The FX-53 continues to lead in Doom 3, however the margins are slim at best. The 3200+ is able to demonstrate a better lead over the 3400+ and 3500+ than is the FX-53 over the 3200+. At higher resolution the game is more graphics bound and CPU speed becomes less relevant. Although leads persist, they are generally smaller except the 3400+ drops about 10fps to the socket 939 processors.
UT2003 benches paint a muddled picture. The 3400+ hangs in there with the 939 chips and even wins a few at 2GHz speeds. After overclocking the 3400+’s lack of memory bandwidth becomes a slight problem and the 3200+ and FX-53 pull ahead slightly. The 3200+ just seems to get stronger and pull closer to the FX-53 as HTT speeds increase. The 3500+ is generally left as the trailing candidate to the other CPUs.
Here we see somewhat of a surprising result. The 3400+ socket 754 just wipes the floor with the 939 chips. Whereas the 3400+ is blessed with 1mb L2 cache and could potentially explain why it is faster than the 3200+/3500+ CPUs, there is no good reason why the 3400+ is faster than the FX-53. I am going to guess that either there is something well optimized in the Chaintech/3400+ combo or something is amiss in my MSI 939 setup in regards to the UT2004 3D game engine.
Far Cry1.1 with the non overclocked CPUs is showing similar results to previous game benchmarks. In Research we see the FX-53 trailed by the 3200+, then followed by the 3400+ and 3500+ respectively. In PC Games Hardware Demo the 3400+ bests the 3200+ at low resolution, but after 1024x768 the 3200+ is winning, and by 1280x1024 the 3400+ is beat by all 3 other processors. The FX-53 again takes the crown, with the 3200+ nipping at its heels
While overclocked the distinction between the FX-53 and the 3200+ fades and we are left with virtually identical numbers. Each CPU trades barbs and in both Research and Hardware demos the 3200+ leaves the parting shot with the very minor win.
Unfortunately for the 3400+, in the overclocked setting the higher ram bandwidth of the 939 chips is overshadowing an otherwise good showing. The 3400+ may have performed better if I could have gotten some ram to run 1:1 at 250HTT on the Chaintech. However, due to clearance issues with the Thermalright XP-120 and difficulty in scaling ram at higher HTT speeds on a 754 board this just wasn’t possible for us to do in this review.
Onto our conclusion ->