Review: MSI GeForce FX5900 Ultra

Here's the base graph to digest before we get on to the good stuff. Remember, it's a 256MB board as well as being clocked higher than the regular FX5900.

600 points from the increased core clock and extra framebuffer memory. If the extra memory has anything to do with the score increase at those settings, I'll be surprised. The test simply doesn't generate enough data for > 128MB to matter. Let's turn on some IQ instead.

Less of a drop when applying IQ for the FX boards, with the Ultra 256MB MSI taking the lead due to core clock. It's worth noting that NVIDIA optimises wrongly for this test in the 44.03 drivers, giving a higher score than should be seen. However a correct score wouldn't and shouldn't be too far away.

An increase in pixel count here, possibly giving the extra memory something to do.

While the regular board drops off, the VTD256 keeps its score nice and high. Clock biased or memory size biased? A bit of both. Here's the PD graph.

Given that we saw no such difference between 9800 versions that we are seeing with the big framebuffer FX5900, what can we conclude? Deducting a performance increase from the differing clocks, it's evident that at least some of the advantage the Ultra has in the PD graph is down to the memory size. But how much of that is exacerbated by the driver?

On to another test to see if we see the same thing again.