NVIDIA GTX 980 review: ‘Max’well performance with impressively low power consumption


Nvidia has launched its latest series of cards which will feature its Maxwell architecture – the GTX 900 series. And the first cards from this series that were announced last week are the GTX 980 and the GTX 970. What happened to the GTX 800 series, you ask? Well, according to Nvidia, they have used the 800 series naming convention in their mobile GPUs, so they decided to go ahead with the GTX 900 series. While the Maxwell architecture was introduced earlier this year, when Nvidia launched theGTX 750Ti, it was more of an entry level card. The GTX 980 on the other hand is a single GPU flagship of the 900 series. So let us see how it fares.
Build and Design
Nvidia seems to be following the ‘if it ain’t broke, don’t fix it’, philosophy when it comes to their flagship card designs. The GTX 980, like the GTX 780TiGTX TITAN and to an extent the GTX 690 before it, has stuck to the similar gray coloured design with slight variations. On a first glance, the designs look identical – the 65mm radial fan is the same, the flourescent green GeForce logo atop the card is the same, the gray/black sheath is the same. The card measures 10.5-inches in length. Under the gray metallic shroud, you have a black-coloured heat-sink though. Under the heat-sink, you have three heat-pipes that transfer heat from the GPU to the heat-sink. This is a departure from the vapour chambers used in the GTX 780Ti and GTX Titan.
The Nvidia GTX 980 comes with a back-plate, a portion of which can be detached when using a GTX 980 multi-GPU setup for better air-flow
The GTX 980 adds in a backplate to the card to protect the PCB on the rear side. In addition to this, Nvidia also allows you to unscrew the top of the back plate just behind the PCIe power ports. According to Nvidia, this helps with the air-flow when you have GTX 980 SLI setup, when there is little space between the two adjacent GTX 980s.
On the display front, you have 3x DisplayPort 1.2, 1x HDMI 2.0 port and 1x DVI port. As is evident, Nvidia is pushing people to adopt the 4K displays for their systems having the GTX 900 series cards. The HDMI 2.0 port also supports 4K at 60Hz. It requires just two 6-pin PCIe power connectors.
Features
As we mentioned earlier, the Maxwell architecture was introduced with the lower end GTX 750Ti which was impressive considering it was a GTX level card with a maximum TDP of 65W and which was powered just by the PCIe port. The GTX 750Ti was based on the GM107 GPU. With the Kepler architecture we had 192 CUDA cores per stream multiprocessor or SMX, which had been modified with the Maxwell architecture. In the GM107 for instance, within a single Maxwell Streaming Multiprocessor or SMM, you had four processing blocks each with 32 CUDA cores and with its own instruction buffer and scheduler which gave a total of 128 CUDA cores per SMM. Thanks to this partitioning of CUDA cores into four separate processing blocks, NVIDIA managed to simplify the scheduling logic and performance per CUDA core. Five such SMM’s made up the graphics processing cluster (GPC).
GM204 block diagram showing four graphics processing clusters (GPC) each with four streaming multiprocessor units (SMM)
With the GTX 980, Nvidia has used the GM204 GPU which naturally is a scaled up variant showcasing the Maxwell architecture. While the GTX 750Ti had just one graphics processing cluster with five SMMs, the GTX 980 will have four GPCs – each of which will have four SMMs with each SMM having 128 CUDA cores. This takes the total CUDA cores count to 128(CUDA cores/SMM) x 4(SMM) x 4(GPC) = 2048 CUDA cores. Also consider each SMM has eight texture units, the total number of texture units on the GM204 are 8 x 4 x 4=128 texture units.
The GTX 980 has a base clock of 1126 MHz and boost clock speed of 1216 MHz. It comes with 4GB of GDDR5 RAM clocked at 7 GHz thereby giving 224GB/s of theoretical bandwidth. Around four 64-bit memory controllers give a 256-bit memory bus width, which is lower than the 384-bit one seen on the GTX 780Ti. There are 5.2 billion transistors on board the GM204, as opposed to 7.1 billion seen on the GK110 (Kepler) GPU. The maximum TDP is set at 165 Watts! Yes, this is even lower than the GTX 780Ti whose maximum TDP was around 250 Watts. All this on the same 28nm manufacturing process, which looks impressive on paper.
Nvidia has also added on some new features.
Dynamic Super Resolution (DSR): With the DSR activated, the scenes will be rendered at a higher resolution and then will be scaled down to the native resolution of your display. For instance, if you have a 1080p display, the the DSR activation will have the GPU render the image to a 4K resolution, and this will then be scaled down to 1080p to correspond to the display. In theory it sounds similar to downsampling, but Nvidia has employed a special filter which helps reducing aliasing artifacts which are common with downsampling. The benefit of using DSR is to improve the image quality of less demanding games which may look stretched on a high-res display.
Multi-frame Anti-aliasing (MFAA): This is a feature exclusive to the GTX 980/970 cards thanks to a hardware feature: programmable AA sample positions. It has been introduced to address the issue faced by graphically demanding games, which get a hit in frame rates when you increase the multi-sampling anti-aliasing (MSAA) to highest levels. The final result can deliver image quality approaching that of 8xAA at roughly the cost of 4xAA, or 4xAA quality at roughly the cost of 2xAA. This lets you get a better image quality without being a performance hit on the frames per second. This feature will be seen in future games.
Voxel Global Illumination (VXGI): Lighting the scenes in a game have been pre-computed as dynamic rendered global illumination was resource intensive. Nvidia GTX 980 / 970 allows a real time global illumination engine which assures realistic reflections.
The GM204 also gets an improved video encoder with support for H.265 codec which allows you to do encoding tasks upto 2.5 times faster than the older Kepler cards. ShadowPlay now allows you to record 4K videos at 60FPS thanks to GM204.
Test Setup
Processor: Intel Core i7-4770K
Motherboard: ASRock Z87M Extreme4
RAM: 2 x 4GB GSkill RipjawsX
Drive: Intel SSD, 80GB (boot drive)
Additional Drive: Western Digital Velociraptor, 150GB (secondary)
PSU: Cooler Master 800W Silent Pro Gold
OS: Windows 7 Ultimate 64-bit
Monitor: ASUS PQ321Q
Drivers: Nvidia 334.09, AMD 14.4
Performance
3DMark 11
3DMark is a computer benchmarking tool created and developed by Futuremark Corporation to determine the performance of a computer’s 3D graphic rendering and CPU workload processing capabilities. The latest version makes extensive use of all the new features in DirectX 11, including tessellation, compute shaders and multi-threading. We used the “Performance” preset for this benchmark.
3D Mark (Fire Strike Extreme)
This is another synthetic benchmark from Futuremark. We test the card on the Fire Strike Extreme test as it is designed for the enthusiast-level graphics performance. It test the DirectX11 capabilities, using extreme tessellation and volumetric illumination along with smoke simulations which use compute shaders and dynamic particle illumination.
Metro: Last Light
Metro: Last Light is a first-person shooter video game that continues Metro’s legacy of being one of the most demanding games for the best of GPUs. The game has a lot of DX11 eye-candy, which really puts a strain on any GPU. All DX11 features were enabled for the benchmark and we used the built-in benchmark for this test.
PS: Some of the cards have not been tested on the 4K display, which we got last week, as we did not have those cards on hand. We will update the charts as soon as we get the cards. 
Bioshock Infinite
Bioshock Infinite developed by Irrational Games uses a modified version of the Unreal Engine. It’s a first-person shooter which takes place in a fictional floating city. We ran the benchmark using the Adrenaline Benchmark tool and used the Ultra and Medium settings. In the Ultra settings you have FXAA On, Ultra texture detail, 16x AF, Ultra dynamic shadows and with the Medium settings you have FXAA On, High texture detail, 8x AF and High dynamic shadows.
Tomb Raider
Tomb Raider is a third-person shooter which is a reboot of the famous series with Lara Croft as the protagonist. In this reboot, she is ship-wrecked on a mysterious island which presents its own set of challenges which Lara has to overcome. The game is based on a customised CrystalEngine. It also features the AMD TressFX technology which adds more realism to hair rendering and physics. We ran the benchmark using the in-built benchmark tool at the Ultimate and Ultra settings. On the Ultimate preset, you have TressFX activated, 16x AF, FXAA whereas the Ultra preset has Normal hair quality, 16x AF, FXAA.
Sleeping Dogs
Sleeping Dogs is an open-world third-person action-adventure game developed by United Front games and Square Enix London Studios. Set in modern-day Hong Kong the game involves you playing as Wei Shan, an undercover cop, who infiltrates a local Triad gang. We ran the benchmark using the Adrenaline Benchmark tool and used the Ultra and Medium settings.
Temperatures
We noted the temperature of all the participating cards at idle and load states. The fan speeds were kept on auto. We noted maximum load temperatures after running benchmarks such as FurMark at 1080p preset, 3D Mark 11, Battlefield 3 and Metro: Last Light and taking an average.
Power Consumption
We used an energy monitoring device to measure the power consumed by the total rig with the graphics cards on them. The Cooler Master 800W PSU was plugged into the power meter which gave out the power consumed in Watts on the display. After powering on the system we let it stay unused for a good five minutes to get the idle power reading and then ran three instances of FurMark 1080p preset to get the maximum load power.
Verdict and Price in India
Looking at the gaming scores, the Nvidia GTX 980 is clearly ahead of its contemporaries. The results on the 4K display are impressive and the GTX 980 with its 4GB GDDR5 memory is more than enough to handle a single 4K display with ease. Sure it takes a hit with all the settings maxed out at 4K resolution, but then it is still ahead of the others. What are most impressive though are the temperature and power numbers. Look again at the total system power consumption chart – just 254 Watts consumed on load! That is seriously impressive for a flagship card.
Features such as Dynamic Super Resolution (DSR) and multi-frame anti-aliasing (MFAA) have the potential to improve image quality and keeping frame rates in high end games higher, while keeping the performance hit in check. We tried playing games with the DSR activated and noticed a hit of around 5-7FPS which isn’t that significant. The MFAA feature was not available at the time of testing and will be updated accordingly in future reviews.
The Nvidia reference card is priced around Rs 46,000 (so expect partner pricing to be around this region) which is not as high the GTX 780Tiwhen it had launched, and that is a good sign. Sure, for those already using the GTX 780/GTX 780Ti cards, it does not make much sense to upgrade as the performance improvement is under 10 percent. The GTX 980 also performs slightly ahead of AMD’s R9 290X which is selling for the same price point. The GTX 980s lower power consumption certainly gives it an edge. For those who are still on the GTX 500 / GTX 600 series, the GTX 980 gives enough incentives for an upgrade if you can afford it. The prices will certainly drop over the months. The GTX 970 also looks impressive specs-wise and is priced almost Rs 20,000 less than the GTX 980. We shall be reviewing that card soon, and let you know which of the two offers a better value proposition. So hold your horses till then.

Comments