A
benchmarking program developed by
FutureMark (formerly
MadOnion) to allow
PC users to gauge their systems' performance. It is intended as a replacement for
Futuremark's widely-adopted
3DMark2001 benchmark package. It is available free to
download, with a 'Pro' version ($39.95 USD) offering some additional
functionality that is not required to run the tests. It is primarily aimed at
gamers, and tests various features of modern
GPUs as well as the general performance of the system's
CPU and
RAM. It also includes tests focussing on 3D sound. Some of the tests make use of features of
DirectX 9.0 and the
Pixel Shader 2.0 instruction set.
The tests include a sequence depicting WWII aerial combat (repeated several times to test different aspects of the system), a fantasy-themed scene involving a fight between a woman and two trolls (showing off real-time dynamic lighting and bumpmapping), a scene showing off Havok-powered ragdoll physics, a fillrate test, and a Nature test (a test requiring full the DirectX 9.0 specification to be met to be run at all). After all the tests are run, the benchmark totals up all the results to give you a score: one number that should conclusively show whether your machine or the other guy's is The Best.
The test is intended to mimic real world situations as closely as possible. At this time, it does not seem that this claim has been delivered on particularly well: Apart from the fact that most benchmarks (other than a timedemo) can only give an approximation equivalent to an in-engine cutscene (machinima) rather than gameplay, the tests do not seem to use real-world game engines (compare the 2001 edition, which used Remedy's engine technology, identical to that used in the contemporary game Max Payne).
Another intention of the package is that none of the tests are optimised to take into account the strengths or weaknesses of a particular graphics chipset (as this is rightly argued would skew the results in favour of such chips). Unfortunately a side effect of this seems to be that the test comes back with drastically lower scores for NVIDIA GeForce4-family cards compared to ATi Radeon 9700 based cards.
While this might be strictly accurate when talking about the hardware implementation of DirectX 9.0 features on the two cards, it is not reflected by the comparitive performance of modern games on the two chipsets. (An ATi system might get anything up to double the score of an NVIDIA one, but that wouldn't translate to double the framerate in an actual game.) Understandably NVIDIA are not happy with this representation of their products and have made public statements to that effect.
It is possible that as 3D card technology marches on, these teething problems will become less impactful on the scores that the benchmark outputs. But as of this writing, your mileage may vary.
http://futuremark.com/products/3dmark03/
http://tech-report.com/etc/2003q1/3dmark03-story/index.x?pg=1