|
3DMark 2001SE
Unlike when looking at an FX5900, there's no second guessing what the FX5600 is doing in terms of its render setup for the test you are running. It's a straight 4 pipeline GPU, with a single texture mapping unit and shader unit per pipe. 4x1, across the board, at all bit depths. There's some confusion as to floating point precision in the shader unit, with NVIDIA alleged to be using 12-bit precision when the application specifically asks for 16-bit data type precision, in certain applications. We can't really test that without a debug version of any given driver, something that's never going to happen. But at the core of things, FX5600 doesn't change. And it's a comparable architecture to 9600 Pro, making our job even easier.
So with that out of the way, here's the base 3DMark 2001SE graph.
We saw the same thing with the original reference board, the FX5600 Ultra in 400/800 configuration is more than a touch quicker than its main rival, in 3DMark 2001SE at least. The PD graph will confirm, so let's look at the IQ graphs quickly.
Much the same as what we saw before. The big pixel test now.
Outwith the mid-range boards, the 9800 Pro ends up being the quicker at the bitter end of our 3DMark testing. With the two competing boards however, the MSI FX5600 Ultra keeps its nose ahead of the ATI board.
With roughly the same performance drop when initially enabling IQ, the NVIDIA board loses less when the pixel count goes through the roof. NVIDIA doing a better job in the driver? Almost certainly, combined with the substantial amount of memory bandwidth it has over the ATI. Whether that better job in the driver is legitimate or not, we can't really say. A new test.
|
