Testing Nvidia’s GeForce GTX 980 4GB Graphics Cards In SLI

Introduction

At eTeknix we were fortunate enough to have received a pair of GTX 980s for Nvidia’s launch day so we thought we’d see what they have to offer in SLI. When running a pair of the same GPUs in SLI it is ideal to get identical graphics cards for the most consistency of cooling, clock speeds, VRAM sizes and so on. In our case that wasn’t possible as we have one GTX 980 from Nvidia which is the reference model (check our review of that here) and we have one GTX 980 from Gigabyte which is their flagship G1 Gaming model (check our review of that here).

For our SLI testing both the GTX 980s were driving an ASUS ROG SWIFT PG278Q display In our review of both GTX 980s we know a single GTX 980 graphics card is more than enough for a smooth 60 frames per second (FPS) at 1440p in the vast majority of titles. However, if you want to effectively use the 144Hz refresh rate of the ASUS ROG SWIFT, you’re going to need to churn out as close to 144 FPS as possible. In this scenario SLI GTX 980s actually make sense because on their own one GTX 980 is not enough to drive such a high frame rate. Of course, it goes without saying you have to have deep pockets to afford a 144Hz 1440p monitor and a pair of GTX 980s – but even if you don’t have that kind of money seeing the numbers is still interesting.

We put both GTX 980 graphics cards onto our Core i7 4960X and X79-based test system ensuring adequate spacing and that both have access to sufficient PCIe bandwidth for SLI operation.

In order to keep both GPUs running effectively we made sure the better cooled Gigabyte GTX 980 sits in the traditional “hot position” of an SLI configuration, the Nvidia GTX 980 reference card is at the end to give it more room to breathe.

Gigabyte’s model comes heavily overclocked so we had to underclock it for the purposes of this testing to make it resemble as close to a reference GTX 980 as possible. By removing 101MHz off the core we were able to match the GPU clock speeds on both cards, although the Gigabyte card (the GPU-Z on the right) has a slightly faster boost speed.

We also didn’t want to run into any cooling issues. We know from our reviews that Gigabyte’s card doesn’t thermal throttle but Nvidia’s reference design does. As a result we set a custom fan profile within MSI Afterburner to make sure both GPUs stay under 80 degrees to avoid clock speed variations. We did this because it’s easier to make sure both GPUs do not thermal throttle than it is to make sure Gigabyte’s GTX 980 thermal throttles in the same way as the reference graphics card. The main thing to note is that we didn’t want significant clock speed mis-matches so this was the smartest option.

Running the SLI configuration in Unigine Heaven for a while gave us an idea of what stable clocks both graphics cards were using. The core and memory clocks were pretty much identical which is great news for getting some accurate test results. We will of course overclock both cards towards the end and see how that helps overall performance.

Test Systems and Procedures

Before we delve into any testing we would like to take this opportunity to overview our test system and thank those sponsors who kindly provided us with test equipment to make our work possible. We offer our thanks to:

Intel for supplying us with a Core i7 4960X processor, which we reviewed here.

ASUS for supplying us with a Rampage IV Extreme motherboard, which we reviewed here.

Be Quiet for supplying us with a Power Zone 1000W modular power supply.

Kingston for supplying us with a Hyper X 240GB SSD, which we reviewed here.

AMD for supplying us with a Dell U2711 Ultra Sharp monitor.

Corsair for supplying us with a Hydro Series H100i liquid CPU cooler, which we reviewed here, and the Corsair Vengeance Pro Series 1866MHz 16GB memory kit, which we reviewed here.

Lian Li for supplying us with a PC-T60A test bench.

Noctua for supplying us with NT-H1 thermal compound.

Test System

• Motherboard – ASUS Rampage IV Extreme X79 LGA 2011 Motherboard
• Processor – Intel Core i7 4960X at Stock 3.6GHz with Turbo Mode disabled
• RAM – 16GB (2 X 8GB) Corsair Vengeance Pro Series 1866MHz at 9-10-9-27
• CPU Cooler – Corsair H100i with Quiet Fan Profile with Noctua NT-H1 Thermal Compound
• Power Supply – Be Quiet Power Zone 1000W
• Main Storage Drive – Kingston HyperX 240GB SSD
• Chassis – Lian Li T60 Test Bench
• Displays – Dell U2711 Ultra Sharp
• Operating System – Windows 7 Ultimate 64 Bit

Driver Details

Performance Summary

To formulate our performance metric we use a weighted average method and each benchmark test gets a weight of 1 out of a total calculation with 9 weights (meaning all resolutions and variations of the same test are only ever one weight). Compute tests, power consumption, temperatures and acoustics are not included in the weighted calculation. The 9 weights are 1 for each test we so that means 6 games (listed below) and 3 benchmarks (also listed below). We calculate the relative score for each card in each test by dividing the cumulative average frame rate or score of each card by the average cumulative average frame rate/score from all the graphics cards in group. We then take those 9 relative scores and average those for each card, before dividing those averages by the average of the card being reviewed in the test so that the value of the card being reviewed is 1 and every other card’s performance metric is in relation to it. The end result are numbers that vary around 1 with 0.5 being half the performance of the tested card and 2 being double the performance of the tested card. We then convert those numbers into percentages to make them more readable so 1 becomes 100% and 0.5 becomes 50%.

Hardware Used

• “Killawatt” style electricity usage meter wall plug
• Benetech GM1351 decibel meter

Games Used

• Batman Arkham Origins
• Battlefield 4
• Bioshock Infinite
• Hitman Absolution
• Metro Last Light
• Tomb Raider

Benchmarks Used

• 3DMark
• 3DMark 11
• Unigine Valley

Other Software Used

• Unigine Heaven
• Furmark
• CPU-ID HWMonitor
• TechPowerUp GPU-Z
• Computemark
• Luxmark

3DMark 11

“3DMark 11 is the latest version of the world’s most popular benchmark. Designed to measure your PC’s gaming performance 3DMark 11 makes extensive use of all the new features in DirectX 11 including tessellation, compute shaders and multi-threading. Trusted by gamers worldwide to give accurate and unbiased results, 3DMark 11 is the best way to consistently and reliably test DirectX 11 under game-like loads.” From 3DMark.com

3DMark 11 scales very well as most synthetic benchmarks tend to do and we nearly end up with perfect scaling. In the Extreme test the graphics score goes from 5300 to 10500, perfect scaling would have been 10600.

3DMark

“The new 3DMark includes everything you need to benchmark your hardware. With three all new tests you can bench everything from smartphones and tablets, to notebooks and home PCs, to the latest high-end, multi-GPU gaming desktops. And it’s not just for Windows. With 3DMark you can compare your scores with Android and iOS devices too. It’s the most powerful and flexible 3DMark we’ve ever created.” From Futuremark.com

3DMark (2013) shows similar scaling; most Futuremark 3D benchmarks handle SLI well. Let’s see how that translates in real-world games.

Batman: Arkham Origins

“Batman: Arkham Origins is a 2013 action-adventure video game developed by Warner Bros. Games Montréal and released by Warner Bros. Interactive Entertainment for Microsoft Windows and the PlayStation 3, Wii U and Xbox 360 video game consoles. Based on the DC Comics superhero Batman, it follows the 2011 video game Batman: Arkham City and is the third main installment in the Batman: Arkham series. The game, played from a third-person perspective, focuses on Batman’s combat and stealth abilities, detective skills, and gadgets for combat and exploration. Arkham Origins is the first game in the series with multiplayer capability.” From Wikipedia.

Batman Arkham Origins scales well, around 65%, which seems the going rate for SLI configurations.

Battlefield 4

“Battlefield 4 is coming to PC, powered by the advanced technology of DICE’s proprietary Frostbite 3 engine. Blur the line between game and glory in Battlefield 4. With dynamic destructable environments, vehicular combat, and the chaos of all-out-war with 64 players, Battlefield 4 on PC will be an unmatched interactive experience. In addition to its hallmark multiplayer, Battlefield 4 features an intense, dramatic character-driven campaign that starts with the evacuation of American VIPs from Shanghai and follows your squad’s struggle to find its way home. There is no comparison. Immerse yourself in the glorious chaos of all-out war, found only in Battlefield.” From Battlefield.com

Battlefield 4 revealed nearly perfect scaling especially at the higher 1440p resolution.

Bioshock Infinite

“BioShock Infinite is a first-person shooter like you’ve never seen. Just ask the judges from E3 2011, where the Irrational Games title won over 85 editorial awards, including the Game Critics Awards’ Best of Show. Set in 1912, players assume the role of former Pinkerton agent Booker DeWitt, sent to the flying city of Columbia on a rescue mission. His target? Elizabeth, imprisoned since childhood. During their daring escape, Booker and Elizabeth form a powerful bond — one that lets Booker augment his own abilities with her world-altering control over the environment. Together, they fight from high-speed Sky-Lines, in the streets and houses of Columbia, on giant zeppelins, and in the clouds, all while learning to harness an expanding arsenal of weapons and abilities, and immersing players in a story that is not only steeped in profound thrills and surprises, but also invests its characters with what Game Informer called “An amazing experience from beginning to end.” From Bioshockinfinite.com

Bioshock is one of the games that scales the worst, but still managed an additional 50%. Interestingly at 1440p we managed bang on 144 frames per second, pretty sweet for a 144Hz monitor I’d say.

Hitman Absolution

“Hitman: Absolution is the fifth title in the Hitman series by producer, IO Interactive, published by Square Enix and co-published by Eidos. The game was released for thePlayStation 3, Xbox 360 and Microsoft Windows on November 20, 2012. Hitman: Absolution runs on IO Interactive’s proprietary Glacier 2 game engine, combining classic Hitman game mechanics with new mechanics.” From hitman.wikia.com

Hitman Absolution scaled well….almost too well. At 1440p we had close to 100% scaling but even then the frame-rate is a considerable way off the required 144.

Metro Last Light

“Metro: Last Light (formerly Metro 2034) is a first-person shooter and horror video game developed by Ukrainian studio 4A Games and published by Deep Silver for Microsoft Windows, PlayStation 3 and Xbox 360. It was released in May 2013. The game is set in a post-apocalyptic world and features a mixture of action-oriented and stealth gameplay. The game exists in the universe of the novel, Metro 2033, and its sequels, written by Russian author, Dmitry Glukhovsky, but does not follow any direct storylines from the books. Metro:Last Light takes place one year after the events of Metro 2033, proceeding from the canonical ending from the novel, ending where Artyom chose to call down the missile strike on the Dark Ones.Metro: Last Light features technology which boasts of lighting effects and improved physics claimed to set a new graphical benchmark on the PC and consoles.” From Wikipedia.org

Metro Last Light is nothing short of a GPU destroyer. Even with SLI GTX 980s that showed 90% scaling we still couldn’t push north of 60 FPS. In theory you’d need 4-Way GTX 980 SLI that’s overclocked just to get near 144 FPS.

Tomb Raider

“On 5 March 2013, Square Enix released Tomb Raider, billed as a reboot of the franchise. In Tomb Raider, the player is confronted with a much younger Lara Croft who is shipwrecked and finds herself stranded on a mysterious island rife with danger, both natural and human. In contrast to the earlier games Croft is portrayed as vulnerable, acting out of necessity, desperation and sheer survival rather than for a greater cause or personal gain.” From Wikipedia.org

Tomb Raider is ROG SWIFT ready; 177 FPS with around 80% scaling at 1440p.

Unigine Valley

“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.” From Unigine.com.

Unigine is a GeForce-friendly test so I am not surprised to see scaling is nearly at 100% in this test. It’s amazing how much difference adding another card can make. It goes from a slightly juddery 40 FPS to a very smooth 74 FPS.

Performance Summary

With our reviews stuffed full of a large number of benchmarks and games at different resolutions and using different settings it can be difficult to get a feel for the overall performance of a graphics card we are testing. As a result we’ve created a performance summary metric. The metric is simple, but my explanation isn’t, if you’re interested in that explanation then see the test systems and procedures page.

Rounding up all our benchmark results and averaging them out we find the Nvidia GTX 980s scale by about 70% – a very respectable figure.

Compute Performance

ComputeMark (DX11 Compute)

ComputeMark is first 100% DirectX 11 Compute Shader benchmark and ultimate GPU burner. From ComputeMark.com.

Computemark doesn’t support a second GPU so we find the result is within margin of error of a single card. The result is a little higher because the Gigabyte GTX 980 is GPU1 and as mentioned on the first page it has a slightly higher boost clock than the reference version.

Luxmark (OpenCL Compute)

LuxMark is a OpenCL benchmark tool. The idea for the program was conceived in 2009 by Jean-Francois ‘Jromang’ Romang. The idea was quite simple, wrap SLG inside an easy to use graphical user interface and use it as a benchmark for OpenCL. From LuxRender.net.

Luxmark scales with multiple GPUs and we saw perfect scaling: a great result for OpenCL fanatics.

Overclocking and Overclocked Performance

We already know how well both our GTX 980s overclocked from their individual reviews and you can see the links to those reviews on the first page of this article. As a result we set both graphics cards to the maximum clock of the slowest GPU overclock which just so happened to be 1347MHz core and 2015MHz memory: still very impressive overclocks. In testing this gave about an additional 10-15% of extra performance which is a strong result. Generally overclocking in SLI configurations can be tough but in this instance it was nothing short of a breeze.

3DMark

Battlefield 4

Noise Levels

The noise levels produced by a graphics card is an increasingly important consideration for PC users and Gamers these days. While fan noise is unlikely to ruin the gaming experience, no one likes a noisy graphics card and no one will argue with the fact that quieter is better. Many users are willing to sacrifice temperatures to gain a silence advantage, but with better cooling solutions being developed it is increasingly common to be able to get both better cooling and better acoustic performance than a reference solution on most custom cooled graphics cards from Nvidia and AMD partners. That said both AMD and Nvidia have stepped up their game too with regards to acoustics on their reference coolers. Ultimately, acoustics will always be a big deciding factor when there is often very little differentiation between graphics cards using the same GPUs. To acquire the noise levels of graphics cards in our benchmarks we run three different load scenarios for 5 minutes and take an acoustic reading using our decibel meter at the end of those load scenarios. The measurement is taken approximately 30cm horizontally away from the graphics card and the CPU fan is disabled and other background noise minimised to reduce interference. The three load scenarios are desktop idle, Furmark load and Unigine Heaven load. The reading in Unigine Heaven is always taken in the first scene of the sequence after one loop is completed.

The noise of our configuration is a direct reflection of the custom fan profile we created. Even so, it still runs quieter than a single R9 290 or R9 290X which is mildly amusing.

Power Consumption

With electricity becoming increasingly expensive across most parts of the world the need for computer components to become power efficient has never been more relevant. Graphics cards are often the most power hungry components inside a desktop system so having an efficient graphics card is very important to keeping power bills under control. Power is often correlated to heat and so lower power consumption means a graphics card is likely to run slightly cooler and put out less heat into your system meaning your other components will run cooler with improved longevity. AMD and Nvidia have both made power consumption an integral part of the way graphics cards dynamically overclock so the need for graphics card vendors to use efficient VRM and PCB designs is becoming important to maximise performance. We take power readings after 5 minutes of three different load scenarios: desktop idle, Furmark load and Unigine Heaven load.

Power consumption is impressive for an SLI configuration. Both GTX 980s consume only marginally more than a single GTX 780 Ti GHz Edition graphics card yet deliver significantly more performance in a real-world style application like Unigine Heaven. Under Furmark power consumption soars mainly due to the high consumption of the GTX 980 G1 Gaming.

Temperatures

The cooling solution which graphics card vendors choose to implement is one of the main differences that consumers have to contend with when choosing a graphics cards. Apart from their acoustic properties, the thermal properties of graphics card coolers are extremely important. Lower temperatures are always better and with AMD and Nvidia opting to use dynamic overclocking algorithms that take temperature into account it is important that graphics card vendors use high performance cooling solutions in order to maximise performance. The era of graphics cards reaching dangerous temperatures are now in the past but the importance of lower temperatures still remains. Lower temperatures mean better stability, longer component longevity and lower fan speeds .We take temperature readings after 5 minutes of three different load scenarios: desktop idle, Furmark load and Unigine Heaven load. We always record actual temperatures and make a note of the ambient. In the case where more than 1 GPU is used an average is created.

Our custom fan profile ensured the temperature stayed well below 80 degrees under all load scenarios. The temperatures you achieve in an SLI configuration will obviously vary depending on clock speeds, the card’s chosen, the spacing used, what kind of case you have and so on.

Final Thoughts

Multi-GPU systems seemed to have gained a negative reputation in the past for bugs, glitches and a lack of performance scaling. I myself have never seen anything to support that argument and today looking at these GTX 980s in SLI it is clear to see that scaling is very strong and the argument for choosing SLI is valid. We didn’t encounter any significant bugs or issues in testing and the 70% scaling figure we can quote from our results is very reasonable. Furthermore, with overclocking, an additional 10% can be had on top of that and that wasn’t very hard to achieve either.

GTX 980s in SLI offer an incredible amount of performance for a surprisingly low amount of power consumption and if you absolutely need to drive framerates to the maximum you can’t go too wrong with this type of setup. The kind of use-case scenarios for GTX 980s in SLI that we would suggest are numerous. You could use GTX 980 SLI for:

• Generating enough frames to match 144Hz displays at 1080p or 1440p (especially G-Sync displays)
• Powering 4K gaming to the magic 60 FPS level on the current wave of 60Hz 4K monitors
• Driving the Nvidia Surround gaming experience across multiple 1080 or 1440p displays at playable frame-rates

Nvidia GTX 980 SLI is downright expensive, $1100 in total just for the GPUs, but considering the kind of display real-estate you’d need to make use of such GPU power this is realistic for the target audience. SLI will always be for a niche market but you can be rest assured if you are part of that niche market it does deliver what you pay for. My only word of warning would be that 2-Way configurations are known to scale very well but if you were to start scaling up to 3 and 4 way configurations the experience might not be as seamless.

Thank you to Nvidia and Gigabyte for providing these review samples.