Futuremark's 3DMark has been seeting the standard in testing PC's on gaming perfromance for more than a decade now. Today the guys from Finnland launched their latest software suite. The new 3DMark is not only capable of testing 3D performance of PC's as well as Android and iOS devices, it also features CPU/GPU temperature and load monitoring. Overall it sounds like a really interesting package.
When it
comes to their latest benchmarking suite the team behind 3DMark from Futuremark
has been thinking a lot about how to improve their software. Nowadays the latest
version of their benchmark ships with lots of monitoring features as well as the
possibility to do cross plattform comparisons. Although today it's only possible
to use the desktop benchmarks, Futuremark pointed out that they are working hard
to finish the versions for Android, iOS and Windows RT.
In order to allow comparing a broad range of systems and devices when it
comes to 3D performance, Futuremark chose to split their benchmark into three
different and individual tests. In older times, like for example with 3DMark 11
or 3DMark Vantage there used to be one single benchmarking suite, at the end of
which you got a final score which used to be an average over all the
subroutines. With the latest version of 3DMark you get three individual routines
which apply different load to a system that is to be tested. The names of these
three benchmarks are "Ice Storm", "Cloud Gate" and "Fire Strike". The first one
will be compatible with all plattforms but "Cloud Gate" and "Fire Strike" only
run, using the dekstop version of Windows.
Ice Storm
Ice Storm is
Futuremarks cross platform benchmark, which is being used to compare
smartphones, tablets, ultrabooks as well as entry level PC's. The test consists
of two parts where as the first part, the graphics test, again is based on two
individual tests. As a last sub routine there is the physics tests. Ice Storm
unter Windows uses the DirectX 11 Engine but it has been reduced to the DirectX
9 feature set. Therefore it's suitable to test Android as well as iOS devices.
On these two platforms the OpenGL ES 2.0 interface is being used.
In the first graphics test 530'000 vertices need to be calculated per frame
which results in 180'000 triangles. In the very same moment 4.7 million pixels
are processed per frame.
In the second graphics test there is a higher pixel load than in the first one.
In total there are 12.6 million pixel that need to be processed per frame. The
reason for this difference can be found with post processing effects that have
been added: Bloom, streaks and motion blur are applied. On the other hand the
number of vertices has been reduced to 75'000 per frame.
Last but not least there is a physics test. The goal of this test is to apply as
much load a possible to the CPU. According to the number of cores/threads the
devices is offering soft and rigid bodies are calculated that collide with
eachother.
Engine Features
Traditional forward rendering using one pass per light.
Scene updating and visibility computations are multithreaded.
Draw calls are issued from a single thread.
Support for skinned and static geometries.
Surface lighting model is basic Blinn Phong.
Supported light types include unshadowed point light and optionally shadow mapped
directional light as well as pre-computed environmental cube.
Support for transparent geometries and particle effects.
16-bit color formats are used in illumination buffers if supported by the hardware.
Cloud Gate
Cloud
Gate is an entirely new benchmark that is perfectly suitable to test Windows
notebooks and home PC's for their graphics performance. Futuremark recommends to
test setups with integrated graphics processing units with this benchmark. Again
Futuremark uses the DirectX 11 Engine but limits it to the DirectX 10 feature
set.
During the first graphics test 3.0 million vertices per frame have to be
processed. Geometry shaders have to take care of 450'000 primitives and 1.1
million triangles need rasterization. Volumetric illumination is disabled and
the scene features particle effects. When it comes to post processing bloom and
depth of field are being used. On average 18 million pixel per frame are
processed.
In the second graphics test 1.8 million vertices per frame are processed and
340'000 primitives need to be take care of by the geometry shaders. There are
690'000 triangles that need rasterization. Furthermore there is simple
volumetric illumination but not particle effect that is calculated.
A closer look at the physics test shows that there are 32 simulated workloads.
Ever world comes with 4 soft bodies, 4 joints and 20 rigid bodies, which all
collide with oneanother. One core/thread has to take care of one simulation.
Physics effects are processed by the CPU, keeping the GPU load to a minimum.
Engine Features
Multithreading
Tessellation
Lighting
Surface illumination
Volumetric illumination
Particle illumination
Particle based distortion
Depth of field
Lens reflections
Bloom
Anti-Aliasing
Smoke simulation
Fire Strike
Fire
Strike is a fully fledged DirectX 11 Benchmark, which has been designed to make
high-performance PCs sweat. During the test many recent DirectX 11 features are
used to stress todays most powerful GPUs. This test has also been split into two
different parts: the two graphics tests and the physics test.
The first graphics test focuses on geometry and illumination effects. Particles
are painted with half resolution and dynamic particles are activated. In total
there are 100 spot lights, which cast shadows and 140 more light points, which
don't cast shadows. On averge there are 3.9 million vertices and 500'000 input
patches have to be invoked. Per frame 5.1 million triangles need to be
rasterized. Furthremore 1.5 million compute shaders are invoked per frameto
enable particle simulations and post processing effects. On average 80 million
pixels have to processed.
The second Fire Strike test is mainly about particle and GPU simulations.
Particles are rendered with full resolution and dynamic lighting effects are
enabled. The GPU has to render two smoke fields, six spot lights are casting
shadows and 65 more don't cast shadows. On average there are 2.6 million
vertices and 240'000 input patches which need processing per frame. In other
words tehre are 5.8 million trinagles per frame and 8.1 million compute shaders
nee to be invoked per fram. In total 170 million pixel are processed.
Also with Fire Strike there is a physics test. In this case there are 32
simulations, whereas one simulation is applied to one core/thread. All physics
effects are processed by the CPU except for soft body vertex data, which is
processed by the GPU.
Article in English
Artikel in Deutsch
Article en français
Articolo in italiano
Re: New 3DMark Tested - 25 GPU, CPU and Memory Scaling