3DMark 2013 Graphics Cards Performance Review: AMD and Nvidia Based Products in the New Benchmark

Today we are going to talk about the new graphics suite for video cards testing. Let’s see how fast Nvidia’s and AMD’s entire line-ups of products will perform in it.

by Sergey Lepilov
02/26/2013 | 11:30 PM

For over 13 years the Finland-headquartered Futuremark Corporation has been developing and releasing software for benchmarking graphics cards and showcasing new-generation game engines (besides working on some other popular test suites like the well-known PCMark). Although each version of 3DMark is always met with some criticism, there is no other GPU benchmark that could rival its popularity. The overwhelming majority of print and electronic media use 3DMark to test graphics cards of the same or different generations. The ease of use and broad availability combined with support for all modern graphics technologies are the trademark of Futuremark's product. Unveiled on the 6th of February 2013, the latest 3DMark – it’s called just like that without any numbers or anything in its name – carries on the glorious traditions. Let’s see what it’s like and how it runs on modern graphics cards with GPUs from both AMD and Nvidia.

3DMark (2013): What’s New?


The new 3DMark is available in three versions. The free Basic Edition can be used to run three graphics tests (Ice Storm, Cloud Gate and Fire Strike) and publish results online using a free user account. The Advanced Edition costs $25 and offers the same tests but you can run each of them individually. It also unlocks the Extreme mode for the Fire Strike scene, auto-saves results and allows to enable several benchmarking cycles. And finally, the Professional Edition opens up all setup options for $995. It can also output benchmarking results in XML format. In this review we will describe the Professional Edition to you.

So, 3DMark welcomes its user with a brief description of its features and supported platforms:

Here you can launch all the available tests and demo clips.

The Tests screen allows choosing any of the three scenes, read its brief description and run each of them individually.

The Custom screen offers detailed settings for each test scene. You can enable or disable individual subtests, select resolution and other settings.

The first test scene is called Ice Storm. Not a heavy load by today’s standards, it is designed for benchmarking smartphones and other Android-based gadgets or low-performance desktop PCs. It consists of two brief graphics subtests and one physics effects subtest:

Graphics test 1

Graphics test 2

Physics test

Ice Storm requires 128 megabytes of graphics memory to run. Frankly speaking, its visuals are rather primitive, so even weakest graphics cards deliver a high frame rate in it. But again, Ice Storm is meant for smartphones and other mobile gadgets, so it is just difficult enough for such devices. It runs on a Direct 11 engine limited to DirectX 9 capabilities.

The second test is called Cloud Gate and requires 256 megabytes of graphics memory. It is a more resource-consuming thing with lots of various settings:

Consisting of two graphics and one physics subtest too, Cloud Gate is far more beautiful than Ice Storm. And the beauty means higher graphics load.

Graphics test 1

Graphics test 2

Physics test

Running on a DirectX 11 engine, Cloud Gate is limited to DirectX 10 features and serves to benchmark basic desktop PCs and notebooks.

It is Fire Strike which is the most sophisticated scene in the new 3DMark. It requires at least 1 gigabyte of graphics memory and has a huge amount of settings:

Making full use of all DirectX 11 capabilities and defaulting to 1920x1080 (as opposed to the first two scenes which run at 1280x720 pixels), it consists of four subtests: two graphics, one physics and one combined subtest.

Graphics test 1

Graphics test 2

Physics test

Combined test

The combined one is especially beautiful, showing a brief fight of two alien creatures. Fire Strike is designed for high-performance graphics subsystems including multi-GPU ones (CrossFireX and SLI). However, if you find this not enough for your system, you can run Fire Strike at the Extreme settings (2560x1440, at least 1.5 GB of graphics memory) which can get even a GeForce GTX 690 to its knees.

After each test scene is complete, 3DMark issues detailed results on each scene/subtest, illustrating them with fps graphs.

It’s handy for comparing different computers. You can publish your results on the official Futuremark website and compare them with other users'.

The Professional Edition of 3DMark has a number of additional settings related to image quality and XML exporting:

And finally, the last window Help is where you can check out your program version and specify some extra settings (disable sounds and choose one of the four available interface languages, for example).

We won’t delve into details about how each score is calculated because it’s unimportant. Let’s just get right to our testing and see what modern graphics cards can do in the new 3DMark.

Testbed Configuration and Testing Methodology

All participating graphics cards were tested in the new 3DMark 2013 in a system with the following configuration:

So, we have six graphics accelerators on AMD GPUs and six on Nvidia GPUs. The “red” camp is represented by products from three manufacturers: HIS, Sapphire and AMD. Here they are:





The honor of the “green” camp will be defended by six graphics cards from another three makers: Nvidia, Asus and Gigabyte:





As you may have already noticed, ten graphics cards out of twelve have increased default frequencies. However, we decided not to lower them to the nominal reference values, because the performance difference will be insignificant, and besides, most of the currently selling graphics cards are exactly the ones with slightly increased speeds anyway. So, we will test all the products “as is”.

In order to lower the dependence of the graphics cards performance on the overall platform speed, I overclocked our 32 nm six-core CPU with the multiplier set at 37x, BCLK frequency set at 125 MHz and “Load-Line Calibration” enabled to 4.625 GHz. The processor Vcore was increased to 1.49 V in the mainboard BIOS:

Hyper-Threading technology was enabled. 16 GB of system DDR3 memory worked at 2 GHz frequency with 9-11-10-28 timings and 1.65V voltage.

The test session started on February 7, 2013. All tests were performed in Microsoft Windows 7 Ultimate x64 SP1 with all critical updates as of that date and the following drivers:

All tests were run with standard settings. We did not adjust the benchmark launch settings in any way. Let’s check out the obtained results.


Ice Storm

As expected, Ice Storm cannot reveal the full potential of most of the graphics cards we test today since their performance is limited by the CPU and platform. Even though we use a six-core CPU, the top-end cards deliver identical performance whereas the dual-processor GeForce GTX 690 is but a little faster than the single-processor Radeon HD 7970.

The scores differ more in the graphics subtest but, as we’ll see later, do not reflect the actual picture of performance well enough.

Anyway we can note that while the graphics cards don’t differ much in the top price segment, the Nvidia-based products are superior in the low-end segment.

The Ice Storm physics subtest uses one CPU core, leaving the GPU idle. Here are the results:

Our testbed runs at the same speed irrespective of what AMD-based card we install. The Nvidia-based cards make the output more variable. The benchmark developers have actually warned that this might be the case since this load is too low for modern desktop PCs.

Cloud Gate

The Cloud Gate scene is more complex than Ice Storm, yet not complex enough to reveal differences between modern graphics cards. The graphics subtest provides some interesting numbers, though:

The GeForce GTX 690 stays in the lead. It will remain the world’s fastest graphics card until the GeForce Titan or dual-processor cards from AMD come out. Next goes the Radeon HD 7970 which is closely followed by the GeForce GTX 680. The GeForce GTX 670 competes with the Radeon HD 7950. In the medium price range we see the Radeon HD 7870 racing against the GeForce GTX 660 Ti and the Radeon HD 7850 against the GeForce GTX 660. All of them are factory-overclocked, of course. The pair of entry-level AMD-based cards is much slower than the GeForce GTX 650 Ti, however, so AMD can only make them attractive by means of lower prices.

There’s nothing to compare in the physics test which now uses all of the six cores of our CPU:

We can only estimate the benchmark’s accuracy here.

Fire Strike

Fire Strike is quite a different story. This scene uses tessellation, shadow maps, post-processing filters and lots of other effects. As a result, the graphics subsystem load is so high that we can easily compare graphics cards between each other, especially in terms of graphics scores:

The GeForce GTX 690 is 50% faster than the Radeon HD 7970 here. The latter, in its turn, is better than the GeForce GTX 680 but then the picture changes. The GeForce GTX 670 is not slower but faster than the Radeon HD 7950. The GeForce GTX 660 Ti is equal to the Radeon HD 7870, and the GeForce GTX 660 to the Radeon HD 7850. Judging by the 3DMark results, AMD should fill the gap between the Radeon HD 7850 and the HD 7770 with some new product because the GeForce GTX 650 Ti is preferable to the Radeon HD 7770 among the most affordable gaming graphics cards.

The physics subtest doesn’t need our commenting:

AMD-based products look preferable in the combined subtest that loads both the CPU and the GPU.

Excluding the Radeon HD 7770 again.

Fire Strike Extreme

The Extreme settings of the Fire Strike scene show the same standings, except for the downright slow Radeon HD 7770 and HD 7750.

The graphics subtests agree with the overall scores:

The physics subtest doesn’t have anything new to tell us.

The GeForce GTX 690 is ahead in the combined subtest but its advantage over the Radeon HD 7970 is hardly impressive.

The GeForce GTX 680 and GTX 670 are third and fifth, respectively, separated by the Radeon HD 7950. The next two pairs of AMD- and Nvidia-based cards are comparable to each other again whereas the Radeon HD 7850 and GeForce GTX 660 have absolutely identical scores.

Here is a table with the full test results:



The new 3DMark is just as good as Futuremark’s earlier products. Fully-functional support of DirectX 11 and modern graphics technologies, flexible settings and the opportunity to publish your results online or export them into XML format upon completion make it a handy and even indispensable tool for benchmarking any graphics card. Moreover, considering the rising popularity of smartphones and other mobile devices, 3DMark is now capable of testing their graphics capabilities, too. The Ice Storm scene is clearly not meant for modern discrete graphics cards but may be used for CPU benchmarking. The Cloud Gate scene is better for GPU benchmarking, yet it is best suited for entry-level and mainstream products. The most resource-consuming and beautiful scene is Fire Strike, especially with the Extreme settings. It is the only 3DMark test we will be able to use to benchmark future generations of GPUs and graphics cards.

As for today’s hardware, we haven't seen anything surprising in the new 3DMark. The dual-processor GeForce GTX 690 is easily ahead in every test since we don’t have a GeForce GTX Titan or a dual-processor Radeon HD 7990 in this test session. The Radeon HD 7970 GHz Edition is superior among single-GPU products, beating the GeForce GTX 680 in every graphics and combination subtest. This pair is followed by the GeForce GTX 670 and Radeon HD 7950 which are comparable to each other, never breaking apart by more than 1 or 2%. Identical performance can also be observed in the pairs of Radeon HD 7870 vs. GeForce GTX 660 Ti and Radeon HD 7850 vs. GeForce GTX 660, although the AMD products are preferable in terms of pricing. Unfortunately, junior AMD-based solutions are too weak for 3DMark and fail to compete against the GeForce GTX 650 Ti. We also didn’t include any slower graphics cards into the today’s gaming test session, which proved to be the right decision (judging by the results of the HD 7750).

We can also add that the physics effects tests do not depend much, if at all, on the graphics subsystem and are thus suitable for benchmarking CPUs and platforms in general.