Nvidia GeForce GTX 1060 (Desktop) Review

But back to the GTX 1060: Nvidia deserves respect for the dimensions and the power supply. While a standard GTX 980 still required two 6-pin connectors, the new model – similar to the GTX 960 and RX 480 –  needs only one 6-pin connector. The PCB is also a few centimeters shorter compared to the GTX 980, while the weight has dropped from 1036 to 838 grams (~2.3 to ~1.8 lb; -19 %). Nvidia specifies the TDP with 120 watts (GTX 1070: 150 watts, GTX 1080: 180 watts). The die size of the GTX 1060 is specified at 200 mm² with 1.5 MB L2-cache. You also get 4.4 billion transistors, 48 ROPs and 80 Texture Units on the chip. For a better comparison: The GTX 960 with a die size of 227 mm² has only 1 MB cache, 2.94 billion transistors, 32 ROPs and 64 TMUs. More information about the Pascal architecture is available in our dedicated article.

Nvidia will have to live with some criticism for the missing SLI bridge. Even though the author is not fond of SLI systems (micro stutters, driver dependency, etc.), it seems to be an unnecessary limitation to waive the SLI. Desktop users are used to a lot of freedom when they configure their own systems. In our opinion, there is no criticism for the connections. The GTX 1060 has three DisplayPorts (1.3/1.4 ready), one HDMI 2.0b ports as well as a dual-link DVI output. Except for DVI, all ports can drive UHD monitors at 60 Hz. The power adapter should provide at least 400 watts.

Nvidia sent us the ForceWare 368.64 for this review, which was very stable. If there were graphics errors or other bugs, they were usually caused by Windows or the applications (e. g. scaling and mouse problems when we did not use the native resolution of our 4K monitor).

Even though synthetic benchmarks are not always representative of actual gaming loads, you can still get a good idea of the performance levels and the differences between GPUs. We start with the Fire Strike Test of 3DMark 13. With a GPU score of 12984 (normal) and 2893 points (ultra), respectively, the GTX 1060 is about 8% faster than the Radeon RX 480, which manages 12186 and 2678 points. The advantage over the old GTX 960 is between 64 and 118%. The GTX 980 – both the desktop and the notebook version – are roughly on par.

The new chip does not stand a chance against its Pascal siblings. While the GTX 1070 is around 40% faster, the GTX 1080 is more than 60% ahead. The situation is similar in 3DMark 11, although the GTX 1060 falls slightly behind the AMD Radeon RX 480 in this case.

If you only play in Full HD (1920x1080 pixels), the GTX 1060 delivers convincing results. We tested 26 games, and 23 of them ran smoothly with the highest details and activated anti-aliasing. Only  "Anno 2205", "Assassin's Creed Syndicate", and "XCOM 2", which are not perfectly optimized, dropped below 40 fps in our Ultra preset. The GTX 1060 is ahead in some games, while the RX 480 in others. Overall, we can see a slight tendency towards the GTX 1060. This supports the predictions ahead of the launch.

Fans of the 4K resolution should get a more powerful GPU. Hardly any modern game ran smoothly with high details. Only "Dirt Rally", "Mad Max", "FIFA 16", "Rainbow Six Siege" and "Overwatch", which are not the most gorgeous titles, are playable in our opinion. 4K would require an SLI system with two high-end GPUs, a GTX 1070 or even better, a GTX 1080. AMD's Radeon RX 480 also has problems with 3840x2160 pixels – even with more VRAM and a wider memory interface.

The 192-bit interface of the GTX 1060 does not seem to be a big problem, which is also a result of the improved compression. At least we could  not notice any problems with our extreme settings. The performance is more limited by the number of shaders (1920 @GTX 1070, 2560 @GTX 1080).

One decisive factor over the next few years will be the DirectX 12 performance. There are some promising alternatives to reduce the CPU overhead with Mantle and Vulkan, but so far DirectX has always turned out to be the mainstream API. Looking at the previous GPU generation, AMD benefitted much more from DirectX 12 (Ashes of Singularity, for example). Currently, there are not enough suitable applications for a final verdict.

Our tests show two results (with reservations). 1) Pascal works slightly better with DirectX 12 than Maxwell. 2) AMD is still ahead. While the DirectX 12 mode in "Rise of the Tomb Raider" reduces the frame rate on all GPUs, the Radeon RX 480 manages a few percent more frames in the 2016 version of "Hitman". Once again, the frame rate of the GTX 1060  drops slightly. We will have to wait a couple of months for a comprehensive verdict. Whether the new API can justify its premature praise and repeat the success of DirectX 11 will also depend on time and the leisure of game developers.

There is no clear winner in the section compute performance, which used to be the strength of AMD chips. Let us have a look at LuxMark 2.0 based on OpenCL, for example. With 1773 vs. 1662 samples/s, the GTX 1060 is 7% faster than the Radeon RX 480 in the Room scene. However, AMD has the edge in the Sala scene, where the Polaris chip is 34% ahead (3468 vs. 2584 samples/s). For comparison: The GTX 1080 is between 30 and 60% faster than the GTX 1060. We also have a quick look at ComputeMark. Expect for one test, the DirectCompute application prefers the Radeon RX 480, which is again clearly beaten by the GTX 1080.

CAD applications are not an important topic for gaming GPUs. There are special and very expensive professional graphics cards such as Nvidia's Quadro series or AMD's FirePro series with optimized drivers for this purpose. SPECviewperf 12 shows the GTX 1060 on par with the GTX 980. A GTX 1080 is almost 50% faster, a GTX 960 40% slower. The Maxwell Quadro GPUs in the table are difficult to compare due to very different results in the individual test sequences.

There is not a big difference between the GPUs when we look at the temperature and noise development. Up to 82 °C (~180 °F) in the stress test and 32 to 34 °C (~90 to 94 °F) while idling are neither very warm nor very cool. The GTX 1060 does reach up to 98 °C (~208 °F) at the bottom according to our thermal camera Optris PI 640, but unlike the GTX 980 (up to 77 °C/~171 °F), the GTX 1060 does not have a cooling plate at this spot. The PC was also optimized for a quiet operation. The higher surface temperature of the Pascal chip can also be explained with the different clock behavior. While the core of the GTX 980 dropped from the default 1126 MHz to 937 MHz (=throttling) under maximum load, the GTX 1060 always utilized the Turbo (~1582 MHz).

We also use FurMark to determine the maximum noise level. The GTX 1060 produced 38 dB(A) from a distance of 50 cm (~20 in) – a decent result (case closed), while we measured 40 dB(A) for the GTX 980. Note: All measurements only apply for the Founders Edition. Custom models from other partners (MSI, Gigabyte, Zotac, etc.) will be equipped with other coolers, which can control temperature and noise development better.