LOSTCIRCUITS

Benchmark Results

Before we address the question of whether AGP 8X is beneficial for performance, there is another question that we need to ask and that question is whether the benchmarks that we are using are actually showing any differences in AGP transfer rates. We know that there is a benefit of AGP 4X over AGP 2X, at least, we think that we know but what is it that the benchmarks are telling us?

3DMark2001SE

We all know it, we all love it and in the course of watching it a few hundred times, we have all but forgotten what it really tells us and more importantly, what it does not tell us. As a sanity check, we were running 3DMark2001SE in a number of configurations. For demonstration we are going to concentrate on three benchmarks here, that is Car Chase Low and High Detail and Lobby High Detail. The rationale is that we need some calibration for the raw processor speed in the form of the Low Detail run that does not stress the AGP interface very much and then go into the high detail settings to add load on texture transfers and see how it all plays together.

ASUS V9280S vs. ASUS V8460 (Ti4600) vs. ASUS V8420 (Ti4200). The V9290S was run at either AGP 8X, or 4X (with 8X mode disabled in the BIOS) or with pin# A3 and A11 taped over to force AGP2.0 (4X) mode. The Ti 4600 was run in AGP 4X Mode and with AGP 8X disabled in the BIOS which results in AGP 2X mode. The Ti4200 was run in AGP 4X mode only. The results make it pretty clear that there is no benefit even from moving from AGP 2X to 4X. Neither is there any improvement in moving from AGP 4X to 8X, on the contrary, even though the delta is small, 4X mode consistently yielded the higher score. However, moving from AGP2.0 to AGP3.0 mode did yield a small (1.5%) performance increase.

At high detail, all cards perform basically the same except for the two "extracurricular" settings on the V9280S that, using the 4X AGP setting takes a small but significant lead while falling behind in forced AGP 2.0 mode.

At high detail in the Lobby, everything comes out similarly except for both 4200 trailing slightly in AGP2.0 mode

Most other parts of 3DMark2001SE did not show any difference for the different AGP settings. What the entire series of benchmark shows is that there is no impact of the AGP transfer rate on 3DMark2001 performance. Does that mean that AGP 4X has no benefit over AGP 2X or does that mean that 3DMark2001SE is optimized for anything but AGP transfers? By extension, is low dependability on AGP a feature of D3D?

Next Page:    => UT2003, The Saga Continues =>