AquaMark3 has been designed to benchmark current and next-generation PCs with a real world game
engine, the krassTM Engine. Having been used in Massive Development’s recent games AquaNox and
AquaNox 2: Revelation as well as in Spellforce by Phenomic Game Development, the rendering
component of the krassTM Engine delivers high-performance graphic effects with superior quality.
AquaMark3 utilizes recent hardware features of the DirectX 9 API, such as PixelShader 2.0, while
staying fully backward compatible to DirectX 8 and 7 graphics hardware. Additionally users can
configure the graphics features in fine detail in order to scale the performance appropriately for their
target system.

Unlike synthetic benchmarks AquaMark3 allows benchmarking in a real-world scenario with an engine
and art assets representing the complexity of current state-of-the-art games. Its art assets are based
on those used in AquaNox 2: Revelation, but have been advanced to be slightly more complex
regarding their average polygon and vertex count so as to represent games that will be released within
the next 6-12 months.

AquaMark3 benchmark results do not only represent the performance of the tested system with
games based on the krassTM Engine, the measured performance is generally transferable to other
state-of-the-art game engines and rendering middleware as well. The vast majority of engines being
used in recent game production cycles work in a similar and widely-accepted way of systems
engineering.

AquaMark3 TRISCORETM (Score Measurement)
The score measurement runs the benchmark with default settings (high quality) and measures the
AquaMark3 score. The AquaMark3 score comprises three independent values (KTRISCORETM): the
overall system performance, the performance of the graphics subsystem, and the CPU performance.
Of those three values the overall system score is the most important measurement, as it is solely
based on the number of frames rendered during a certain amount of time. The graphics and CPU
scores are less precise as AquaMark3 is not a synthetic benchmark measuring raw processing power.
It is necessary to take into account that the graphics score does not measure the performance of the
graphics card only.
The complete graphics score also comprises CPU and system memory, as the
graphic card driver and the operating system also contribute to the result. Graphic card drivers, though
running on the CPU, are seen as a part of setup of the graphical processing pipeline and therefore
don’t contribute to the CPU score. Nevertheless graphics and CPU performance are useful indicators
to see whether the interaction of graphics card and CPU match the needs of action-based computer
games.

One of the serious problems in handling huge outdoor scenes in real-time 3D games is the detection
of the potential visibility set of objects to be rendered (referred to as PVS). To extract the PVS it is
necessary to test if an object is being occluded by other objects. For indoor scenarios with a welldefined
geometry and short view ranges, this problem has been partly solved by algorithms like BSP
rendering or portal techniques. Unfortunately, these solutions don’t apply well to outdoor scenarios
(especially dynamic ones). The krassTM engine solves this problem for the occlusion between objects
and terrain, which is the most important portion of the problem. Think of an entire city consisting of
hundreds of individual buildings for instance: If that city is occluded by a hill, no object needs to be
rendered. AquaMark3 solves this problem by implementing a software rasterization process of the
terrain geometry into a depth buffer. Before an object is rendered a conservative estimation of its
bounding shape is transformed and tested for full occlusion. Because this task is purely CPU driven
and no data is shared with relevant simulation data, it fits well to a multi-threading concept.

Before the rendering starts, every object is checked for its visibility with respect to the terrain. This
query decreases the total amount of objects to be rendered significantly (~20% - 90%) and so leads to
less load on the graphics hardware.

In games particles are a common technique for volumetric effects with high visual quality yet
efficient algorithmic computation. This chapter demonstrates a high number of particles being used for simulating dust and smoke. Other applications include explosions and weapon effects. In AquaMark3 each particle is approximated by a mass point with one degree of freedom for its rotation.

This chapter deals with the use of partial environment mapping on objects with highly specular
materials. The environment map is modulated with a transparency map to determine the relationship
of the incoming light contribution with the diffuse material computation.

To enrich the lifeless atmosphere of the sea bed the terrain system can be combined with a highly
efficient animated plant rendering system that can generate a large number of similar, but colored and
dynamically lit plants. Systems of this type become more and more popular, because they can be
seen as a Level of Detail concept for simulated artifacts e.g. the plants in our case. In AquaMark3 this
system is designed to fill the gap between rigid and animated objects. Each plant is animated
individually and the animation supports the recognition of huge waves rolling over the sea floor.
AquaMark3 nearly puts the total computational load onto the graphics hardware by utilizing a concept
which we call “Hardware Driven Geometry Instancing”. Therefore the only task of the CPU is to deliver a plant’s size, position, rotation and color. This concept enables processing a huge amount of geometry on the graphics hardware. In this chapter, the number of polygons per second goes up to 25 million depending on the graphics hardware and the CPU, and so reaches the highest polygon throughput in the entire benchmark.

The AquaMark3 terrain system enables scenes with large viewing distances while still having complex surface materials with several detail and light textures. The AquaMark3 terrain database is represented by a discrete height map defined on a rectangular grid that covers the entire game world.

The height data taken from satellite surface scans has been adapted manually to fulfil our requirements: it can be used to build the level structure and its tile can be repeated unlimited times in each direction without cracks.

The objects in this scene show a broad range of different material and lighting effects used in
modern game engines. AM3 tests two things: It applies many vertex and pixel shader changes straining the graphics hardware and its many texture lookups (determining the lighting contribution and material reflectivity) also strain the texturing and rasterization components.

AquaMark3 supports a 3D volumetric fog system providing a 2D texture for every object in the
scene to be rendered. This texture should also be software-based to deal with visibility queries coming from ray casting systems which determine if an object can be “seen” through the fog from arbitrary viewpoints.

This section demonstrates different material shaders within single objects, different shaders
representing different materials individually. Materials are determined by the following parameters:
• Number of simultaneous caustic maps
• Mask application of environmental maps
• Number of simultaneous light sources
• Number of simultaneous detail maps
• Fog map

This scene demonstrates the application of the particle system for large explosions. For each
particle a textured rectangle is drawn that gets blended over the previous frame buffer contents. The test stresses the graphics hardware by the high overdraw of textured areas as many particles overlap each other. To ensure correct screen results, the entire particle set is sorted in back to front order for each frame with respect to the viewer’s position.

As a developer of interactive high-end 3D gaming applications and benchmarks we obviously have to collect in-depth Know-How about a specific hardware from it's manufacturer.

This type of cooperation is common practice for all major developers. It serves the sole purpose to utilise the latest technologies in the correct way and thus giving the customer the best gaming and performance experience possible.

You may decide to support a specific hardware platform or a specific hardware feature by a game application. However this not an option for any serious benchmarking application.

While developing AquaMark3 we take precautions in a way so that all IHV's (Independent Hardware Vendors) with a vital interest in AquaMark3 are treated equally. So there is no reason for those vendors to conduct any cheating attempts on AquaMark3.

Once a benchmark is released there are obviously multiple possibilities to cheat its results (both for any IHV and any user). We took precautions to prevent and detect cheating attempts in this post release phase and during the lifetime of AquaMark3. Unfortunately we can not prevent this in all cases by the application itself.

The most powerful anti cheating tool will be the AquaMark Result Comparator (ARC), the online database and the AquaMark3 forum. We will immediately inform all customers about our evaluation of the latest cheating attempts (if they should happen).

As a developer of high end 3D gaming applications we will of course be very specific and accurate when we have to judge if a specific optimization is of general interest or if it only targets AquaMark3 in a questionable way.

We are always open to discuss our policy with interested users in this forum.

Alexander Jorias / Managing Director Massive Development
Ingo Frick / Technical Director Massive Development