System setup and notes
Software
- Windows XP Professional Build 2600.xpclient.010817-1148
Notes
We'll examine the benefits of high latency and high MHz against lower latency and lower overall bandwidth. The semi-unlocked P4 3.2GHz allows us to run in the following combinations:
3021MHz - 201.4FSB - DDR402.8 - 15x multiplier - 2-2-2-6 timings: These settings indicate just how a well-tuned 3.0GHz 800FSB CPU would perform with tight, low latency DDR400 memory. We're thinking along the lines of Corsair's very own TwinX LL here.
3041MHz - 233.9FSB - DDR467.8 - 13 x multiplier - 3-4-4-8 timings: We're attempting to keep the CPU speed relatively constant by lowering the multiplier and raising the FSB and memory speeds. We're thinking of a pseudo 2.6GHz 800FSB CPU overclocked to 234FSB and using Corsair's TwinX XMS3700 matched memory kit in this instance.
3007MHz - 250.6FSB - DDR501.2 - 12x multiplier - 3-4-4-8 timings: Here is how we try to compare the Corsair TwinX XMS4000 matched memory kit to the other two setup's results. This would indicate a 2.4GHz 800FSB overclocked to 250FSB and running the TwinX kit in synchronous mode.
The testing runs along the lines of the standard XMS4000 set. The differences this time, though, are in the use of a newer, performance BIOS for the EPoX Springdale motherboard and a Radeon 9800 Pro with the Catalyst 3.7 driver and control panel. Expect benchmarks to be a little different than before, then.
Issues
Corsair states that the modules have been tested, as a pair, on an ASUS P4C800 Canterwood board at settings of DDR500, 3-4-4-8 latencies and 2.75v voltage. We first tried the TwinX XMS4000Pro in an EPoX 4PDA2+ Springdale motherboard. SPD settings were detected as 3-4-4-7 at DDR400 speeds. We manually inputted the requisite 3-4-4-8 settings, raised the VDIMM voltage to 2.8v and set the board up at 250MHz FSB (synchronous RAM). Whilst it would load Windows XP consistently, the modules would refuse to run any memory-intensive tests with stability. Swapping out the modules for some OCZ PC4000 EL DDR RAM showed that the motherboard and constituent components were capable enough. The XMS4000Pro was then tested in an ABIT IC7-MAX3 and DFI i865PE Infinity. Both motherboards reported similar findings to the EPoX, that is, the memory would boot at its rated speeds but wasn't totally stable. Re-testing in the EPoX Springdale board, this time with 2.9v, seemed to cure most of the stability-related problems. However, let's be clear about this, this isn't how it should be, and testing on three proven motherboards backs up the case.
We then purchased an ABIT IC7-MAX off the shelf and tested the modules in that. They appeared to run a lot, lot better, and managed to hit the asking speed of DDR500 with 2.7v load voltage. It's puzzling, annoying and frustrating for modules to work considerably better in one board than a number of others.
We did manage to conduct our benchmarks on the resident EPoX 4PDA2+ with 2.9v VDIMM. We weren't ecstatic about putting that voltage through the ProSeries, though. Even 2.9v would fail test 5 in Memtest86, a pre-OS memory-checking utility. We'll come back to this issue when summarising.
