Corsair PC3500 XMS ReviewBosco - February 8, 2003
Dual Channel with ghetto northbridge setup
I told you this was going to be a quick one, the only benchmarks being used for this review are SiSoftSandra 2003 Memory Bandwidth buffered and 3D Mark 2001SE build 330. Once again, keep in mind that what you are seeing is 256MB of the Kingston vs 512 MB of the Corsair. If anything, I believe this makes it harder on the Corsair, as, in general, a system can run with stability easier on 1 dimm than on 2. I'll go processor by processor so you can see the scaling.
On the Kingston Value RAM, the best overclock I had been able to get out of the 1700+ TBred A was 10.5 X 166 = 1743 MHz, with quite relaxed 2.5-3-3-7 memory timings. The memory voltage options in the MSI K7N2-L's bios DO NOT WORK, this is a known issue, you get 2.6v VDimm no matter what you set it at - at least that is one variable taken out of the testing. Inserting the Corsair sticks in dimms 1 & 3 ultimately yielded a stable 11 X 169 = 1870 MHz, per CPUID. An extra 127 MHz, or 9% over default for the cpu - and with dual channel memory at 2-2-2-5 FAST timings. Bonus. I am certain that with better cpu cooling, higher numbers could be reached. And, nothing but a black screen when trying anything over 10.5 X 166 with the Kingston - well, it would boot into Windows occasionally at 169 fsb but was unstable on the desktop, even at 3-4-4-7 - most likely due to being unable to feed it more voltage. In all cases, memory is run in sync with front side bus on AMD cpus. For some reason, this cpu on this motherboard just hated fsb beyond 166.
The most I've been able to get out of my P4 2.66 and Asus P4PE on stock air cooling has been 155 X 20 = 3100 MHz. This is with DDR 388, 2.9v to the Kingston, and 2.5-3-3-7 timings. As I'm not willing to push the vcore beyond 1.75v without better cooling, I have ended up with the same overclock on the Corsair but with much better DDR413, 2-2-2-5, and only 2.6v to the RAM. I did boot into Windows and benchmarked SiSoft Sandra memory bandwidth at 160 fsb and 427DDR but neither more relaxed timings nor more voltage would make it stable for 3D benchmarks - likely a cpu bottleneck rather than a memory bottleneck, in my estimation. Please note that both 133 X 20 benchmarks are with DDR 355 and Turbo RAM and system timings.
I received a 2100+ TBred B very late in testing. I went straight to the Corsair modules with this one and wow is an appropriate word to use here. I am still testing, but as I write this I am at 182 X 13 = 2369 MHz per CPUID and 100% stable with 2-2-2-5 timings on dual channel and 1.775v (1.82-1.84v actual) to the cpu. 37% overclocked on mediocre air cooling anyone? Memory run in sync @ DDR 364, as no benefit will be found running higher memory clock asynchronously with front side bus.
Awesome. Believe the hype....well, if it's true maybe it's not hype? For sure the above testing has not stressed the very limits of this RAM - in the Intel case, it's cpu limitations. In the AMD case, it's motherboard limitations - or at least that is my belief, based on both personal experience and tons of reading in forums and reviews around the net. The Corsair PC3500 XMS gave me the ability to go well beyond all my previous bests.