Like the PC4000, the XMS4400 is designed to run exclusively on a Canterwood or Springdale chipset board, at a 1:1 memory ratio. Therefore our test rig is exclusively Intel this time.
First up, testing procedure. The motherboards we tested the memory with are the Asus P4C800E and the Abit IC7-Max 3 - the two most popular overclocking boards around at the moment. Naturally the processor is the now practically unavailable 2.4C Pentium 4 - but you will almost certainly need one, or a very good 2.6, plus extreme cooling (we are using a Vapochill) to get to Corsair's rated speed. FSB was initially set to 250MHz just for a warm up. At these speeds CAS 2.5 was no problem, although the rest of the timings still had to remain at a conservative 7-4-4. At those speeds the memory was 100% stable no matter how many loops of memtest we looped it through. Increasing to 260MHz saw us have to reduce the CAS latency to 3 in order to maintain stable operation, although a couple of errors in memtest on the IC7 Max 3 were observed - it was rock solid in windows.
At 265MHz the memory was unstable in 3D applications on the IC7 MAX 3 platform, and the system would randomly reboot when doing something as mundane as launching outlook or moving MP3s to another folder. On the P4C800 it was 100% stable - even in memtest. I have to conclude that if you have an Abit IC7 based motherboard, there's simply is no point in getting Corsair’s TwinX PC-4400. You are much better off getting the best quality PC-4000 you can, with the most aggressive timings available. If however you have the Asus, get ready for a treat!
At 275MHz the system was rock solid using 2.8V of current. However, I felt that we had not yet stretched the Corsair. Feeling brave, I increased the voltage to 3V - something I do NOT recommend if you just splashed out £300 on the most expensive DDR on the market! At this voltage I was able to boot at 290MHz. In memtest however, I was showing about 50 errors a pass. Feeling the sticks proved to be something of a hazard, as they were nearly burning hot at 3V. After letting them cool I lowered the voltage to 2.9V and strangely enough I completed 4 full passes at 290MHz before a single error crept in. To those not familiar with the test, one error in four passes of memtest is a stable system as far as I am concerned - it was able to run 3dMark 2001SE looped for 12 straight errors without a crash.
As you can see at this speed you get some SERIOUS memory bandwidth, with just below 7GB/s of raw throughput. Rather than repeat myself, cast your eyes back on our PC 4000 review (don't worry it launches in a new window). You will see that the benefit of running PC4000 vs PC3200 at 5:4 was not really that much of a benefit in most benchmarks, i.e. 500MHz at relaxed timings vs 400MHz at super tight timings was not showing much of an increase. This proved to scale nicely, with real-world performance at very high speeds being virtually analagous to a 5:4 ratio at very aggressive latency. With 550MHz rated XMS4400 however, Corsair have raised the bar. In order to compete in the same 5:4 ratio manner, you would now need above PC3500 specs (440MHz) at CAS 2 2-2-5 to compete. Virtually NO memory currently available will do this, especially now winbond BH-5 based memory is practically impossible to get. At the lofty hights of 290Mhz as we saw in this review, you would need similar timings at 233 (PC3700) speeds - something we have not seen yet bar a handful of "magic sticks". Without doubt, ultra high speed, high latency memory does have large benefits at this kinds of FSB.
