Battlefield 4, the GTX 980 matches GTX Titan Black on the minimum frame rate (66fps) but trumps it on average at 1080p. Moving up to 1440p, however, sees the Titan Black and the GTX 780 Ti overtake the GTX 980 on the minimum frame rate. It’s not a major difference, but it may well be due to the GTX 980’s lower memory bandwidth; there’s a particularly gruelling section of this benchmark where a character’s face is very close to the player’s viewpoint, and the texture and detail on it is very high at and above this resolution. It is the section where all cards dip to their minimum fps numbers, and it may be that the greater bandwidth on the two GK110 parts helps them stay slightly faster. However, at 4K, the GTX 980’s 28fps minimum compares favourably to the 25fps of those two cards. The R9 290X and R9 290 also do well here with 28fps and 27fps respectively. As such, it seems likely that the combination of 64 ROPs and a 4GB of frame buffer are the crucial element for success at 4K here – only cards that match both criteria stay above 25fps.

Click to enlarge
In BioShock Infinite, it’s a simpler story. The GTX 980 is the fastest single GPU card at every reoslution, where it is at least 30 percent ahead of the R9 290X. The GTX Titan Black manages to match its minimum frame rate of 35fps at 4K, but the GTX 980 still wins out on average.

Crysis 3 again gives pole position to the GTX 980 every time, and this time its minimum frame rate is never matched. This time, however, it isn’t quite as fast compared to the R9 290X – its advantage sits somewhere between 17 and 19 percent at playable resolutions, which is still a significant margin.

Our final game, Skyrim, is of course no challenge for the GTX 980. It averages over 120fps at 1440p with all settings at full, and at 4K it’s almost able to hit a constant 60fps, though it doesn’t offer any performance advantage over the GTX 780 Ti or GTX Titan Black here, and its lead over the R9 290X is a meaningless 2fps.

Click to enlarge
The Unigine Valley 1080p result is mostly just so that most people can easily compare their own set-up to the results here. However, with a score of 4,890, the GTX 980 trumps both the R9 290X and the GTX 780 by a juicy 25 percent. Of course, you don’t buy a card like this to play on a 1080p screen, so it’s good that even at 1440p the GTX 980 is still the single GPU king, albeit only slightly, as it beats the GTX Titan Black and GTX 780 Ti by just 47 and 100 points respectively. Meanwhile, it has 23 percent and 28 percent leads on the GTX 780 and R9 290X respectively.

The power consumption figures are frankly phenomenal. The GTX 980 just manages to land itself on the right side of 300W (total system power consumption) under load, which makes it incredibly efficient compared to anything else on the chart (except the GTX 750 Ti). Its figure of 299W is more than 80W below that of the GTX 780 Ti and GTX Titan Black, both of which it manages to match or even beat in nearly every benchmark. Better yet, it is 110W less than the R9 290X, which is beats by up to 30 percent in some tests. Granted, even for hardcore gamers this difference isn’t going to drastically alter your annual energy bill, but it’s still not an insignificant difference. The two flagship cards couldn’t be much further apart when it comes to performance per watt.

Click to enlarge
The thermal performance is exactly as expected. As with the GTX 780, GTX 780 Ti and GTX Titan Black, the GTX 980’s fan only spins as fast as is needed to keep the card at 80°C, resulting in a delta T of around 60°C for all four. Again, this is lower then the R9 290 cards, which run at 95°C by default. The card’s cooler under load is audible, but quiet, and there’s no noticeable noise from the power components either.

The 18 percent core and 11 percent memory overclocks were certainly very pleasing. At 1440p, the new speeds were enough to boost performance by 17 percent in all three benchmarks (Battlefield 4, Crysis 3 and Unigine Valley). Also, despite the increased temperature limit, the core temperature only climbed by 2 or 3°C and though the fan had to work harder it wasn’t by much – certainly not enough to make it loud or annoying. Our system power consumption did increase by 51W to 350W under load (remember we increased the card’s power limit to 125 percent), but even this figure is still below the GTX 780 and well below the R9 290-series cards, both of which it utterly dominates when running at these frequencies.

Conclusion

Without doubt, the GTX 980 is a great graphics card. By and large, it is unmatched in the performance arena, and there’s certainly nothing that even comes close to it at this end of the market when you look at performance per watt. It is also very well crafted and it looks and feels like a true premium product.

Of course, it has to be said that performance per watt is not always a great concern in the enthusiast desktop space, especially at this high end of the market, but equally we’re certain that the GTX 980’s efficiency and overclocking credentials are going to win over a good number of people, especially those who are interested in quiet computing and/or small form factor builds, as here the lowered heat output is highly beneficial. These are also two areas that are not well suited to the hot-running and power hungry AMD Hawaii-based cards.

Click to enlarge
With all that said, we can’t ignore the pricing. The MSRP puts the GTX 980 at £429, which is a serious amount of cash to splash on a single GPU; we think it would have looked far more appealing had Nvidia managed to get it to below £400. As it stands, it’s good value compared to the GTX 780 Ti, which still retails for around £460, though only a fool would consider the GTX 780 Ti a benchmark for good value. Meanwhile, the GTX 780 and R9 290X tend to go for about £330, though some are even closer to the £300 mark. This means the GTX 980 would need to offer performance that was consistently 30 percent higher than these cards to be aligned with their value. As we saw, in a handful of cases, it does do so, but the majority of times it does not; the advantage is likely to be closer to half that, especially when you’re playing at the higher resolutions these cards are designed for.

Still, performance isn’t all that you’re paying for. There’s the efficiency for one thing, but also new technologies and future proofing in the form of things like the, HDMI 2.0, multiple DisplayPort outputs, H.265 encoding, G-Sync (though bear in mind AMD’s FreeSync movement is picking up traction and support) and hardware acceleration of exciting new graphics processing techniques, including DirectX 12 features.

Click to enlarge
Essentially, if you can afford to, and you buy this card, you’re going to receive it and be one happy bunny. However, if you’re chasing value for money, then it’s not your best option; you’ll definitely want to take a look at our round-up of three GTX 970 cards right here though.

SOURCE:http://www.bit-tech.net/hardware/graphics/2014/09/19/nvidia-geforce-gtx-980-review/1