Corsair will be the first to admit that these sticks are designed to run almost exclusively on a Canterwood or Springdale chipset board, at a 1:1 memory ratio. For this very reason, we have decided to drop the usual cross platform testing, and go exclusively Intel for a change.
First up, testing procedure. The motherboard is an Asus P4C800 (the IC7 from Abit has issues running faster than 257MHz in 1:1 synchronous), the processor a 2.4C Pentium 4 specially picked out by the nice chaps at Chillblast, and a host of other hardware not relevant to this review. FSB is set at 250MHz for both the XMS 4000 and the XMS3200LL we are comparing it to, although the former is set at 1:1, with the latter at a 5:4 ratio. We were able to increase the XMS4000 to CAS2.5, 7-4-4 with a little extra volatge, so those results are shown as well as the stock timings (CAS 3, 8-4-4). XMS3200 was running at CAS 2 - 5-2-2 and 200MHz.
First up the mandatory Sandra tests. Right away you can see that in terms of raw memory bandwidth the XMS 4000 comes out in the lead. Note that at CAS 2.5 Sandra performance is unchanged, which speaks volumes about its usefulness as a memory benchmark. Unbuffered tests for those who want to compare, are gathered by deselecting the eight tests from "Enable MMX (Integer) Benchmark" down to "Enable SSE2 (Float) Benchmark", as well as "Enable Buffering/Block-prefetch Benchmarks" in the bandwidth test options.
3DMark 2001SE is a much more useful test than its newer brother, which is, frankly, a joke. The test was run at default settings, as although 640x480x16 would be more revealing, you won’t be trying to impress your mates at those settings. As you can see, the inclusion of XMS4000 makes little or no difference over the very-low-latency XMS3200.
Quake III is highly dependant on the memory subsystem for performance, especially at the 640x480x16 resolution selected. The XMS4000s high latency is clearly hurting it in this test and the benefits from switching to CAS 2.5 results in an 18fps gain. However, the XMS3200 still comes out on top.
Comanche 4 is an odd benchmark, it is highly stressing on the graphics card, as well as the subsystem, mainly because it is poorly coded in every way imaginable. In this test, (640x480x16, no sound) we can see that the XMS4000 is favoured over the XMS3200 by a reproducible margin. Interestingly the difference between CAS 2.5 and CAS 3 is non-existent, showing that Comanche 4 favours raw bandwidth over latency.
