Benchmarks

Synthetic Benchmarks

    We will again have no 3D Mark 2001SE in this review, but rather look at the very bandwidth intensive Villagemark as well as the game based DroneZmarK.   First lets look at Villagemark, we have the 16GB/s bandwidth of the Parhelia, the extreme memory efficiency of the Kyro II, the 8.2GB/s of the MSI GeForce 4 MX, and lastly the 11.7GB/s of bandwidth of the MSI card when its overclocked.  So lets see how the MSI card does in this Power VR based benchmark.

    With the obvious exception of the Kyro II, with its almost 100% memory efficiency what can we see.  The MSI card can't seem to improve very much, as a 42% increase in memory bandwidth only gives a 13% improvement in average frame rate.   Compared to the 16GB/s of the Parhelia the MX needs at least 18GB/s, and probably more, to get near to the frame rate the Parhelia attains.  It seems that the MX doesn't perform very well in memory bandwidth intensive applications, but as this is only one synthetic test, we cannot conclude anything about the card as of yet.

    Here the results are slightly different, with the Kyro II plummeting from first all the way to last place.  The AA and ansiotropic filtering of the MX is much better than that of the Kyro II, 'only' losing 75% of its performance while the Kyro II loses over 93% of its average frame rate.  The Parhelia shows the efficiency of its AA technique only losing about half of its average frame rate.  Now what kind of 'real world' triangle throughput can the MSI card give?

    As we mentioned in the Parhelia article, Fraps and DroneZmarK do not really like each other, so therefore we will only show the average frame rate and triangle throughput as well as the maximum and minimum frame rates.  Lets see how it performs at 1024 without any antialiasing or ansiotropic filtering.

    We can see that the MSI card performs very well in this test, at stock speeds it is a paltry 5fps behind the Parhelia, and has a much higher minimum frame rate.  When overclocked the MSI card gains an extra 22% in the way of frame rate, and another 20% in the way of raw triangle output, out of the extra 45% improvement in both GPU's and Memory speed.

    When AA and ansiotropic filtering are enabled, we see the MSI card looses a fair bit in the way of performance, about 66%+ of it to be exact.   Overclocking this card yields better improvement than with no AA, as the improvement is almost perfectly in line with the increase in core/memory speed, showing a lack of CPU dependence in this test.  However synthetic benchmarks are not what we use everyday, rather lets look at some real games that are played.

Previous Page - Benchmark System

Next Page - Jedi Knight II