Corsair XMS4000 on MSI 875Neo post-mortem

REMEMBER OUR our recent test of Corsair XMS4000 DDR500 RAM on MSI 875 "Neo" mainboard, where Neo seems to cheat the BenchMatrix without any slow motion flying kicks - the RAM works at an astounding CL2-2-2-6 at DDR466?

Well, OCZ, Corsair's competitor in the "enthusiast's memory" arena, alerted me that I could have been fooled by the crafty mainboard, which shows you one thing on the screen, but something very different happens in real (oh, I thought only CNN and BBC do this?). In this case, they claim, it may display 2-2-2-6 on BIOS screen, but in real it still sticks to SPD pre-programmed settings of 3-4-4-8. And there was no one else achieving this!

But hold on - when I ran the benchmarks, the results did differ between different latency settings, after all. On the other hand, OCZ could have a valid point - hey, no one else can get this result, something must be fishy! Is MSI cheating the users by making them perceive they run better latency settings than what is really running? Who knows?

So, we retested the Neo to see whether it really cheats the evil BenchMatrix.

This time, we went all the way and set the settings to UltraTurbo, and two different settings - CL3-4-4-8 at DDR466 with SPD enabled first, and CL2-2-2-6 at DDR466 with SPD timings disabled afterwards.

I changed the benchmark this time, using Sandra MAX3, a newer replacement for the old Sandra 2003. So I ran the memory bandwidth benchmark, obtain a fairly low set of results (see Pic 1), and confirmed statement on the screen that Sandra recognises the DIMMs as CL 3-4-4-8. Then I restarted the system, set the timings in BIOS to CL2-2-2-6 at DDR466 with SPD timings disabled, and booted Windows again.

When running the Sandra MAX3 memory bandwidth benchmark again, it showed the same poor results as just before that - no difference at all compared to CL3-4-4-8 SPD settings! All that despite running the benchmark thrice in both cases. And yes, Sandra was recognising the DIMMs as 3-4-4-8 SPD settings. So, it must be that the new settings really didn't take effect after restart.

So, I shut down and turned off the system, and turned it on again, while leaving it at 2-2-2-6 settings. Guess what, when running Sandra MAX3 memory bandwidth benchmark this time, the results were some 4 - 5 % better in all three runs (see Pic 2), and even the memory bandwidth efficiency was recongnised as 2% better! So, definitely, there was a performance improvement commensurate with better latency - so the new settings did take effect after power off and on, didn't they? However, Sandra still recognised the DIMMs as 3-4-4-8. Could it be that Sandra would read the SPD settings (which were ignored now by BIOS) rather than actual BIOS settings?

In summary, Neo was a nice guy and didn't cheat the BenchMatrix - it seems to me that MSI does use these settings for real, but you have to power on and off the system to really enable them. How come the Corsair XMS4000 DIMM pair we tested is so damn good at DDR466 true CL2 then? After all, it is not one isolated piece, but two matched ones "in sync"...

The Sandra results are here and here.

Hey, tell us if anyone else has got such results! In the meantime, I will try more benchmarks, and maybe more Canterwood boards (if the board vendors are willing to send them, hah). µ