Review: Intel's 915P and 925X w/ LGA775, DDR-II and PCI-Express

by Ryszard Sommefeldt on 19 June 2004, 00:00

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qayu

Add to My Vault: x

3D Tests

Testing 3D performance turned out to be slightly suboptimal since I was unable to secure an almost identical 3.4E or 3.4EE-based Canterwood system, to go up against the Alderwood.

Instead, the FX-53 provides sufficient CPU grunt to test performance against the 3.4EE on LGA775, neither CPU able to influence the mighty 6800-series performance in GT or Ultra form over a couple of easy to run tests. The three tests were designed to highlight two things. First, FX-53 performance up against 3.4EE on Alderwood, to determine which is the better enthusiast platform for top-end 3D gaming. Secondly I wanted to see the performance difference between AGP and PCI Express systems, when CPU limited with equal graphics cards (bar their interface), to see if PCI Express conjures up any magical increase.

So 3DMark 2001SE is used to provide the CPU-focussed overview and 3DMark03 and Painkiller are used to compare AGP and PCI Express.

3DMark 2001SE



The AGP-based system, FX-53 with 1GB of DDR-I RAM on SK8V, comfortably outpaces the LGA775 system (3.4EE, 1GB DDR-II, Alderwood desktop board).

The 3.4EE doesn't provide enough CPU grunt combined with Alderwood, to oust the reigning hardcore gamer's CPU of choice.

3DMark03



While the AGP-based system still has the edge, generally things are as close as you'd like to call them in a real world scenario. Being limited by the GPU in 3DMark03 has the nice effect of letting us compare the interfaces, given identically clocked GPUs in the same family.

The FX-53's wins are accounted for by its superior performance in 03's Game Test 1, the only test the CPU is able to influence. For all intents and purposes, the AGP and PCI Express performance of NV40/NV45 is equal, in a generic gaming situation.

Painkiller



The PCI Express system is within 2.5% of the performance of the AGP system when equipped with 6800 GT and there's less than 1% deviance when we equip the systems with 6800 Ultras. Exactly as you'd predict, given the bus usage pattern of current applications, especially games. GPU-to-host writebacks are rarely the bottleneck in current games.

Summary

While it's a giant cop-out, a proper roundup of PCI Express graphics performance will have to wait for a separate article, time constraints and part supply limiting what I could do in this article. Look out for that in the near future.