Sign in with
Sign up | Sign in
Nvidia GeForce GTX 970 And 980 Review: Maximum Maxwell
By , Igor Wallossek,
1. Introducing GM204: There's A New Maxwell In Town

Nvidia introduced its newest graphics architecture, code-named Maxwell, about seven months ago in the mid-range GeForce GTX 750 and 750 Ti. While those cards don't deliver the high performance of a flagship gaming rig, they introduce unheard of energy efficiency to the PC graphics space. The Maxwell-powered GeForce GTX 750 Ti is a poster child for great performance per watt, achieving frame rates on par with graphics cards doubling its 65 W thermal ceiling.

The company gets there by applying the same advancements made to its Tegra family on GeForce, maximizing what each piece of a GPU can do given an unforgiving power budget.  But enthusiasts don’t really lust over efficiency alone. Mostly, they want breakneck speed. And we’ve wanted a performance-oriented implementation of Maxwell ever since the GeForce GTX 750 Ti surfaced.  Guess what Nvidia’s giving us today? Its GeForce GTX 980 and 970. If you game, this is the Maxwell you've been waiting for.

From a high level, the GM204 GPU in these cards looks like the GeForce GTX 750 Ti's GM107, scaled up appropriately. Like GM107, it leverages four partitioned blocks per Streaming Multiprocessor (SMM). Each SMM plays host to 128 CUDA cores, 32 load/store units, eight texture units, and a single PolyMorph engine. Two megabytes of L2 cache is shared across the entire GPU. The obvious difference, though, is the number of SMMs: while GM107 has five, the GeForce GTX 980’s GM204 features a staggering 16 of them. That adds up to a total of 2048 CUDA cores, 512 load/store units, 128 texture units, and 16 PolyMorph engines. Naturally, the second-fastest GeForce GTX 970 has some of those resources disabled, totaling 13 SMMs, 1664 CUDA cores, 416 load/store units, 104 texture units, and 13 PolyMorph engines.

Despite their similarities, the GM204 benefits from a lot of upgrades compared to GM107. For example, each SMM in the GeForce GTX 980 has a 96 KB chunk of shared address space for compute tasks, up from the 750 Ti's 64 KB. The new GPU's L1/texture cache size is now 24 KB per pair of CUDA core processing blocks (48 KB per SMM), doubling GM107.

Unlike GM107, the GM204 GPU features four Graphics Processor Clusters (GPCs) instead of one. That means it benefits from four times the number of raster engines. Of course, high-end graphics cards require a beefier back-end to handle all of that data throughput, and the GeForce GTX 980 utilizes four render back-ends capable of handling 16 full-color ROP operations per clock, adding up to 64. Four 64-bit memory controllers create an aggregate 256-bit bus. By the way, you may have noticed that the GeForce GTX 970's 13 SMMs don't divide equally into four GPCs. Nvidia says that there is no predefined recipe of SMMs per GPC in the 970, and each GPU may be configured differently.

The GeForce GTX 980 runs at a 1126 MHz base clock rate with a rated 1216 MHz GPU Boost frequency, while the GeForce GTX 970 has a 1050 MHz base and 1178 MHz typical GPU Boost rate. Both models carry 4 GB of GDDR5 RAM set to 1750 MHz, or 7 GT/s, yielding 224 GB/s of theoretical bandwidth.

It probably makes the most sense to compare GM204 to the Kepler-based GK110 in Nvidia’s GeForce GTX 780 Ti, even if we know it probably won’t go down in history as the most potent GPU based on Maxwell. The 780 Ti has 2880 CUDA cores, 240 texture units, 48 ROPs, and a 384-bit aggregate memory bus with 336 GB/s of memory bandwidth. Every one of those attributes (except for ROPs/clock) overpowers the GeForce GTX 980's specifications. But keep in mind that Maxwell is a more efficient architecture able to do more with fewer resources. For a quick refresher, you may want to have a look at our GeForce GTX 750 Ti Review: Maxwell Adds Performance Using Less Power.

The point is that we expect GeForce GTX 980 to compete on the 780 Ti's level, maybe even beating it. The GeForce GTX 980’s raster processing appears particularly well-endowed, with 64 full-color ROPs per clock and 72 Gpixels/second of fill rate. Compare this to the GeForce GTX 780 Ti's 48 ROPs per clock and 44 Gpixels/s. That should give GM204 an edge when it comes to high resolutions (like 4K) and anti-aliasing.

Before we go on, though, let's touch on Maxwell's focus on efficiency. Have a look at the GeForce GTX 980's thermal design power limit of 165 W. That's not a typo, and it's almost 100 Watts shy of the GeForce GTX 780 Ti. It's been a long time since we've seen a sub-200 W performance leader in the graphics card world. The GM204's 398 mm2 die size is less than 80% the size of GK110, despite being built on the same 28 nm fabrication node. Consider that this second generation Maxwell GPU is home to 5.2 billion transistors vs Kepler's 7.1 billion. The GeForce GTX 970 is even more of a power miser with a 145 W TDP. If these new GeForce cards can deliver performance on par with their predecessors, Nvidia has accomplished something special.

Regardless of Maxwell’s more efficient graphics architecture, the GeForce GTX 980 suffers a 33% drop in peak memory bandwidth compared to the GeForce GTX 780 Ti. Nvidia mitigates this deficit by implementing an improved memory compression scheme. GM204 sports an improved third-generation delta color compression mode that provides more options to the hardware. Using 8x8 blocks of 64 pixels, compression analysis is performed on 2x4 blocks in order to achieve the best compression possible in 8:1, 4:1, and 2:1 steps. If 2:1 compression isn't doable, that block isn't compressed. Successfully compressed blocks save even more bandwidth every time the memory is read. Nvidia claims that this increases effective memory bandwidth from 224 GB/s to approximately 300 GB/s compared to the GeForce GTX 680 Kepler-class cards. We'll see if this is enough to allow GeForce GTX 980 to stay competitive with the formidable GeForce GTX 780 Ti.

We should mention GM204's upgraded tessellation performance, too. The company's PolyMorph engine 2.0 design was leveraged across Kepler and even the Maxwell-based GM107. But that changes with the GM204's upgraded third-gen PolyMorph engine design. Nvidia representatives claim up to a 50% increase in performance at high tessellation factors per engine due to architectural improvements.

Nvidia's new GPU has some unique display options as well. The GeForce GTX 970 and 980 are the first graphics cards in the world that support the HDMI 2.0 standard, allowing for full 4K support at 60 Hz as an alternative to HDMI 1.2's 30 Hz limitation. The reference card design also includes two DisplayPort 1.2 outputs and a dual-link DVI connector, while the GPU is capable of driving four 4K displays simultaneously using a display engine that supports resolutions up to a forward-looking 5K.

Nvidia also updates its latest GPU with H.265 (HEVC) encoding support, providing a nice future-looking upgrade. As for existing H.264 support, keep in mind that Maxwell's fixed-function video encoder, already available on the GeForce GTX 750, is improved with 2.5x the throughput compared to previous-generation Kepler hardware. Nvidia's ShadowPlay software now features 4K video support at 60 FPS on the GM204, and this capability could possibly be enabled for other Maxwell-based cards in the future.

Nvidia wouldn't tell us whether GeForce GTX 980 is using an uncut version of the GM204 processor, so it's possible that we might see this GPU resurface with even more (or less) CUDA cores in a future product.

2. New Features

Nvidia's GM204 includes a smorgasbord of new functionality, and it's been a while since I've reviewed a GPU that brings so many features to the table. Some of these features will provide an immediate benefit to users right out of the box, while other functionality is definitely more forward-looking in scope and implementation. Let's start with the immediate benefits and work our way outward.

Dynamic Super Resolution

Dynamic Super Resolution (DSR) is a feature that has the potential to reduce jaggies at the edge of pixels, an effect known as 'aliasing'. Put simply, DSR is a form of anti-aliasing.

Those of you familiar with anti-aliasing may be aware of the oldest, most visually appealing, and probably the most demanding anti-aliasing technique available: SuperSampling Anti-Aliasing, or SSAA. SSAA renders a game at a higher resolution than the game is set to display, and downsamples the result to the target resolution. As a result, aliasing artifacts are greatly reduced. The biggest problem with SSAA is that it is so demanding on hardware resources that most game engines ignore it entirely. It can sometimes be forced through the driver, but this can be unreliable.

What Nvidia offers with DSR is a way to deliver supersampling antialiasing while bypassing the game's built in anti-aliasing limitations completely. With DSR, the GeForce card fools the graphics engine into believing that it the monitor can display higher resolutions than it is actually able to. The GeForce GTX 970/980 takes that high resolution information and downsamples it to the monitor's native resolution in real time. This provides the anti-aliasing benefit of SSAA without the need to program support into the software. Nvidia claims that as long as a game engine is capable of rendering to higher resolutions, DSR will work.


What's the downside? Just as SSAA is a performance killer, and so is DSR. Selecting a virtual 4K resolution in game to view at 1080p will saddle you with the same performance penalty you'd get by actually rendering to a 4K monitor. There might even be a slight extra latency added during the downsampling step, and we'll test for that later. Because of this, DSR is most promising for use with light game engines or older titles. If you're getting 120 FPS in Counter-Strike at maximum detail settings, for instance, DSR might be an excellent option to squeeze more visual fidelity out of the game without dropping the frame rate too much.

The other limiting factor for DSR is interface scalability. Not all games that can technically handle a 4K display will be coded to support it from a user interface perspective. Many modern titles have flexible and even user-selectable user interface scaling, but older games that would otherwise benefit from DSR may prove to be practically unusable thanks to on-screen buttons and indicators that are too small to see at 3840x2140. Of course, DSR isn't limited to a particular resolution, so users could opt for other resolutions such as 2560x1440 although it would not be as effective from an anti-aliasing perspective.

DSR is enabled through GeForce Experience software, and the driver panel provides controls for tweaking the sharpness of the down-sampling filter. While this feature will launch on the GeForce GTX 970 and 980, it will be moved over the rest of the GeForce product line over time.

We think this is a great technique that could bring much-needed visual improvement to many, less visually advanced game titles. It's especially compelling it turns out to be as backward-compatible as we're told it is.

 Multi-Frame Sampled Anti-Aliasing

Multi-Frame Sampled Anti-Aliasing (MFAA) is yet another method to reduce pixel stepping artifacts, also introduced with the GeForce GTX 970 and 980. Like DSR it is based on an older technique, and part of its appeal is how Nvidia has implemented it in way that is transparent to the game engine to ensure compatibility. The key difference is that DSR is designed to increase fidelity in games that are graphically maxed out, at the cost of performance. MFAA, on the other hand, is intended to increase performance without sacrificing visual fidelity.

At its core MFAA is very similar to an anti-aliasing mode introduced with the Radeon X800 called Temporal MSAA. AMD has since abandoned this mode as its implementation had flaws, but Nvidia appears to have resurrected the technique with a simple improvement that might make all the difference: a temporal synthesis filter.

If you're not familiar with Temporal MSAA, it's a method to increase the number of apparent sample positions by switching the MSAA filter pattern every alternating frame. Thanks to visual persistence in the human eye, the result looks similar to higher numbers of filter samples. For example, 2x sample Temporal MSAA can appear similar to 4x MSAA, which requires twice the number of filter positions per pixel and results in a lower frame rate. The problem with Temporal MSAA as it was implemented in Radeon cards, though, is that visual artifacts can occur during subtle movement. In addition, high frame rates are a must. When performance drops below 60 FPS, the effect is turned off.


Nvidia may have solved both of these problems with the use of a temporal synthesis filter that takes into account both pixel samples over time and movement in the scene, and then averages out a result. This allows MFAA to work at frame rates below 60 FPS, and should minimize artifacts during camera movement.

Of course when the user's camera moves very quickly MFAA won't be able to keep up, but the beauty of this scenario is that aliasing artifacts are not something that users can really experience when the camera is moving that fast. Put simply, when MFAA breaks down due to a lot of movement, it probably doesn't matter because you won't be able to notice. If it works like Nvidia says it does, MFAA can provide 4x MSAA quality at the performance cost of 2x MSAA.

While the company told us to expect it soon, MFAA is not enabled for the GeForce GTX 970 and 980 launch. We have to wait for more details on the time frame and implementation.

VR Direct

With DSR and MFAA explained, we'll begin to consider features that are a work in progress and probably won't see availability for at least a few months. The first of these features is VR Direct, Nvidia's blanket of technologies designed to enhance experience with virtual reality (VR) head mounted displays (HMDs) like the Oculus Rift.

If you're a VR enthusiast, you may have already considered the tremendous potential benefit that DSR could bring to the resolution-challenged Oculus Rift. With a single 1080p screen split to 960x1080 resolution for each eye, the potential to minimize aliasing at this relatively low resolution is obvious. MFAA is also an anti-aliasing option that might make sense for use with an HMD, as the light computational load of 2x MFAA has a lower latency cost compared to true 4x MSAA.

Speaking of latency, Nvidia has already been working to eliminate this problem on the VR graphics pipeline. Latency can be a crippling factor when it comes to HMDs, decoupling the user from their movement and pulling them out of the experience. Asynchronous Warp is a technology that Nvidia designed to combat HMD latency by doing as much of the graphics work as possible ahead of time, re-checking the HMD position closer to render time, and adjusting the frame to better sync with the user's orientation as soon as possible.

In addition, the company mentioned the possibility of an auto stereo driver. One of the first things I thought when I saw the Oculus Rift was  how great it would be if Nvidia coded support for the HMD into the 3D Vision driver, allowing users to experience virtual reality in existing game titles that don't have native support. Clearly someone at Nvidia had the same idea, and I look forward to seeing what the company comes up with.

DirectX 12

Microsoft has committed that its new graphics API will ship in time for the holiday season in 2015, and developer access to the current build is already available. Full implementation might be more than a year away, but it's somewhat relevant because Nvidia claims that the GeForce GTX 970 and 980 are DirectX 12 compatible.

If you've heard anything about this API update, you probably know that one of the core features of DirectX 12 is reduced CPU overhead by giving developers greater control over resources. This includes features such as pipeline state simplification and command multithreading that allow for greatly improved processing allocation across multiple CPU cores.

DirectX 12 isn't just about leveraging CPU resources more efficiently, though. It's also about new and more efficient rendering features and techniques for developers. These include rasterizer ordered views, which allows for correct and consistent rendering of overlapping semi-transparent objects ordering for the first time; tiled resources, allowing for the 'megatexture' technique to reduce memory use for textured tiles or sparse volumes; and conservative rasterization, a method for testing the contents of an entire pixel instead of just a single point sample, which can be useful for scenarios such as collision detection.

There's a lot of interesting technical data regarding these features, but considering the compressed GeForce GTX 970 and 980 launch we don't have a lot of time to dig into the specifics. We'd rather concentrate on Nvidia's new graphics cards and revisit DirectX 12 as we get closer to its launch. By the way, many of DirectX 12's features will be released in  a DirectX 11.3 update in order to bring them to Windows 7, too.

Voxel Global illumination

Nvidia's Voxel Global Illumination (VXGI) is perhaps the most ambitious feature included in the GM204 GPU (it wasn't included in the GM107 used in the GeForce GTX 750 series). VXGI represents a wholesale rethinking of the way games are made, a transition to a realistic global illumination lighting model that players have never experienced in real time before, and a tremendous computational load.

VXGI uses voxel space and voxel cone tracing to deliver an approximation of realistic lighting, one that is much faster to compute when compared to standard 3D ray tracing methods. Let's start by defining some of those terms.

'Voxel' is a combination of the words 'volume' and 'pixel', and represents a volume in a three-dimensional grid in space. Imagine the room you're currently in was completely filled with invisible boxes stacked tightly together all the way to the ceiling. Each of those boxes represents a voxel, a volume in 3d space. The voxel grid is used as a data structure to store light emission and opacity data.

Cone tracing is a technique that uses different resolutions of voxels along a light path to store data about how the light affects objects in its path. This is computationally efficient as the size of the voxels grows (the resolution drops) as the light cone travels further from its last bounce.

Rendering is computed in three steps: the voxelization of the scene, the injection of light, and finally, the final gather. During the final step the geometry is rendered, light is evaluated, and indirect light is gathered to produce a photorealistic result. 

Nvidia demonstrated VXGI with the use of a moon landing demo that realistically recreates a famous picture of the original moon landing. It looks absolutely authentic for a real-time lighting demo. Having said that, it was demonstrated on a system running two GeForce GTX 980 cards in SLI, but it didn't look like a very complex scene and featured only a single light source. The potential computational requirements to run VXGI in a full-fledged game gives us pause.

In many ways VXGI is a technology that is more game-developer oriented than consumer oriented at this point.  It potentially simplifies the way games are created: with current game engines, lighting effects need to be baked on to textures in a scene. With VXGI everything would be rendered for the player in real time, removing significant steps in the creation process and allowing developers to make changes to a game level without adding workloads. VX3D is attractive as a development kit for game creators to play with this new lighting model.

I think global illumination and path tracing represent the inevitable future of real-time photorealistic game graphics. From a consumer standpoint though, VXGI is in the early chicken-or-egg scenario that every disruptive technology suffers. I'm somewhat skeptical that the GeForce GTX 970 and 980 are fast enough to perform the tremendous number of calculations required by voxel global illumination.  Having said that, perhaps VXGI can be used to augment current lighting models for more realism without completely replacing them. I really look forward to seeing how this technology will be integrated into the Unreal 4 engine, which Nvidia told us will happen by Q4 of this year, and I'm anxious to see the effect it has on visual fidelity and frames-per-second performance.

3. Nvidia GeForce GTX 980 Reference Card

At this point, let's examine Nvidia's reference design for the GeForce GTX 980. It's dressed a slight variation of the same classy outfit that the company debuted in the GeForce GTX 690. This isn't a particularly large card but it does have significant weight to it, some of which is attributable to the metal backplate.

The Speeds And Feeds
Form Factor
Dual-slot design
Length (from Slot Panel to End)
268 mm (10.5")
Height (from Slot to Top)
111 mm (4 3/8")
Depth 1 (from PCB to Front Cover)
36 mm (1 5/16")
Depth 2 (from PCB to Back Plate)
3 mm (1/8")
Weight
1030 g
Connectors
1x DL-DVI-I, 1x HDMI 2.0, 3x DisplayPort

The 65mm radial fan is the same model used on the original GeForce Titan. Indeed, the design of the GeForce GTX 980 is almost indistinguishable from the titan at a casual glance (until you see the model number emblazoned on the side, anyway). The main visual difference is that the cooling fins behind the transparent cover are painted black.

The back plate has a nifty little door that can be removed if the user wishes to improve cooling when placed beside another card in SLI. As a point of interest, the Titan does not have a back plate.

Note the two 6-pin PCIe power connectors: it's odd to see so few pins on a high-end card. The GeForce GTX name glows in green when the card is in operation, of course.

Output options consist of three DisplayPort 1.2 ports, a dual-link DVI port, and a single HDMI port. This is the first graphics card with an HDMI 2.0 plug by the way, which allows for 4K video at 60 Hz. With the previous HDMI standard, two ports need to work in tandem to handle the bandwidth required by a 4K 60 Hz display.

4. Gigabyte GTX 980 WindForce OC

Now let’s take a look at the technical specifications of Gigabyte's GeForce GTX 980. It’s overclocked from the factory and, according to the manufacturer, uses a low-leakage GPU able to overclock well at lower power consumption. Indeed, our test sample fared particularly well in the power measurements, demonstrating the Maxwell architecture's potential once manufacturing is refined and there's less variance from chip to chip.

A total of six video outputs (of which four can be active at any given time) are great for multi-monitor configurations. Gigabyte keeps the second DVI connector that didn’t make it onto Nvidia's reference design. And HDMI 2.0 support makes the new card a bit more future-proof.

The Speeds And Feeds
Form Factor
Dual-slot design
Length (from Slot Panel to End)
297 mm
Height (from Slot to Top)
108 mm
Depth 1 (from PCB to Front Cover)
35 mm
Depth 2 (from PCB to Back Plate)
5 mm
Weight
1196 g
Connectors
1x DVI-I, 1x DVI-D, 1x HDMI 2.0, 3x DisplayPort
Pros
- GPUs binned for better overclocking and lower power consumption
- Total of six video outputs (better connectivity)
- Dual-slot design, well-suited for SLI
- Lots of cooling headroom
Cons
- RPM under full load is too high (BIOS update expected to fix this)

The 600 W-rated cooler first used on Gigabyte's GeForce GTX Titan Black WindForce OC should be able to handle anything this new card throws at it. After all, it was well-behaved on top of the big GK110 GPU.

Unfortunately, the backplate doesn't help with cooling; its only purpose is stabilizing the PCB.

Air blows up and down, rather than back. That's better than shooting it all toward your hard drives and SSDs, though we'd much rather see a high-end graphics card push its waste heat out of the chassis instead.

Two 8-pin auxiliary power connectors are supposed to provide stable overclocking. The WindForce label lights up in blue.

A total of three DisplayPort connectors, two dual-link DVI connectors, and an HDMI 2.0 connector fill the slot panel almost completely.

5. Gigabyte GTX 970 WindForce OC

The Gigabyte GTX 970 WindForce OC’s specifications are quite different from the 980's in some places. According to Gigabyte, the 970 also uses GPUs binned for better-than-average overclocking potential and lower power consumption. Still, our tests show why this graphics card costs less than the flagship.

Fortunately, it retains the higher-end board's six video outputs (of which four can be actively used at a time). And again, Gigabyte keeps the second DVI connector that’s missing from the reference design. HDMI 2.0 makes this new card somewhat more future-proof.

The Speeds And Feeds
Form Factor
Dual-slot design
Length (from Slot Panel to End)
297 mm
Height (from Slot to Top)
108 mm
Depth 1 (from PCB to Front Cover)
35 mm
Depth 2 (from PCB to Back Plate)
5 mm
Weight
797 g
Connectors
1x DVI-I, 1x DVI-D, 1x HDMI 2.0, 3x DisplayPort
Pros
- GPUs binned for better overclocking and lower power consumption
- Total of six video outputs (better connectivity)
- Dual-slot design, well-suited for SLI
- Suitable cooling headroom
Cons
- RPM at idle is too high (35 percent)

Gigabyte's GTX 970 WindForce OC sports the WindForce cooler that we know from the company's older GeForce cards. It’s still a solid cooler that’s both thin and fairly quiet. It’s been updated with the new rotor blades found on the WindForce 600 W.

Once again, the backplate doesn’t contribute to the graphics card’s cooling. It only provides some stability and a nice finishing touch we aren't used to seeing on Nvidia-based boards.

The vertically-oriented fins send waste heat toward the top and bottom, and not toward your hard drives and SSDs. It'd be better to exhaust it entirely, though.

One 8- and one 6-pin PCIe power connector are supposed to keep the Gigabyte GTX 970 WindForce OC stable when overclocked. As on the 980, this card's WindForce label lights up blue.

You get a total of three DisplayPort connectors, two dual-link DVI connectors, and an HDMI 2.0 connector on the rear bracket. Together, they take up almost all the available space.

6. EVGA GTX 970 Superclock ACX 2.0

EVGA sent us its superclocked take on the GeForce GTX 970, outfitted with the company's ACX 2.0 cooler. It's slightly smaller in length than the GeForce GTX 980 reference card, but significantly lighter at 639 grams. While its 1165 MHz nominal clock rate is substantialy higher than the 1050 MHz reference spec, this card's 1317 MHz boost clock is impressive and one of the highest factory overclocks we've seen.

The Speeds And Feeds
Form Factor
Dual-slot design
Length (from Slot Panel to End)
243 mm (9.5")
Height (from Slot to Top)
111 mm (4 3/8")
Depth 1 (from PCB to Front Cover)
35 mm (1 5/16")
Weight
639 g
Connectors
1x DL-DVI-I, 1x DL-DVI-D, 1x HDMI 2.0, 1x DisplayPort

EVGA's card is actively cooled by two 87mm axial fans. The company claims to have focused on details like fan power draw and blade quality to provide 26% lower temperatures and 250% less fan power consumption than a reference fan and cooler setup, with 36% less noise.  Fan draw is part of any graphics card's power requirement, so it's interesting that the company has taken this approach to efficiency. EVGA even suggests that the power consumption of fans can directly affect overclocking ability.

On the top of the card you'll find two 6-pin power inputs as well as dual SLI connectors, which suggests that the card supports triple- and quad-SLI configurations.

The GTX 970 Superclock features three small (~5mm) and two large (~8mm) heat pipes that worm through nickel-plated fins and tie in to an aluminum cooling block.

For monitor connectivity there is a single DisplayPort 1.2 output, DVI-I and DVI-D outputs, and a single HDMI port. Like the GeForce GTX 980, the 970 is also equipped with the new HDMI 2.0 standard.

7. Test System And Benchmarks

Benchmarking Hardware And Software

The GeForce GTX 980, 9970, and 780 Ti were tested with the 344.07 launch driver, band all of the Radeon cards were outfitted with the Catalyst 14.7 release candidate for testing.

We selected a variety of newer game titles with high detail settings at a resolution of 1920x1080 in order to give the GeForce GTX 970 and its competitors a solid, real-world workload that this class of card should be able to handle. Frankly, it surprised us with its formidable capabilities.

The factory-overclocked EVGA GeForce GTX 970's core clock was dropped to the 1050/1178 MHz nominal/boost reference specification in order to show what a typical specimen should be able to accomplish. Keep in mind that there is no reference cooler for this card, so all GeForce GTX 970s will be unique in this respect.

Two of the games we're testing have an option to use a Mantle code path, so we're running those benchmarks (Thief and Battlefield 4) with Mantle enabled and disabled to measure the API's impact.

High-end graphics cards require a substantial amount of power, so XFX sent us its PRO850W 80 PLUS Bronze-certified power supply. This modular PSU employs a single +12 V rail rated for 70 A. XFX claims continuous (not peak) output of up to 850 W at 50 degrees Celsius.

We've almost exclusively eliminated mechanical disks in the lab, preferring solid-state storage for alleviating I/O-related bottlenecks. Samsung sent all of our labs 256 GB 840 Pros, so we standardize on these exceptional SSDs.


Test System
CPU
Intel Core i7-3960X (Sandy Bridge-E), 3.3 GHz, Six Cores, LGA 2011, 15 MB Shared L3 Cache, Hyper-Threading enabled.
Motherboard
ASRock X79 Extreme9 (LGA 2011) Chipset: Intel X79 Express
Networking
On-Board Gigabit LAN controller
Memory
Corsair Vengeance LP PC3-16000, 4 x 4 GB, 1600 MT/s, CL 8-8-8-24-2T
Graphics

Reference GeForce GTX 980
1126/1216 MHz GPU, 4 GB GDDR5 at 1753 MHz (7000 MT/s)

EVGA GeForce GTX 970

1165/1365 MHz GPU, 4 GB GDDR5 at 1753 MHz (7012 MT/s)
(underclocked GPU to reference 1050/1178 MHz specification for benchmarks)

Nvidia GeForce GTX 780 Ti
875/928 MHz GPU, 3 GB GDDR5 at 1752 MHz (7008 MT/s)

Nvidia GeForce GTX 770

1046/1085 MHz GPU, 2 GB GDDR5 at 1752 MHz (7008 MT/s)

AMD Radeon R9 280X
850/1000 MHz GPU, 3 GB GDDR5 at 1500 MHz (6000 MT/s)

AMD Radeon R9 290
947 MHz GPU, 4 GB GDDR5 at 1250 MHz (5000 MT/s)

AMD Radeon R9 290X
1000 MHz GPU, 4 GB GDDR5 at 1250 MHz (5000 MT/s)
SSD
Samsung 840 Pro, 256 GB SSD, SATA 6Gb/s
Power
XFX PRO850W, ATX12V, EPS12V
Software and Drivers
Operating System
Microsoft Windows 8 Pro x64
DirectX
DirectX 11
Graphics Drivers
All GeForce Cards (except GTX 770): Nvidia 344.07 Launch Driver
All Radeon cards: AMD Catalyst 14.7 RC 1
GeForce GTX 770: Nvidia 340.52 WHQL
Benchmarks
Watch Dogs
Version 1.04.497, Custom THG Benchmark, 90-sec FRAPS, Driving
Arma 3
V. 1.26.126.789, 30-sec. Fraps "Infantry Showcase"
Battlefield 4
Version 1.3.2.3825, Custom THG Benchmark, 90-Sec
Assassin's Creed IV: Black Flag
Custom THG Benchmark, 40-Sec
ThiefVersion 1.6.0.0, Built-in Benchmark
Grid Autosport
Version 1.0.101.4672, Built-In benchmark
Far Cry 3
Version 1.05, Custom THG Benchmark, 55-sec FRAPS

We'd like to voice our appreciation to Damian and the folks at Memory Express in Canada who helped us with some last-minute equipment requirements in order to perform the benchmarks for this review.  Thanks, gents!

8. Results: Battlefield 4 And Thief

Battlefield 4

We'll begin our benchmarks with one of the most popular first person shooters available today, powered by the technologically advanced Frostbite 3 graphics engine: Battlefield 4.

Nvidia's new GeForce GTX 980 puts achieves an impressive showing, easily sailing past the GK110-powered GeForce GTX 780 Ti and earning the top spot. What's more, the GeForce GTX 970 bests the Radeon R9 290X in DirectX, and AMD's single-GPU top dog can only achieve parity when it leverages the company's Mantle API.

But 1080p is low-hanging fruit for Nvidia's new graphics cards, so let's up the ante to 4K. Unfortunately, we had to pull back the level of detail from Ultra to High in order to allow the GeForce GTX 980 to achieve our target 30 FPS minimum/40 FPS average frame rate. 

The GeForce GTX 980 still leads the pack, but by a smaller margin. The GeForce GTX 970 falls back a bit relative to the Radeons, but still manages a strong showing considering its relatively low MSRP.

Thief

Thief is another modern title that can be difficult for today's graphics hardware to handle. We'll begin at 1080p with the very high detail preset invoked.

The GeForce GTX 980 achieves the best showing as it did in Battlefield 4, and the GeForce GTX 970 is right behind the GeForce GTX 780 Ti. Even when outfitted with the Mantle code path, the Radeon R9 290X can't beat Nvidia's new $329 card. The GeForce GTX 970 is shaping up to be an impressive value proposition. But what about Ultra HD resolution?

Once again, we had to dial back the visual settings to keep things smooth, trading the very high detail preset for the normal detail preset. Still, it looks very impressive at 3840x2160. A note about frame time latency: Thief's graphics engine suffers from a little lag across all of the cards we tested for this benchmark, unfortunately, but the good news is that we haven't found it to be noticeable while actually playing the game.

9. Results: Arma 3 And Grid Autosport

Arma 3

And now let's put everyone's favorite hyper-realistic combat simulator, Arma 3, to the test:

It looks like the GeForce GTX 980, 970, and 780 Ti are hitting a platform bottleneck at 1080p when the ultra detail setting is used in conjunction with 4x MSAA. Nvidia's newest products continue to shine in this benchmark and we have a pretty clear expectation of performance by now. Note how the Radeon R9 280X suffers from some significant frame time variance spikes, it's not bad enough to be a deal breaker but you can notice it while playing.

When we increase resolution to 3840x2160, the GeForce GTX 970 falls behind the pack slightly where raw frame rates are concerned, but it maintains a playable pace and a relatively flat frame time variance. The same can not be said for the Radeon cards, which suffer from some judder at this high resolution. 

Grid Autosport

Grid Autosport is the newest installment in the popular racing series, and we've recently added it to our benchmarking suite. As usual, Codemasters did a wonderful job with their game engine, allowing us to push all of the details to their maximum settings, crank MSAA up to 8x,  and maintain incredibly smooth performance.

Note how none of these cards spend a significant amount of time below 60 FPS. It should come as no surprise by now that the GeForce GTX 970 and 980 can easily handle 1920x1080 at ultra high detail settings in almost every game we throw at them. Let's see how these cards handle those same settings, including hardware intensive 8x MSAA, on a 4K monitor.

All of the cards tested managed to maintain the minimum frame rate above 30 FPS, and the majority kept the average frame rate above 40 FPS. Those that missed the mark did it by a small margin, meaning that this game is quite playable with 8x MSAA at 4K!

10. Results: Assassin's Creed IV, Watchdogs, Far Cry 3

Assassin's Creed IV: Black Flag

Despite its status as an Nvidia Gameworks partner title, Assassin's Creed IV: Black Flag is known to perform well with the Hawaii GPU in the Radeon R9 290 series. Let's see how the new GeForce cards compare:

At 1080p the high-end graphics cards run in to a clear platform bottleneck, so it's difficult to call out any real winners amongst these super elite products. Perhaps bumping up the resolution will separate the men from the boys:

It remains a tight race but the GeForce GTX 980 has demonstrated a clear trend of achieving best-in-class performance at 4K, and manages over 40 FPS average in this case with very high details enabled.

Watchdogs

Watchdogs can be brutal on medium-range hardware, but it does not present an insurmountable burden for the top-tier graphics hardware we're throwing at it today.

Even with the high detail preset enabled, all of these graphics cards handle 1080p without too much problem, but you can see that this game engine is plagued with significant frame time variance spikes. This is most likely caused by loading chunks of the large game world as we travel through it in a speeding vehicle.

Increase the resolution to 3640x2160, though, and we need to settle for the medium detail preset. Even so, the frame time variance spikes increase and impact the player's experience. The issue isn't distracting enough to make the game unplayable, though.

Far Cry 3

The final game in today's benchmark tests, Far Cry 3 remains a sterling example of a beautiful graphics engine with lush foliage that is affected by your avatar as you walk through the jungle.

The new GeForce cards stand tall here, as they have throughout the rest of our benchmark suite. You can see that Radeon graphics cards demonstrate considerably more frame time variance than their competitors, which can manifest as micro stutter and be an annoyance in this game.

With the resolution increased to 4K, details must be dropped to the medium preset in order to maintain playability with even the fastest single-GPU cards. The GeForce GTX 970 and 980 distinguish themselves yet again, although the Radeon frame time variance issue we saw at 1080p has curiously become less of a problem.

11. A New Power Consumption Test Setup

Nvidia's newest architecture presents us with a whole new set of challenges for measuring power consumption. If the maximum of all four possible rails are to be measured exactly (to find out Maxwell’s power consumption reduction secrets), then a total of eight analog oscilloscope channels are needed. This is because voltage and current need to be recorded concurrently at each rail in real-time. If the voltages are measured and then used later, the result may be inaccurate. So, how did we solve this problem?

We enlisted the help of HAMEG (Rohde & Schwarz) to search for a solution with us. In the end, we had to use two oscilloscopes in parallel (a master-slave triggered setup), allowing us to accurately measure and record a total of eight voltages or currents at the same time with a temporal resolution down to the microsecond.

The measurement intervals need to be adjusted depending on the application in question, of course, in order to avoid drowning in massive amounts of data. For instance, when we generate the one-minute graphs for graphics card power consumption with a temporal resolution of 1 ms, we have the oscilloscope average the microsecond measurements for us first.

We use a riser card on the PCIe slot (PEG) to measure power consumption directly on the motherboard for the 3.3 and 12 V rails. The riser card was built specifically for this purpose.

In addition, we separately measure the voltage and current at each of the two individual PCIe power connectors.

Test MethodologyNo-contact current measurement at all rails
Direct voltage measurement
IR real-time monitoring
Test Equipment2 x HAMEG HMO3054, 500 MHz Four-Channel Oscilloscope with Data Logger
4 x HAMEG HZO50 Current Probe
4 x HAMEG HZ355 (10:1 Probe, 500 MHz)
1 x HAMEG HMC8012 DSO with Data Logger
1 x Optris PI450 80 Hz Infrared Camera + PI Connect
Test SystemIntel Core i7-5960X, 4.2 GHz
16 GB G.Skill Ripjaws DDR4-2666 (4 x 4 GB)
MSI X99 Gaming 7
2 x Transcend SSD370 (System, Applications + Data, Storage)
be quiet! Dark Power Pro 1200 W
Microcool Banchetto 101

Nvidia’s GPU Boost Accelerates Maxwell

Everything makes sense in theory, but we still want to know how Maxwell achieves better efficiency at this magnitude. Kepler already adjusted the GPU’s voltage quickly and exactly depending on its load and temperature, and AMD’s PowerTune did the same thing as well. It turns out that Maxwell refines the formula further. With its shaders fully utilized, the new architecture's advantage over Kepler practically vanishes. So, Maxwell depends on its superior ability to adjust to changing loads, and, consequently, it’s able to tailor the power consumption even better to the needs of the application in question. The more variance there is, the better Maxwell fares.

To illustrate, let’s take a look at how Maxwell behaves in the space of just 1 ms. Its power consumption jumps up and down repeatedly within this time frame, hitting a minimum of 100 W and a maximum of 290 W. Even though the average power consumption is only 176 W, the GPU draws almost 300 W when it's necessary. Above that, the GPU slows down.

Now, how do the PSU’s important 12 V rail's current and voltage behave under these conditions? We add them up for this purpose.

The PSU doesn’t supply a constant 12 V supply voltage. Spikes that empty the secondary capacitors and the slower power supply that tries to fill them back up again cause voltage fluctuations that the graphics card's five phases must take care of. All of these interactions make our measurements more complicated than a log of averages would suggest.

12. Power Consumption In Detail

Idle Power Consumption

[EDIT] Our original Nvidia GeForce GTX 980 reference sample suffered from a BIOS issue that caused a higher-than-expected power draw. We flashed the card with the reference BIOS and have updated the charts below with the new results. [/EDIT]

There’s a huge difference between the two Gigabyte graphics cards. Nvidia’s reference board is almost exactly in the middle between them. The GTX 980 WindForce OC gives us our baseline by idling just under 10 W, while the derivative 970 consumes almost 20 W. This (relatively) massive difference was measured on the same platform using the same measurement methodology. Nvidia’s reference graphics card comes in at 16 W, right between the Gigabyte boards.

Our best guess is that Gigabyte has a “golden sample” at work here. But it goes to show what could be possible from Maxwell if production is refined further. We weren’t able to get our hands on a reference Nvidia GeForce GTX 970, so we had to use a downclocked version of a board partner’s card instead.

Power Consumption at Idle
Gigabyte GTX 980
Windforce OC
GeForce GTX 980
Reference
Gigabyte GTX 970
Windforce OC
PCIe Total3.99 W
3.78 W
4.56 W
Motherboard 3.3 V0.18 W
1.93 W1.32 W
Motherboard 12 V5.05 W
10.05 W
13.50 W
Total 9.22 W
15.76 W 19.34 W

Gaming Power Consumption

These findings further illustrate what we said on the previous page about Maxwell and its ability to regulate GPU voltage faster and more precisely. The gaming-based power consumption numbers show just how much efficiency can be increased if the graphics card matches how much power is drawn to the actual load needed. The version of the GeForce GTX 980 that comes overclocked straight from the factory manages to use significantly less power than the reference version, while offering six percent more performance at the same time.

Power Consumption while Gaming
Gigabyte GTX 980
Windforce OC
GeForce GTX 980
Reference
Gigabyte GTX 970
Windforce OC
PCIe Total137.38 W
143.85 W
132.70 W
Motherboard 3.3 V1.07 W 2.81 W2.27 W
Motherboard 12 V34.98 W39.04 W
42.84 W
VGA Card Total173.43 W 185.70 W
177.81 W

Let’s take a look at the measurements in detail. The large load fluctuations are plain to see.

Nvidia GeForce GTX 980 Reference Card Measurement Results and Graphs

Gigabyte GTX 980 WindForce OC Measurement Results and Graphs

Gigabyte GTX 970 WindForce OC Measurement Results and Graphs

Stress Test Power Consumption

If the load is held constant, then the lower power consumption measurements vanish immediately. There’s nothing for GPU Boost to adjust, since the highest possible voltage is needed continuously. Nvidia's stated TDP becomes a distant dream. In fact, if you compare the GeForce GTX 980’s power consumption to an overclocked GeForce GTX Titan Black, there really aren’t any differences between them. This is further evidence supporting our assertion that the new graphics card’s increased efficiency is largely attributable to better load adjustment and matching.

Power Consumption Under Max. Load
Gigabyte GTX 980
Windforce OC
GeForce GTX 980
Reference
Gigabyte GTX 970
Windforce OC
PCIe Total237.63 W
122.70 W
192.46 W
Motherboard 3.3 V0.98 W1,83 W 2.15 W
Motherboard 12 V41.07 W 52.83 W
48.36 W
VGA Card Total279.68 W 177.36 W242.97 W

The values above have potential consequences for the everyday operation of these graphics cards, as they represent what can be expected when running performance-hungry compute-oriented applications optimized for CUDA and OpenCL.

That's not the only offering that makes a good impression, though. Nvidia's reference GeForce GTX 980 does well too, as long as you don’t focus on the idle power measurement. This isn't the same result as custom models with higher power targets (up to 300 Watts for the GTX 980), though, when compute-based stress tests are run. A taxing load just doesn't give Maxwell room for its optimizations to shine.

Note that the GeForce GTX 980's stress test power consumption is actually a few watts lower than the gaming result. This is likely due to throttling that kicks in when we hit the thermal ceiling.

Finally, let’s take a look at the detailed measurements. The much smaller load adjustments are illustrated graphically below. Pay particular attention to the reference card’s small drops when it hits the thermal limit.

Nvidia GeForce GTX 980 Reference Card Measurement Results and Graphs

Gigabyte GTX 980 WindForce OC Measurement Results and Graphs

Gigabyte GTX 970 WindForce OC Measurement Results and Graphs

Maxwell’s performance can almost be compared to that of a compression tool. Already compressed data (our compute-heavy numbers) can’t be compressed any further, whereas text (games) can be compressed massively.

13. Power Consumption Overview

[EDIT] We originally posted Power Consumption Torture (GPGPU) results that showed a simulated GeForce GTX 970 reference card pulling over 240 Watts. This does not represent Nvidia's reference GeForce GTX 970 board because our data point was simulated with a Gigabyte GTX 970 card that has a non-reference ~250 Watt power target, unlike the reference board's ~150 W power target.

We have since pulled that data since it does not represent Nvidia's reference GeForce GTX 970 card. On the other hand, as far as we know there are no actual GeForce GTX 970 reference card designs for sale as each manufacturer has put their own spin on this model. None of the manufacturers we have talked to have released a GeForce GTX 970 card with a ~150 Watt power target as of this time, opting instead to give this product more performance headroom.

This is an issue we are keeping a close eye on, and we will follow up with a detailed investigation in the near future. We are curious to see if a reference-based GeForce GTX 970 will perform in the same league as the cards we have tested with higher power targets, but it would certainly make more sense in an HTPC or for use in smaller form factors. In the meantime, we have removed the 'simulated' GeForce GTX 970 data point from the following charts. [/EDIT]

Now it's time to compare our measurements to some of the other relevant graphics cards. Gigabyte's GTX 980 WindForce OC stands out yet again, especially when it comes to our idle and gaming readings. It’s amazing what a specially-selected Maxwell GPU can do.

That's not the only offering that makes a good impression, though. Nvidia's reference GeForce GTX 980 does well too, as long as you don’t focus on the idle power measurement. And the party ends as soon as you look at the compute-based stress test results. A taxing load just doesn't give Maxwell any room for its optimizations to shine.

When it comes down to it, it's possible for our most taxing workloads to take Maxwell back to Kepler-class consumption levels. In fact, Gigabyte's factory overclocked GeForce GTX 980 actually draws more power than the GeForce GTX Titan Black without offering a substantial performance gain in return. As you can see below, the reference GeForce GTX 980 draws substantially less power, though.

14. Efficiency

Frankly, though, the GeForce GTX 980's primary purpose is gaming. So to be fair, we're only looking at the card's efficiency in that context. After the launch, we'll spend more time covering the other disciplines in a follow-up story.

Maxwell does very well, even without Gigabyte’s golden sample factored in. This observation is based on the power consumption measurements in our 2014 VGA charts. If Nvidia's reference GeForce GTX 980 sets the bar at 100 percent, then all of the other Maxwell-based cards fall in line right above it. Non-Maxwell-based boards show up below. Not surprisingly, Gigabyte’s golden sample ends up at the top, and even manages to pass a GeForce GTX 750 in the process, which is no small feat for a card built using a large GPU.

Thinking back to the maximum versus average power consumption findings for gaming, one fact becomes abundantly clear: AMD’s issue is not absolute performance or the efficiency of its architecture, but rather that PowerTune technology can’t adjust the power consumption quickly or finely enough depending on the actual load. This is exactly where Nvidia scores most of its points with Maxwell.

But even Nvidia can’t change the laws of physics. The new cards’ power consumption during compute-based testing demonstrates this fact very clearly. There are some ways around these laws, however, and the company's engineers seem to have found them. A gamer would say that they simply have the better skills.

15. Temperatures And Noise

Temperatures During Normal Operation

Lower power consumption should result in lower GPU temperatures under load, which in turn should necessitate a simpler and quieter cooling solution. That’s the theory anyway, so let’s have a look at the GeForce GTX 980 in reality. We’re focusing on its behavior in a closed case, given the number of reader requests we've received about this topic.

Interestingly, we’ve identified a method to confirm what our sensors tell us and to spice up our usual temperature graphs a bit in the form of the PI450 by Optris. This piece of equipment is an infrared camera that was developed specifically for process monitoring.

It supplies real-time thermal images at a rate of 80 Hz. The pictures are sent via USB to a separate system, where they can be recorded as video. The PI450’s thermal sensitivity is 40 mK, making it ideal for assessing small gradients.

No Biggie: 15 Minutes at Idle

Let’s first take a look at the system’s state after the motherboard and graphics card warm up for 15 minutes at idle. We keep the chassis' side panel closed during this test, of course, only opening it briefly to snap a quick picture with our infrared camera.

Barbeque: 1-Hour Demanding Gaming Loop

Our second scenario changes the infrared picture significantly, really driving home differences between each graphics card's cooling solution. We begin with Nvidia's reference GeForce GTX 980.

The direct heat exhaust (DHE) cooler does its job and exhausts thermal energy from the case. The hottest part of the card’s shroud is above the voltage converters, which can hit 90 degrees Celsius or more inside of an enclosure. There’s really no thermal impact on the motherboard, though. It heats up mostly due to its own voltage converters and platform controller hub.

On Gigabyte's GTX 970 WindForce OC, the voltage converters are in the same spot (on the left) as the reference card. Consequently, the hottest spot is in the same place. Gigabyte’s board is a bit shorter than that of the reference design, too.

It’s plain to see that there’s a different cooling solution at work here. The motherboard gets its share of waste heat due to the graphics card’s axial fans, positioned above the cooler's fins. These temperatures don't endanger the motherboard in any way, though they can affect overclocking of other components in your system, particularly if you use multiple cards in SLI.

The factory-overclocked Gigabyte GTX 980 WindForce OC sports a redesigned board as well. The voltage converters are located to the right of the GPU, where they were located in the past. It’s easy to spot them in the infrared picture. Greater power consumption results in a higher measured temperature.

Overall, Maxwell’s cooling solutions don’t pose any problems in a closed case, which a lot of enthusiasts will be glad to hear (since that means slower-spinning fans and a quieter PC).

Temperatures During Gaming Loop
Open Bench Table
Closed Case
VRM
Gigabyte GTX 970 Windforce OC
61-62 °C
64-66 °C
91 °C
GeForce GTX 980 Reference
80-81 °C (Limit)
80-81 °C (Limit)98 °C
Gigabyte GTX 980 Windforce OC60-61 °C
64-65 °C
93 °C

Noise

Our noise measurements are performed using a calibrated high-end studio microphone positioned perpendicular to the middle of the graphics card in question at a distance of 50 cm. This distance, in conjunction with the microphone's very strong cardioid directionality, represents a compromise between avoiding noise due to fan turbulence and avoiding ambient noise, which can never be completely eliminated. We like to perform our noise measurements at night for this reason.

So how do the GeForce cards perform, acoustically?

Noise Level (Open Bench Table)
Idle
Gaming Loop
Gigabyte GTX 970 Windforce OC
31.8 dB(A)
41.8 dB(A)
GeForce GTX 980 Reference
31.4 dB(A)
43.6 dB(A)
Gigabyte GTX 980 Windforce OC30.4 dB(A)
40.2 dB(A)

The bottom line is positive when it comes to noise as well. No matter if you’re looking at Gigabyte’s card or the reference model, Maxwell can be cooled safely and quietly without resorting to liquid cooling during gaming.

16. Verdict

If you've skipped to the last page to see our conclusion, I'm not going to waste your time: both the $330 GeForce GTX 970 and $550 GeForce GTX 980 represent tremendous, earth shattering value from a price/performance perspective. But don't take my word for it, have a look at the following charts that represent the average performance we recorded from these cards:

The GeForce GTX 980 not only sets a high-water mark for single GPU performance, but at $550 it does the trick for $50 less than the GeForce GTX 780 Ti costs at e-tail. That's a mere $70 more than the Radeon R9 290X, by the way. There are two important implications with this product: first, that a dual-GPU setup is no longer the only viable path to playability on 4K monitors. Second, the fastest single-GPU graphics card can now be purchased under $600. That's significantly cheaper than the $1050 GeForce Titan Black, the previous title holder.

As for the GeForce GTX 970, it may not bring performance to a new level, but its price definitely makes high-end performance more accessible. At $330 it delivers frame rates comparable to the Radeon R9 290X, a card that currently sells for an average price of about $170 more. We did not expect to see a product this potent at a price tier significantly below the $400 Radeon R9 290, which has suddenly become less appealing.

Nvidia, you had us at the price/performance ratio. But you didn't stop there: the power draw is very low when you consider the performance these cards deliver in return. Anti-aliasing features such as Dynamic Super Resolution (DSR) and Multi-Frame Sampled Anti-Aliasing (MFAA) promise to increase visual fidelity and increase the frame rate, respectively, in a manner completely transparent to the games we play. The point is that we can enjoy these features without relying on developers to implement them on a per-game basis, and that is a good thing. In addition, the technologies under the company's VR Direct initiative speak directly to the virtual reality enthusiast in us, and we can't wait to try them out. As for Maxwell's ability to handle Voxel Global Illumination (VXGI), we think it's a valuable step toward photo realism in games. On a final note, it goes without saying that DirectX 12 compatibility is a desirable feature from a long-term perspective.

Of course, nothing is perfect. The GeForce GTX 980 might be the fastest graphics card on the planet with a solitary GPU onboard, but you're still going to have to lower detail settings most of the time in order to make 3640x2160 a playable resolution. In addition, our expert Igor Wallossek has some insights into the GeForce GTX 970 and 980's power draw as he found micro-spikes and load scenarios that draw significantly more than the TDP these cards are rated for. As for features, we couldn't test MFAA as it isn't yet implemented in the GeForce driver. At this point we're taking Nvidia's word that this new anti-aliasing method will be as effective, efficient, and backward compatible as the company suggests. We weren't even able to test DSR to our satisfaction thanks to the late arrival of our sample cards. Finally, as promising as VXGI may be, we're extremely skeptical that this generation of hardware is powerful enough to handle meaningful real-time voxel cone tracing in a modern game with complex geometry.

But the beauty of the GeForce GTX 980 and 970 is that, regardless of any minor annoyances, they have got it where it counts: higher frame rates for significantly less cash than we're used to paying. That's a formula that speaks directly to our gamer hearts, and inspires us to award both of Nvidia's new graphics cards with the highest appropriate honors. For the GeForce GTX 980 we bestow Tom's Hardware's Elite award, stealing the prized mantle-piece (pardon the pun) from the Radeon R9 290X. As for the GeForce GTX 970, Tom's Hardware's Smart buy award is a perfect fit for this card, which provides strong performance for an incredible value. It may have taken a while to get here, but the second-generation Maxwell GPU was worth the wait.