NVIDIA GTX 980 Review

NVIDIA finally release their next-generation graphics card. Join us as we unveil their new flagship GPU based around the GM204’s 28nm Maxwell architecture.

Product on Review: GeForce GTX 980
Manufacturer & Sponsor: NVIDIA
Price: £429 GBP

NVIDIA’s Maxwell architecture, a successor to the Kepler architecture seen in the majority of 600- and 700-series GPUs, was first introduced in February with the GTX 750 & 750Ti. However these are entry-level gaming parts which, despite showing some cool technology and excellent power efficiencies, weren’t tempting for many enthusiasts; rightly so in fact, they still had the GTX Titan Black and later Titan Z to dine out on. Maxwell has therefore been only at the back of our minds – something to look forward to, but with trepidation over whether NVIDIA could scale up the promise of the GTX 750 to a far bigger GPU.

It’s theorised that the Maxwell architecture was originally designed for the 20nm process node, a node which has apparently been delayed well beyond original projections. NVIDIA couldn’t afford to wait on TSMC, especially with new graphics standards and use cases such as DirectX 12 and Occulus’ Rift Consumer Beta arriving in the not-to-distant future, necessitating the release of a major architecture revision without a process node shrink. Whether or not it was designed for 20nm, 28nm was what they had to work with.

To put that into perspective, this set of circumstances hasn’t occurred since the pre-Tesla (200-series) era, and effectively never for major performance parts. That’s a big deal; often processor performance boosts have been dependant on the reduced voltages, higher frequencies and denser transistor packing made possible with a process node shrink.

Immediately therefore the question looms in everyone’s mind: without a process shrink can performance Maxwell parts show any sort of edge over their Kepler counterparts. Well, as it turns out, yes they can. But let’s not get ahead of ourselves.

Released today are the first two such performance parts: the GeForce GTX 980 and GeForce GTX 970. Note the leap in numbering – the 900-series was chosen over the 800-series to bring their desktop and mobile architectures in line over the next few releases. Given the flack GPU manufacturers have taken for re-labelling old cards in new line-ups that’s a welcome decision.

Although NVIDIA are claiming major performance increases over Kepler (and AMD’s lineup, naturally) the most technologically significant improvement is in power consumption. The required number of PCIe connectors tips you off – two six-pin parts places this on the order of a 200W part. In fact the GTX 980 has a TDP of 185W, which even accounting for the different ways manufacturers define these figures is a major step forward in perf/watt.

The flip-side of course is NVIDIA’s pricing scheme. In the past their launch pricing has reflected the performance compared to the completion, which naturally has tended to gradually increase the amount consumers have had to pay over time. This time the new flagship is launched with an MSRP of £429 (although pricing will vary based on partner costs), close to a £100 premium over the GTX 780 and over £170 more than the GTX 970 (which is a relatively benign £259); comparisons with the GTX 780 Ti are difficult due to the low stock levels across the UK. Whilst this is tempered by no additional associated outlay – for many an existing ~600W PSU will be more than enough to power each card – it will tend to diminish the urge many will have to upgrade.

For their part NVIDIA feel that both new GPUs represent value for anyone upgrading from the 600-series and earlier, especially premium 500-series parts. Not only is this due to higher performance but also new technologies such as G-SYNC, Shadowplay and more which were introduced with Kepler and have been expanded upon with Maxwell.

We’ll get to the performance results later in the review, but before then it’s worth going through the plethora of new and thus far unique features introduced with Maxwell, some of which are available immediately and some still awaiting final validation.

Specification

Even at a surface level analysis the GTX 980 is a major step up over the GTX 680/770, which you’ll find many sites referring to in the briefing material for the new Maxwell GPUs. Most notable is the increased CUDA core count, up by a third to 2048, but also intriguing is the increased number of ROPs (up from 32) and Level 2 Cache (not listed, but increased to 2MB from 512KB). In theory this will aid both overall rendering speed and anti-aliasing performance significantly, a definitely plus for image quality enthusiasts.

Comparisons to the GTX 780 are more difficult to make. The number of shaders is down from 2304, as is the memory bus width. However the core frequency, memory frequency and amount of video memory is significantly higher. Guessing just how close the two cards will be in terms of performance is difficult, however by naming it the GTX 980 it’s likely that NVIDIA intends the card to at least go toe-to-toe with the smallest of the Big Kepler parts.

However, as we’ve mentioned the most eye-opening figure is that TDP – 165W. We can see where the card makes its savings by taking a look at the GM204 block layout.

GK104 organised the GPU by way of four graphics processing clusters (GPCs), and the GM204 does much the same. However whilst Kepler split each GPC into two streaming multiprocessors known as SMXs Maxwell has a much more aggressive partition: four SMMs of 128 CUDA cores each with yet further internal partitions. Interestingly the GM107 (seen on the 750Ti) made use of five SMMs per GPC. The Geforce GTX 980 is a fully unlocked GM204 part, whereas the GTX 970 makes use of 13 SMM and a total of 1664 CUDA cores.

The original GK104 was the GPU within four discrete 600-series parts including both GTX 660 variants and the GTX 670. Having proved so successful NVIDIA are highly likely to use similar measures to flesh out the rest of the 900-series mid-to-high-end card range; in essence, expect a GTX 960/960Ti based on the GK204 in the not too distant future.

Each Maxwell SMM is subdivided four ways into what can be thought of as a further sub-SM, such that each SMM now has four sets of instruction buffers, warp schedulers, dispatch units and registers rather than a single set per SMX in Kepler. This, in addition to offloading some scheduling to the CPU where appropriate, is why the card can be so power-efficient and perhaps more importantly make use of resources more intelligently.

One further development NVIDIA have made to the memory architecture is the implementation of Third Generation Delta Colour Compression. This new lossless algorithm allows them to make better use of the memory bandwidth available, which is essential for higher resolution rendering, texture sizes and IQ settings. Benefits realised by using this algorithm largely depend on the scene but internally NVIDIA estimate that the GTX 980 has between 17 and 25% more effective use of memory bandwidth than the GTX 680.

As well as hardware improvements NVIDIA have also implemented new technologies at the driver level which can make better use of Maxwell’s resources, perhaps most notably DSR and MFAA…

Features – DSR & MFAA

Dynamic Super-Resolution

Super Sampling Anti-Aliasing (SSAA) isn’t a new concept, but its adoption in games or graphics hardware over the years has been patchy due how processor-intensive the process is. It relies on sampling the scene at a resolution much higher than your native screen resolution and then blending pixels together to output a ‘normal’ resolution frame, reducing ‘jaggies’. It’s often colloquially referred to as the ‘king of AA techniques’, but has given way to MSAA, MLAA and NVIDIA’s own FXAA, each of which are faster and less computationally expensive even if the quality isn’t as good.

Progress in graphics technology and horsepower has meant that a game which would bring a GPU to its knees five years ago can be comfortably played on mid-range hardware these days, whilst top-tier hardware push frame rates well into the hundreds. That’s a lot of horsepower going to waste when it could, if the game supported it, be used to improve image quality significantly.

Relatively recently enthusiasts have used a homebrew technique to increase the resolution which a game engine renders at, and then apply post-processing techniques to reduce the output resolution down to the monitor’s native. Known as downsampling, it’s essentially a form of SSAA. Unfortunately it’s very hit-and-miss, often relying of 3rd-party applications and prone to throwing up OS-level errors; and of course it’s very computationally expensive. However the rewards – beautiful images from older game engines – make the effort worth it.

Today GPUs such as the GTX 780Ti and Titan are being produced which are capable of and designed for 4K resolutions in even modern titles. However most gamers still use 1080p monitors on a day-to-day basis, making a $500+ GPU obscene overkill for all but the most taxing of titles. Because of this NVIDIA are going to start to support downsampling at the driver level through a method they’re calling Dynamic Super-Resolution (DSR).

Composite scene, DSR sample on right

Using DSR (which can be enabled automatically through GeForce Experience or the driver application) the game is rendered at a higher resolution and then downscaled to the monitor native resolution. Unlike previous methods however tt’s fully integrated into the driver, meaning not only ubiquitous game support but also a far more stable process than the bespoke solutions previously used.

Whilst the potential image quality benefits are significant it’s also worth noting that DSR would be an excellent measure of assessing whether your setup is capable of supporting a higher resolution monitor in gaming. The performance impact of the downsampling process is apparently only 1-2% (i.e less than a frame per second), so the framerates you get running DSR at 4K on your 1080p monitor should be extremely close to those you get for 4K natively. We’re not sure if NVIDIA intended this, but we’ll certainly take it.

DSR is part of NVIDIA’s launch driver for the GTX 980 and 970, but will also be rolled down to Kepler in the near future. The current implementation is still a little rough around the edges and NVIDIA haven’t supplied a software means of easily showing the downsampled output frame as all in-engine screenshots will be at the supersample resolution, but it shows a great deal of promise.

Multi-Frame Anti-Aliasing

Optimised anti-aliasing techniques is one of the ways GPU manufacturers have sought to differentiate themselves beyond pure frame rates and GFLOP statistics. Recognising that MSAA was insufficient to the needs of shader-based rendering and SSAA was too expensive both AMD and NVIDIA have generated their own methods to combat aliasing, to greater or lesser success. NVIDIA’s most well-known iteration is Fast Approximate Anti-Aliasing (FXAA), initially requiring in-engine implementation but later enabled at the driver level. More recently Temporal Anti-Aliasing was debuted, cleaning up an image based on frame-by-frame changes to pixels around edges. Each has their pros and cons; a good way to get into a fight on enthusiast boards is to claim that one method is flat out better than the other.

Today Multi-Frame Anti-Aliasing (MFAA, not to be mistaken for AMD’s Morphological Filtering AA) is being announced as a Maxwell-exclusive feature. As the name suggests MFAA is a post-processing anti-aliasing technique which operates by analysing a frame within the context of the previous frame, checking aliasing and rendering colours based on a proprietary algorithm in hardware.

The foundation of MFAA is multi-pixel programmable sampling, a means by which the matrix of sampling positions can be randomised to reduce data processing artefacts caused by repeating the same sample position from frame-to-frame.

MFAA seeks to be as good as MSAA, but without the performance impact algorithms and with image characteristics that are substantially different from FXAA. The latter has garnered a reputation for slightly blurring images, reducing the overall image quality, and so an effective alternative is probably due at this point.

Sadly MFAA is not quite ready for prime-time and will be issued as part of a driver update at a later date. The obvious question-mark is just how high the image quality presented is, and if it can operate well when the differences between frames is relatively large. However NVIDIA have clarified that the technique will work in tandem with alternate-frame rendering, an SLI mode available when using two or more matched NVIDIA GPUs.

First Look

The GTX 980 looks near identical to both the TITAN, and GTX780 variants we saw in the last round of NVIDIA card releases so there is no improvement in aesthetics. Not that it is needed you understand because we really do like this design with the perspex window framed with a fully enclosed silver shroud complete with contrasting black inserts. If it isn’t broke then why fix it?

The addition of a backplate is a very smart move by NVIDIA. No longer are the components exposed to falling screws, errant water drips from poorly sealed CPU barbs! The printed circuitboard designs is also attractive to look at and while it may look plastic, it actually painted aluminium. Very nice.

The same illuminated ‘GeForce’ profile is found on the GTX 980 which again we liked. illumination colour choice is however limited to green, green or green. So if you like green, you’re in luck. If not, you will have to learn to love it.

The cards dimensions work out at:

To power the card you will need a 550W PSU with twin 6pin (75w) PCIe Power cables.

Being a range topping card, SLI is a pre-requisite. Sadly, NVIDIA are still using physical bridges to link up more than one GPU rather than electronic as with their competitors, AMD.

The I/O area is a change from the norm with 3 full size DisplayPorts being provided along with both a HDMI and DVI port. The I/O shiled is also a little different having a crosshatch/cheese wheel style vent.

Closer Look

There is a small insert in the backplate that can be removed without the need to remove the whole of the backplate. We are not entirely certain what this is for other than to access a couple of voltage read points but it could be a hint at a later design that serves no purpose with the GTX980.

Upon removal though it became clear the backplate was more for aesthetic/protection purposes rather than for any cooling benefit as there are no components required to be cooled on the rear of the PCB.

Removing the cooler was an arduous affair with a number of screws required to be removed for the backplate which is is screwed through to the main heatsink. The heatsink itself is a mirror of the previous design with a vapor chamber linked to the main finned heatsink module. Air passes through these fins from the fan only to be exhausted out of the rear of the case. This is a design we prefer as it prevents hot air from being re-circulated within the PC case thereby heating up other components.

The first thing that struck us about the PCB design was not what was visible but what was evident by it’s omission, namely the missing chokes and mosfets which leads us to believe these areas are reserved for an as yet to be unveiled GTX 980Ti?

The VRM features a 4+2+1 design with space for an extra two power phases. This is less than both the 780Ti and TITAN so clearly there have been power efficiences made here which hopefully, will not affect any overclocking exploits.

A revised voltage controller is used on the GTX 980. Sadly our version had no markings however we can tell you that this device will allow the end user to monitor and adjust voltage through supported software.

Memory comes courtesy of Samsung and carries the product code of K4G41325FC-HC28 which tells us the GDDR5 SGRAM is packaged as 4GBit, is rated to 3500MHz at 1.5v.

Finally we reach the beating heart of the GTX 980.

Test Setup & Methodology

How we test graphics cards

Testing a graphics card is an extremely lengthy process. We sometimes see reviews where a graphic card has been tested using out of date equipment, out of date games and even out of date drivers. While the latter can be excused due to the sheer amount of driver updates making a thorough review impossible we will however notify the reader when we have used a different driver to the normal for example a pre-release driver. For the most part however we will always use the same hardware and latest drivers so this does certainly need to be taken into consideration when viewing our testing results. Each game and benchmark is tested times with the lowest and highest scores omitted out of the 5 results and the average then taken from the three remaining to give us our final result for each resolution and setting used.

HardwareWith the latest Intel chipset release we decided to update our ageing X79 setup to the very latest X99 chipset. Not only that but we will be using the flagship Core i7-5960X Extreme Edition processor clocked to a realistic, achieveable 4GHz to ensure we have no CPU bottlenecks. With 16GB (4x4GB) of GDDR4 at 2666MHz and a 500GB SSD there should be minimal paging so the benchmarks we run today will hopefully give a true reflection of the graphics card’s performance.

High End Graphics cards on test:

NVIDIA GTX 980 4GB GDDR5
(Forceware 340.52)
SAPPHIRE TRI-X R9-290X 4GB GDDR5(Catalyst 14.3)

Special thanks go to Intel, Corsair, MSI and AOC for providing the extreme spec components used for our test bench:

Motherboard: MSI X99S GAMING 9 AC
CPU: Intel Core i7-5960X Extreme Edition @ 4GHz
RAM – 16GB (4x4GB) Corsair LPX Vengeance (Red) DDR3 @ 2666MHz 16-17-17-35 C1
Power Supply – CORSAIR RM Series™ RM1000
Hard Drive – CORSAIR Force Series™ LX 512GB SATA 3 6Gb/s SSD
Cooler – CORSAIR Hydro Series™ H105 240mm Extreme Performance Liquid CPU Cooler
Monitor – AOC 28″ 4K U2868 PQU

Benchmarks
If you are a regular reader of our reviews you will know we like to test the latest hardware with the latest games and benchmarks on the market. Here are the games we have chosen:

Battlefield 4
Watchdogs
Company of Heroes 2
Thief
GRID: 2
Metro: Last Light

We also take into consideration the benchmarkers out there so have included 3 of the more popular synthetic benchmarks available:

Unigine: Valley
Unigine: Heaven 4
Futuremark 3DMark Firestrike

Let’s see how today’s graphics card performed…

Temperature, Acoustics and Power Consumption

Power Consumption

To test for power consumption we take a measurement from the plug socket. It is important to note that the figures below represent total system power use and not just the GPU. The CPU was always in a fixed state (4GHz overclock) so did not affect the additional power load when the GPU was placed under 100% load using Furmark which was ran for 15 minutes at which point a load reading was taken. Idle readings were taken 5 minutes after system boot to ensure any background services and applications had loaded.

As you can see above, the power figures were extremely good, especially when compared to the competition. Clearly Maxwell is a very power efficient architecture.

Temperature

To test temperatures we measured idle temperatures after booting windows, letting all applications finish loading and ran a few benchmarks. Once the benchmarks were complete we left the card to reach a cooling plateau where we then took the idle temperatures. For the load tests, we would normally run Furmark for 20 minutes, taking the absolute maximum temperature attained however we found that this throttled the card and resulted in spurious results. So we set Heaven running continuously for 20 minutes and used this as a temperature result.

Not exactly efficient cooling here. While not too bad at reference clockspeeds, the card did start to suffer (and throttle back) when we overclocked the card to its limit. We fully expect aftermarket coolers to rectify this though however it is certainly worth mentioning that the ‘old’ cooler is nearing it’s limit with the current architecture.

Acoustics

As this is the same cooler used on previous examples, we were just as impressed as we were previously. However, we had to adjust the fan profile when overclocking as the temperatures were getting a little uncomfortable. This in turn increased the noise levels.

Overclocking

For overclocking today’s sample we will be using MSI’s Afterburner tool as this gives us all of the options we require included the additional voltage controls.

Above is a screenshot of GPU-Z showing the reference (factory overclocked) settings.

…and manually attained overclock speeds.

Overclocking the GTX 980 was very easy. We maxxed out the voltage controls and found the core limit (+295MHz). We then put the core back to stock and found the memory limit (+360MHz). After testing with a few runs of Unigine heaven we then adjusted both to gain an equilibrium and found our best run at +280MHz on the core coupled with +335MHz on the memory.

Above we can see the resulting performance increase of the overclock via Unigine Heaven 4.0.

Futuremark 3DMark Fire Strike (Synthetic)

Fire Strike is the new showcase DirectX 11 benchmark designed for high-performance gaming PCs. It is Futuremark’s most ambitious and technical benchmark ever, featuring real-time graphics rendered with detail and complexity far beyond what is found in other benchmarks and games today.

Unigine: Heaven 4 (Synthetic)

Heaven Benchmark with its current version 4.0 is a GPU-intensive benchmark that hammers graphics cards to the limits. This powerful tool can be effectively used to determine the stability of a GPU under extremely stressful conditions, as well as check the cooling system’s potential under maximum heat output. It provides completely unbiased results and generates true in-game rendering workloads across all platforms.

Unigine: Valley (Synthetic)

Valley Benchmark is a new GPU stress-testing tool from the developers of the very popular and highly acclaimed Heaven Benchmark. The forest-covered valley surrounded by vast mountains amazes with its scale from a bird’s-eye view and is extremely detailed down to every leaf and flower petal. This non-synthetic benchmark powered by the state-of-the art UNIGINE Engine showcases a comprehensive set of cutting-edge graphics technologies with a dynamic environment and fully interactive modes available to the end user.

Battlefield 4 (FPS)

Battlefield 4 is a 2013 first-person shooter video game developed by Swedish video game developer EA Digital Illusions CE (DICE) and published by Electronic Arts. It is a sequel to 2011’s Battlefield 3 and is perhaps currently the most popular online First Person Shooter for PC gamers. It has an extremely powerful gaming engine that will stress components more than most FPS titles available.

Watchdogs (Action/Adventure)

Watch Dogs is a 2014 open world action-adventure video game developed by Ubisoft Montreal and published by Ubisoft. Set within a fictionalized version of Chicago, Illinois, the single-player story follows a hacker and his efforts to seek revenge after the accidental death of his niece. The open world design lets players freely roam Chicago, which includes the urban city, open countryside, and slums. The game is played from a third-person perspective and its world is navigated on-foot or by vehicle. As such this is a very demanding game, especially when graphical options are turned up.

Thief (Stealth)

Thief is a stealth video game developed by the Canadian video game developer Eidos Montreal, and published by Square Enix. It is a revival of the cult classic Thief series of stealth games, of which it is the fourth game. Thief is set in a dark fantasy world inspired by Victorian, gothic, and steampunk aesthetics. Using a modified version of the Unreal Engine 3 it makes for a challenging game to pitch against our range of components.

Company of Heroes 2 (RTS)

Company of Heroes 2 is a real-time strategy video game developed by Relic Entertainment. It is the sequel to the critically acclaimed 2006 game Company of Heroes.As with the original Company of Heroes, the game is set in World War II but with the focus on the Eastern Front, with players primarily controlling the side of the Soviet Red Army during various stages of the Eastern Front, from Operation Barbarossa to the Battle of Berlin. Company of Heroes 2 runs on Relic Entertainment’s proprietary Essence 3.0 game engine and is extremely demanding of both CPU and GPU.

Metro: Last Light (FPS)

Metro: Last Light is a single-player post-apocalyptic-themed first-person shooter video game with stealth and survival horror elements, developed by Ukrainian studio 4A Games and published by Deep Silver. The game is set in a post-apocalyptic Moscow, part of the universe of the novel Metro 2033 and its sequels, written by Russian author Dmitry Glukhovsky. The second of our FPS titles, it runs on the 4A gaming engine which again, has the capability to stress even the best GPU components.

GRID: 2 (Arcade Sim)

Grid 2 (stylised as GRID 2 and known as Race Driver Grid 2 in Japan) is a racing video game developed and published by Codemasters and is the sequel to 2008’s Race Driver: Grid. The game includes numerous real world locations and cities such as Paris. It also includes motor vehicles spanning four decades. In addition, it includes a new handling system that developer Codemasters has dubbed ‘TrueFeel’, which aims to hit a sweet spot between realism and accessibility. It uses the EGO 3.0 engine which while not overly challenging, is representative of games of this genre.

[fullimage=28065]

[fullimage=28067]

Overall Performance

Here we take a look at the overall performance of the graphics card. This figure is determined by finding the average FPS across all of the settings used in a particular benchmark(game) to give us an overall value.

Value For Money

We make a simple calculation for this set of results. For each game we add all of the resulting FPS totals together and then divide by the four sets to give us an overall performance figure. While the figures alone do not give you an accurate picture of how the card will perform in any given scenario, they do tell you, when averaged, which GPU is the more capable across all of the benchmarks and settings tested.

Conclusion

What makes a good graphics card is primarily its speed. The GTX 980 has that base covered, of that there can be no doubt. It took apart the overclocked AMD R9-290X with relative ease. So much so we had to do a double take when looking at the results. Obviously, we expected it to better the R9-290X but not by such a margin. While speed is of course very important, it isn’t everything you should consider when opting to upgrade your old graphics card or indeed if you are looking to invest in your first high performance gaming card.

When looking at flagship graphics cards it is also easy to forgive the price as we expect them to be expensive. While we don’t think NVIDIA have dropped the ball here (as it is cheaper than previous incarnations), it is still a large chunk of cash more than the competition and as shown in our value for money comparisons, it is soundly beaten on on that score. We should however remember that £429.99 is the manufacturers recommended price and this may well differ when it comes to retail prices so these results should be taken with a pinch of salt at this stage in the release.

It is also important to consider how efficient a GPU is both in terms of power and cooling, after all it is all good and well having a fast graphics card but not if you need a small wind farm to power it, or cool it for that matter. Maxwell has shown us that it’s architecture is extremely efficient in turning those watts into what we see on screen with FPS figures that are sure to astound. There is however a trade off here and that is the heat produced.

The GTX 980 is not a blisteringly hot card, nor is it noisy at reference speeds however, when you start ramping up the volts and increasing the clockspeed (with which it will readily oblige), the card does become a little toasty. Sure, we can adjust the fan speed to compensate but that increases the volume of the card, not that it will bother most gamers who use a headset but we thought it worth mentioning regardless. Thankfully, it is nowhere near the volumes levels expelled by the hairdryer cooler on the reference R9-290X. Consider that there will be many custom cooled varieties available and the heat may very well become a none issue but by our testing we found it hotter than both the GTX 780 and TITAN.

We suppose the GTX 980 being a ‘hot’ card (when overclocked) is actually very fitting because the blistering speed it produces is breath taking – more so when overclocked! While it is easy to get carried away with a new release, we feel the GTX 980 is deserving of our performance award simply because performance advantage it has over it’s competitors. So if, like many, you have been holding on to your old flagship card in anticipation of this release then you need wait no longer. Get one as soon as you can. It is as simple as that. On the other hand if you are a newcomer to the world of flagship graphics cards there is little reason to choose anything else.

To summarise:
If you need a graphics card to power the latest games at the highest resolutions. The GTX 980 will not disappoint.

Pros
+ Class leading Performance
+ Power efficient
+ Excellent features
+ Great overclocking potential

Cons
– Physical SLI bridge remains
– Can get hot when overclocked (reference design)