Corsair XMS4000 TwinX memory reviewed

JUST OVER a year ago, memory limped way behind the font side bus (FSB) on Pentium 4 boards. Part of the problem was, of course, the single-channel memory controlled design. Once we moved to dual channels, the twin DDR266 modules in Granite Bay were just fine for a 533 MHz FSB, perfectly matching the FSB bandwidth. Same story with dual-channel DDR400 on Canterwood or Springdale platforms, except that in this case, seeing a true 2-2-2 memory latency was kind of rare until recently.

Now, the FSB is still stuck at 800 MHz, however the memory gang is not waiting for the Tejas, still over a year from now, and its 1066 MHz FSB. Leading vendors like Corsair and OCZ have already reached the DDR500 (PC4000) level. Here we test Corsair TwinX XMS4000 1GB matched module pair - their current top-of-the-line set in Corsair's usual black metal finish.

The test platform was MSI 875 Neo board with 3.2GHz Intel Pentium4 CPU using the WinXP Pro SP1 "bloatware virus". We have already tested this board before with Corsair Platinum TwinX XMS3200 pure CL 2 (2-2-2-6 rated) DIMMs, and achieved excellent memory results. It was tempting to compare whatever best results we can get our of XMS4000 with the previous record holder.

The strategy was - do not change the CPU clock rate, leave it at 3.2GHz, but change the bus multipliers. So, we went boldly ahead with 12X multiplier on 266x4 (1066 MHz effective) FSB instead of usual 16X multiplier on 200x4 (800 MHz effective) FSB. The CPU and everything else worked fine, however memory couldn't run at DDR533 even with 3-4-4-7 latency; system just wouldn't boot - we had to run it at DDR425 instead, then system was stable in Windows benchmark tests. We also tried 250x4 (1 GHz effective FSB) with DDR500 memory setting at 2.85 V memory voltage - didn't boot either.

However, knowing that, on other mainboards, these DIMMs work fine even beyond DDR533, I suspect chipset memory controller to DIMM layout on MSI mainboard did not allow very high memory clock. So we tried another approach - set the CPU multiplier to 14X with 233x4 (933 MHz eff) rate - or a 3.26 GHz CPU clock, and hoped for the best with the latency.

First round, 2.5-3-3-7, worked fine, results within the expected range. Then I set the config to 2-3-2-6, a pretty aggressive setting for DDR466, and it also worked fine, going through all the benchmarks. Finally, I went for the kill and tried 2-2-2-6 settings at the DDR466, not being confident it will pass with lowest latency settings at such a high clock rate. Well, it went through all the way at 2-2-2-6 DDR466 with flying colours! A nice, really unexpected, surprise, I have to say.

Here are the benchmark results.

System	3.2 GHz	3.26 GHz	3.26 GHz	3.26 GHz
CPU	Pentium 4	Pentium 4	Pentium 4	Pentium 4
FSB speed	800 FSB	933 FSB	933 FSB	933 FSB
	2-2-2-6	2.5-3-3-7	2-3-2-6	2-2-2-6
	DDR400 dual	DDR466 dual	DDR466 dual	DDR466 dual
Sandra2003
Memory int MB/s Corsair	5096	5495	5529	5581
Memory FP MB/s Corsair	5097	5377	5448	5475
PCMark2002
CPU	8073	8126	8129	8130
Memory	9844	9816	9972	9985

As you can see, on the benchmark side, there is some benefit from pushing up the FSB, but not as much as expected from a 16% FSB and memory speed increase. In any case, running the Corsair memory at low latency at DDR466 dual channel, and realiably at that through the benchmarks, was quite a feat. µ