RSS Feed

HCW Tech Blog

For the latest info on computer hardware, tech, news, video games, software tips, and Linux, check out our new improved front page: HCW Tech Blog

Reviewed by: Bryan Pizzutti [11.12.01]
Manufactured by: Hercules
MSRP: $119.99


3D Mark 2000

Well, According to this, the GeForce2MX seems to dominate at lower resolutions. However, as resolutions scale up at 16 bits, the Kyro2 begins to assert some authority. When it comes to 32 bit color, the Kyro is unchallenged leader, getting essentially the same performance at 32 bit color as it does in 16 bit mode. This is due to the fact that the Kyro renders everything internally at 32 bit color to begin with.... it will translate down to 16 bit color if that is the requested display mode. The downside of this is very little performance can be gained by switching to 16 bit color.  The upside to this is that, unlike the GeForce 2 MX, there will never be a need to use 16 bit mode with the Prohpet 4500.

Note that the GeForce2 MX does NOT support Environmental bump-mapping, while the Kyro does. However texture compression isn't as well supported on the Kyro2, so it could not complete the 64 MB texture rendering test.

3D Mark 2001

Keep in mind, this benchmark is heavily leveraged towards the very latest generation of high-end cards with hardware T&L engines, such as the GeForce 3 and Radeon 8500.  Most of the tests have a ton of polygons, and the benchmarks are designed so that very few surfaces are hidden very often.

3DMark 2001 actually showed that the Kyro2 supported pixel shaders (probably in software through the drivers) while the GeForce2MX did not. As you can see from the scores, the Kyro does a decent job, considering it doesn't have a hardware T&L engine, but STMicro had better develop one soon if they want to support games in the future.  The Kyro2 showed SERIOUS Z-Buffer (Depth-placement) problems in the high-detail "Lobby" demo....objects that were supposed to be hidden by other objects were visible. This may have been just a DirectX8 issue, since it didn't show itself in any other benchmark


Now that we've shown a benchmark that is biased towards the use of hardware T&L, which the MX features, for the sake of parity we'll show a benchmark that makes use of Kyro 2's Hidden Surface Removal.  VillageMark very simply creates a scene with a LOT of hidden surfaces and tries to get the fastest frame rate possible. It also supports hardware T&L, even though this benchmark was developed specifically for the Kyro chips, which don't have any such engine. So the MX was given as much of a leg-up as possible, but it simply gets overwhelmed by the number of polygons it has to process that never get shown. It even runs in only 16 bit color, so the MX isn't burdened with extra 32 bit calculations (Normally, the MX has an advantage at 16 bit) Since the Kyro2 doesn't bother rendering them, it sails through this benchmark with an easy win.

So what can be concluded from these synthetic benchmarks? Well, now we know some specific strengths and weaknesses of each card.....the Kyro2 is very strong when a lot of rendering is going on that never gets seen. However when games start taking advantage of DirectX 8's feature set, which includes a LOT of T&L work, then the MX will most likely handle it better. Of course, by then the next MX and Kyro chips will have been released. :)

Here's a comparison of Fillrates reported by both 3DMark2001 and Serious Sam:

Even though the Kyro is showing up as the winner in the real-world benchmark, the chip has an 8-layer multitexture setup, so it should be able to apply two or more textures per clock cycle. Wrong. It only applies 1 texture per clock, so the numbers in Sam are accurate (One PASS isn't the same as one CLOCK). But why oh why didn't 3DMark 2001 realize that? I haven't found out yet. The MX has two rendering pipelines, though, and can produce a total of two per clock, so 3DMark2001's THEORETICAL numbers are accurate, provided that it's only a dual-texturing environment.. But Serious Sam says you'll never see more than half of that theoretical performance in the real world...

These benchmarks prove theory, and show that each card can do what it's supposed to.  However the real story lies in the the REAL video games... Let's get on to those...

Next Page: (4)