Futuremark Release 3DMark 11

Today, Futuremark released 3DMark 11, the latest in their long-running series of benchmarking software. 3DMark, as the name suggests, tends to be aimed towards benchmarking gaming PCs and provides users with a score that can be used to compare the relative performance of different systems for gaming. Futuremark has kindly provided us with press copy of 3DMark 11 Professional for a few days, so we can go through some of the features of the new software and reveal some of the first 3DMark11 scores. Unlike some previous versions, the 11 in 3DMark11 stands for ‘DirectX 11’ rather than ‘2011’.

Features and Specifications

As stated, this is a DirectX11 benchmark, which means it will probably thrash even current bleeding edge hardware; in the sample tests below, frame rates were pretty low. As with previous versions, there will be be a free Basic Edition (no expiry date or run limits) and an Advanced Edition ($19.95 RRP, with some extra toys), as well as a Professional version (licensed for commercial use, coming in at a whopping $995). As well as a battery of tests and online results submission, 3DMark11 also features a demo mode (based on the ‘Deep Sea’ and ‘High Temple’ tests) with a soundtrack. While the Basic Edition only allows use of the ‘Performance’ preset (1280 x 720 resolution; scores are designated Pnnnn) the Advanced and Professional versions allow use of the Entry (1024 x 600 resolution; Ennnn) and Extreme (1920 x 1080 resolution; Xnnnn) presets, as well as custom settings.

The only really massive difference, to overclockers, in the Professional Edition is the Image Quality: this allows high quality renders of any given point in any test. Futuremark has kindly provided us with a Professional license, which should allow us to pick out image quality differences between cards, operating systems and drivers in the future.

The Tests

There are six tests: four test the graphics capability of the machine, one is a physics test designed to put the CPU through its paces and one tests the CPU and GPU for physics-related capability. Realistic physics is one of the major benefits of modern games, but do require some extra processing power to deal with properly.

The graphics tests all use a lot of post-processing to add various camera lens effects, such as altering the depth of field.

1. Test 1 is heavy on light sources, shadows and volumetric lighting.
2. Test 2 has some lights, and tessellation of many parts of the scene.
3. Test 3 employs tessellation and some volumetric lighting and shadows.
4. Test 4 also features tessellation and volumetric lighting but casts a large number of shadows.

The following YouTube videos are courtesy of Futuremark.

The two physics tests will stress different parts of the PC.

1. The Physics Test deals with rigid-body physics on a large number of objects, at a fixed resolution. Eye-candy is kept to a minimum here.
2. The Combined Test also tests the CPU with rigid-body physics calculations, but puts the GPU to use on volumetric lighting, tessellation, post-processing and soft-body physics-simulation.

As you can see, all of these tests will be particularly heavy on the GPU; only one test is really designed to assess your CPU. However, this is appropriate given the incredible capabilities of modern graphics chips. Although the minimum specifications suggest you need a 1.8 GHz dual-core CPU and 1 GB RAM you’re definitely going to need more horsepower, as our sample results below illustrate.

How Will Your Rig Perform?

Thanks to my colleague Jeremy (who has more computer hardware than your local store) we have some sample scores, on both Intel and AMD platforms.

Note: 3DMark11 apparently works best when run as ‘Administrator’, although the stability issues will probably be worked out in the first patch.

Pre-DirectX 11 Cards

These don’t work. The program will refuse point blank to start: below is what I was greeted with when trying to start 3DMark11 on my Lenovo ThinkPad W510 (nVidia Quadro FX880M).

AMD Hex-Core Build

• CPU: AMD Phenom II X6 ‘Thuban’ 1100T  (overclocked)
• Motherboard: Asus Crosshair IV Formula
• RAM: 4 GB of G.Skill Flare DDR3-2000
• GPU: Asus Matrix 5870 Platinum

Intel Quad-Core Build

• CPU: Intel Core i7-870 (overclocked)
• Motherboard: EVGA P55 FTW
• RAM: 4GB of G.Skill DDR3-2400
• GPU: AMD HD6870

Conclusions

These sorts of benchmarks are intended to be valid for some time to come, so it’s no surprise that these two very up-to-date machines struggle with this initial version of 3DMark11. It will probably be one or two generations of cards (GT780 anyone?) before this benchmark becomes even close to trivial for hardware. The GPU influence on these scores is massive: a mere 25 MHz on the core and RAM (for a 5870) nets you ca. 300 3DMarks, and 125 MHz on the core and RAM on a 6870 scores an extra 500+ 3DMarks. In stark contrast, 5 to 10% extra CPU speed gets you only 10 – 30 extra 3DMarks. These are just preliminary runs, but I’m sure there will be a lot more trends emerging in the near future.

According to the published formulae, in order to hit 10,000 ‘P’ 3DMarks, you’ll need 40 FPS across the board, or 45 FPS in all graphics tests plus 35 FPS in the physics tests.

Clearly, our benching team will have to get their GTX580s and the like under some liquid nitrogen to see some boost in 3DMarks!

If you’ve downloaded and run 3DMark11 (and we hope you do), please post your score below (‘Leave a Comment’) along with your hardware specification.

– David