Earlier this year, NVIDIA offered us a peek at its next architecture, Maxwell with the launch of the 750Ti. Now a mid-range card seems a rather odd launch platform for a new architecture, but this was done to drive home the power efficiency of the new architecture. And as is custom, NVIDIA is now ready to take the wraps off its top tier products based on the second generation of the Maxwell Architecture, the GTX 980 and its little sibling the GTX 970.
Maxwell, GM204 and feature briefing by Sahil Arora.
[section label=”Maxwell” anchor=”Maxwell”]What is Maxwell?
The GTX 980 and GTX 970 are not the first Maxwell architecture 28nm based graphics cards. Infact, it is the excellent GTX 750Ti and GTX 750, that is the first mainstream desktop Maxwell based GPU. The 750Ti was launched in February 2014, and it marked the end of any new GTX 7xx sseries cards remarkably.
Maxwell is NVIDIA’s new architecture that is designed to work on a considerably lower power draw than the Kepler architecture, which the entire GTX 7xx series uses, except the GTX 750Ti. Maxwell’s low power footprint did wonders to the 750 Ti, which is a “proper” mid-range beast that has a TDP of 60W and works without a discrete power connector. At iLL, we found the GTX 750Ti to be so good, especially for its asking price, that we gave it our “iLL Penny Pincher Award.” We talk about the Maxwell architecture much deeper in our 750 Ti review.
Back then, we knew little about Maxwell, we only knew it in its stripped down form, the GM107, that the 750 Ti is based on. We have been anticipating what a pimped up Maxwell’s would dance like. Well, we do now. And it was hard to contain our excitement till the embargo lifted. Ahem.
The GTX 980 is based on the GM204 chipset. which NVIDIA calls the “full realization of our 10th generation GPU architecture, Maxwell.”
GM204 Hardware Architecture In-Depth
The GM204 consists of 4 Graphics Processing Clusters (GPCs) and memory controllers. The 780Ti’s GK110 has more GPCs than the GTX 980, at five. A GPC is where all the work in a GPU happens, and single GPC in the GTX 980 consists of a dedicated raster engine and four SMMs, which makes up the total to a bulky 16 streaming multiprocessors. Each SMM has 128 CUDA cores, a PolyMorph Engine, and eight texture units. With 16 SMMs, the GeForce GTX 980 ships with a total of 2048 CUDA cores and 128 texture units.
A dedicated PolyMorph Engine on each SMM gives the GTX 980 the ability to have twice the number of geometry units as its direct predecessor.
With the GM204, NVIDIA has achieved in GTX 980 double the SMs (Streaming Multiprocessors) than the GTX 680, which is based on the Kepler GM104 chipset, and this has been done without doubling the die size. The GTX 970 will be shipped with 13SMs compared to the 980’s sixteen.
The texture units in the Maxwell has been bumped up, and lies at 128 (8 texture units per SMM), the same as Kepler. Clock speeds in the GTX 980 have seen a slight increase from the previous series. Texture fill rate has been improved by 12%. Due to the higher clocks, pixel fill-rate in the GTX 980 is more than double that of GTX 680.
Memory clocks are 15% higher than the GTX680. Implementation of how memory works in Maxwell has been revamped too. The GM204’s cache is larger and more efficient than Kepler’s design, reducing the number of memory requests that have to be made to DRAM. Memory compression is implemented in a more efficient way, “amplifying” the DRAM bandwidth in the system.
Features in the GTX 980
Such high specs look fancy, but how will they help you in playing games in your 24″ 1920 x 1080 pixel monitor? A GPU such as a GTX 980 seems like an overkill. That’s what we were always of the opinion of, but its not the case anymore.
Dynamic Super Resolution (DSR)
The most simple explanation of DSR can only be done by citing an example. Say that 24″ 1920 x 1080 pixel monitor belongs to you, and you own a GTX 980. Sure, this card can run any game at full visual settings at the highest resolution you monitor can support. And it’ll do it without breaking a sweat. But a card of this power has an appetite to be utilized to its full potential. That’s where DSR comes into play. DSR processes that game in an upscaled resolution like 4K, and then downscales it back to your monitor’s resolution. This makes the image quality considerable better working on similar principles as Antialiasing.
Many modders use downsampling as a way to achieve a sharper image quality, but doing so requires a lot of fiddling with the game’s system files, so that is eliminated with DSR. It’s nothing new, it’s just added convenience. But a great convenience nevertheless. DSR is integrated in GeForce Experience and can also be “Auto-enabled” for your PC games. This should make bearing those shoddy console ports that much easier.
MFAA – Multi-frame Sampled AA
MFAA is a re-hashed Multisample anti-aliasing (MSAA). MSAA is the most common form of AA, and although it does raise the visual quality, the performance hit is massive. MSAA in in games is always seen as a luxury feature, one which you’ll enable to see the best eye candy but you know you’re dropping a lot of frames per second doing it.
MFAA has been introduced with the logic of giving the same quality as MSAA, but not being a resource hogger in the process. Nvidia claims the visual difference between 2xMSAA and 4xMFAA is not even 99.9%, where the performance gain is almost 30%.
MFAA uses the features of the GTX 980/970 chipset to achieve this. The GM204’s chipset allows programmable AA sample positions, which the MFAA implies. Currently there are no games that support MFAA, but this seems to be a nice feature and will definitely benefit 4K monitors, who almost always have to miss out on MSAA due to unplayable FPS.
[section label=”Specification” anchor=”Specification”]Specification
The card has a bumped up ROP count of 64 and comes with 4GB of Samsung GDDR5 VRAM. Core clock is set to 1126MHz with boost clock at 1216MHz. The VRAM is clocked at an effective 7000MHz. The VRAM should be able to overclock decently and given our experience with the 750Ti, we wouldn’t be surprised if this card had decent overclocking headroom to let the manufacturers have some fun with.
The rear panel of the card sports a DVI-D port, a HDMI port and 3 DisplayPort slots. Of these only 4 can be used at a time. There are two SLI connectors as well.
The launch price for the GTX 980 is ₹46,000 and for the GTX 970 is ₹28,000.
[section label=”Build Quality” anchor=”bq”]Build Quality and Packaging
NVIDIA’s stock cooler design builds on their older design for the 7xx series and the Titan. The fins are now coloured black and the top of the PCB is covered with a backplate. This backplate has a removable piece for allowing airflow into the fan of another card in the cramped space of a SLi setup. Our review sample came in a lovely box as seen in the pictures but the manufacturers should provide basic necessities like a SLi bridge, stickers, Driver DVD, Manual and adapters. The build quality of the card is good though the finish leaves a little to be desired and gives the card a slightly rough look. There is the familiar Green GEFORCE GTX LED to leave onlookers in no doubt as to what beast powers your Gaming Rig.
[section label=”Testing Methodology” anchor=”TM”]Testing Methodology
We tested the GTX 980 not only for the average Frames per Second(FPS) but also for the 99th Percentile Frame time which tells us about the performance of the GPU within the second. Within the Second testing is useful to understand micro-stutter which can render a game unplayable despite FPS being high. Fraps 3.5.99 allowed us to calculate both.
Since the card is a Top End GPU, we decided to use highest possible settings in our benchmark games and compare with other Top End GPUs. We tried to disable CPU dependent settings or minimise their impact where possible. VSync and frame buffering were disabled for testing. All tests were run at 1920×1080 on a single monitor configuration.
Test System:
CPU: AMD Phenom II X4 965 BE C3 @ 3.8GHz
Motherboard: ASRock 970 Extreme 3
RAM: 2x4GB G.Skill Sniper CL9 1600MHz, 2x2GB G.Skill Ripjaws CL9 1600MHz
PSU: Corsair TX650 650W (for GTX 980, Matrix Platinum 780 Ti, Mars GTX 760), Seasonic S12II Bronze 620W (for others)
HDD: 2xSeagate Barracuda 7200.12 1TB, 1xWesten Digital Red 3TB
OS: Windows 8.1 x64
NVIDIA Driver: 334.89 for benchmarks
NVIDIA Driver: 344.07 (Beta Driver) for GTX 980
AMD Driver: 14.3 Beta 1 (Beta Driver)
While this system may not look like a purpose built test rig, we decided to use a normal usage PC so as to better reflect real world scores of the card. The games were also tested with a few applications like Antivirus, Browser, VOIP tool and others running in the background to get a realistic usage scenario. All HDDs were thoroughly defragmented prior to usage. Due to lack of equipment we were unable to conduct acoustic and power testing.
[section label=”3DMark” anchor=”3DM”]3DMark Firestrike
3DMark is an artificial Benchmarking tool whose Firestrike Test is very thorough on DirectX 11cards powering High-end PCs. The full test run for Firestrike includes 2 GPU only tests, a CPU dependent Physics Test and a Combined Graphics and Physics Test. The Tool is also useful for stress testing a GPU when run on loop.
Given that we are looking for the Performance of the Card itself, one should look at the Graphics score and the FPS for Graphics tests 1 and 2. The Physics and Combined tests are CPU dependent which is the limiting factor of our test rig.
[section label=”Game Benchmarks” anchor=”GB”]Game Benchmarks
Batman Arkham Origins
Batman Arkham Origins is a game that’s been supported by NVIDIA and utilises their PhysX technology to handle physics. We used highest possible settings. PhysX was also set to Enhanced. We ran the inbuilt benchmark tool with a FRAPS timed run of 120 seconds. Since this benchmark was picked mainly to showcase the NVIDIA exclusive PhysX performance, one should ignore the AMD card results which are generated sans PhysX.
This game is an Nvidia Card’s Home territory and the results are visible. Disregard the AMD card since its score is sans PhysX. THE GTX 980 almost matches the Matrix for average FPS and minimum FPS, but has a lower 99th percentile frametime, which shows that it is the better card.
The spikes in the graph with corresponding dips in the FPS come from transitions between benchmark areas and should be disregarded. Gameplay was crisp for the duration of the run.
Bioshock infinite
The Unreal Engine 3 powered Bioshock Infinite really pushed the boundaries of visual effects achievable with the ageing engine. It comes with a built in benchmark mode which runs for a lower time period than the standard 120s we used in other tests.
Another close run with the 780Ti on average FPS and this time it’s superior on both the minimum FPS and the 99th percentile frametime.
The graph shows local spikes which aren’t going to trouble your gameplay experience as microstutter. The FPS however changes wildly and is noticeable so it’s recommended to play with vsync on.
Company of Heroes 2
Relic’s Company of Heroes 2 is a tough nut to crack for quite a few GPUs, though it’s dependent on CPUs to a great degree as well. We used highest possible Settings with Physics turned off and Low AA for the short 45s benchmark run the game offers.
Given that this game made even the 780 Ti struggle to provide playable framerates, it is astonishing to see how well the GTX 980 performs in this game. It takes the crown in all three parameters with minimum fps being just below the playable 30 fps threshold.
The frametimes are quite tight with the stray spikes. The FPS also remains mostly playable dipping below the limit near the end. It was quite a smooth run.
Crysis 3
CryEngine 3 was built to push the PC Hardware to its limits with Crysis 3 and deliver Maximum Eye Candy. We used the highest possible settings for the purpose of our test. The test run consisted of a section of the first mission for 120s as the game surprisingly offers no benchmark tool.
This game is so graphics dependent that a change in CPU saw no change in the results. Thus it is indicative of the GTX 980’s true power that it gets the highest average fps, but loses out in the minimum fps. It’s lowest in the 99th percentile frametime however.
There are some noticeable spikes in the graph but they aren’t spaced closely enough to be observable as stuttering. The rest of the graph is quite concentrated. The FPS seems steady for the most part though it has wild swings towards the end.
Far Cry 3
Far Cry 3 is a visually impressive game that lacks a benchmark mode. We used highest possible settings and the test run consisted of a 120s sample from the game’s single player campaign where we spent some time goofing off in the open world.
Finally a game that made the GTX 980 stumble, but not by much. The 780 Ti muscles through this game on all the criteria.
The frametimes are evenly distributed with a few spikes that were noticeable as stutter. The FPS curve shows a lot of dips and valleys which makes vsync a necessity in this game.
GRID 2
GRID 2 from Codemasters is one of the few games in our test suite that has no allegiance to either NVIDIA or AMD, which makes it a neutral candidate to better judge the GPU’s performance. We used highest possible Settings to run the benchmark tool offered by the game.
The GTX980 owns the competition in this game with significantly better scores across our criteria. Given that the game is vendor neutral, this is a good showcase of the card’s raw power.
The frametime graph is tight and the fps graph is relatively smoother. Gameplay was fluid though there were some instances of screen tearing.
Sleeping Dogs
While the game is ageing, Sleeping Dogs is still a stunner when it comes to running benchmarks. We pushed the settings to highest possible and ran the game’s built in benchmark run.
The GTX 980 gets very close to the 780 Ti but only manages to bet it on the 99th percentile frametime.
While the frametimes are more spread out in the outdoor scenes, they are very tight in the final indoor sequence. The FPS varies accordingly too remaining near constant in the end. The spikes are corresponding to loading screens between the areas showcased in the benchmark and can be ignored.
Tomb Raider
Tomb Raider introduced us to a new Lara with fabled TressFX hair that behaves a lot more realistically than the pre-rendered mop we were used to. However the card does not seem to cope well with this technology and we had to drop it in our test run. We used highest possible settings with TressFX on. The Test run was shorter since the benchmark tool offered by the game runs for less than out 120s target time.
Another game which caused the GTX 980 to stumble, it seems NVIDIA’s Hairworks answer to TressFx couldn’t come sooner.
The consistent spiking pattern is caused more by CPU bottlenecking rather than GPU problems.
Total War: Rome II
Total War: Rome II is another CPU heavy game that offers significant visual goodness. It offers a benchmark tool that focuses more on GPU power though and that is what we used for our test run of 120s. We set the game to Extreme and Unit Size to Small so as to reduce the impact of the CPU. Unlimited Video Memory was off so the game could scale down visual settings if it reached a bottleneck.
While the GTX 980 has the best average and 99th percentile frametime score, it dips a little lower than the 780 Ti in the minimum FPS, though it fares much better than the other NVIDIA cards.
The frametimes are distributed all over the place and are partly higher due to the CPU bottleneck. The spikes are too isolated to have any effect on the crisp gameplay run.
[section label=”Acoustics” anchor=”acoustics”]Acoustics
Our acoustics testing consisted of trying to determine how noticeable the noise output from the card was, when kept in a case at 1m distance with the side panel closed, as it might be in a real world scenario. Noise is a very relative characteristic that depends not only one the person hearing, but also the background noise of their surroundings. During the course of our testing, we found that the card barely went over 50% fan speed even at full load, which was silent to our ears. However, setting the fans at 100% rpm makes a noticeable din that sounds something like a toy R/C Car. NVIDIA have come up with an ear pleasing cooler design since the 7xx series that is modified with black fins for this card.
[section label=”Temperatures” anchor=”Temps”]Temperatures
At idle the card hovered in the 42°-44°C range with an ambient temperature of 28°C which is quite warm. The card easily went to 84°C under load. While overclocked it would plateau out at 85°C as the fan kicked up its RPM. This shows that there is room for manufacturers to improve on the stock cooler design for a cooler card.
[section label=”Power” anchor=”Pwr”]Power
The card comes in a dual 6 pin power connector configuration which rates it at 165W TDP though it might exceed that when overclocked. The card requires a good PSU for overclocking. While the Maxwell chips in the 750 TI delighted us with their phenomenal power per watt performance, these new generation Maxwells managed to outperform them. NVIDIA claims a Performance per watt ratio twice that of its older GTX 680 and 1.5 times that of the GTX 780. On idle it consumes 7.6% of TDP and this can go up to 127.7% of TDP when overclocked.
[section label=”Overclocking” anchor=”oc”]Overclocking
Using MSI Afterburner, we were able to push the core clock to 1357MHz with boost clock at 1446 MHz and the Samsung VRAM chips to 1953 MHz (7812 MHz effective) which is a total overclock of about 20.5% over stock for the core clock and about 11.5% for the VRAM. These are impressive figures for overclocking without any alterations to the card or without increasing the voltage. The Voltage can be increased to improve stability of the overclock. All our tests however, were conducted in the stock configuration.
We ran 3DMark to check improvements in the overclocked performance and the graphics score increased from 13147 to 15100.
We also ran the benchmark offered by the game Thief to get a real world idea of the performance gains. We used the Very High settings for our runs which concluded in less time than our standard 120 second runs. One must note that Thief is slightly unreliable as a benchmark due to inconsistency in results across various runs. That said, we got an improvement of about 1-2 fps on average with a very slightly higher 99th percentile frametime.
[section label=”Feature Set” anchor=”FS”]Feature Set
With NVIDIA’s GeForce Experience, the card can support features like Shadowplay and game streaming. The card’s LED strip can also be controlled to display some cool effects. The new card also lets one use a technique called Dynamic Super Resolution that can be used to enhance the visual quality of shoddy console ports.
[section label=”Conclusion” anchor=”End”]Conclusion
The GTX 980 is a brilliant card. Its performance improves on not just the GK 104 parts it’s derived from but also matches or betters the GK110 parts like the 780Ti that we rated so highly. And this is done with much less power than those cards.While the card comes with a host of new technology, the showcase games for it are yet to be released. When they do come, they are sure to pack quite some eye candy and usher in a new level of graphical effects on the most advanced gaming platform, the PC.
The stock cooler is a passable design and has quite some scope for improvement. The overclocking performance of the card is stellar and with its sweet pricing, it gives it even more bang for the buck. The power efficiency of the GPU adds another dimension when judging its value for money, since we also get more value in the operating costs. Considering all the findings of our tests, we award this card the iLL Gaming Gold and Editor’s Pick awards.
Priced at ₹46,000, the GTX 980 is ₹12,000 lower than GTX 780 Ti’s launch price. This makes the card a worthy proposition for people wanting top-notch future-proof gaming performance. Owners of the GTX 680 will find good enough reason to upgrade to the GTX 980. For owners of the GTX 780 and Titan, performance boost isn’t that great, but the host of features introduced in the series are attractive, yet they’re not critical to a gaming experience, so they can give it a pass.
