System setup, test notes, assumptions and drawbacks.
Motherboard: DFI X58 EH6. The X58 and FET heatsinks were checked for contact quality and thermal paste was replaced with Arctic Ceramique where appropriate.
CPU: I7 920 C0. Watercooled.
GPU: 512MB 9800GTX+ (The 512MB is important when dealing with RAM mirroring)
O.S.: Windows 7 RC1 64-bit and XP Pro 32-bit
Tests: Memtest 86 and SPi32M.
RAM performance is closely tied to the CPU now as opposed to a capable motherboard, these results may not be achieveable by just anyone with an I7. Performance consistency, even within the same CPU batch is hit and miss. I believe my CPU to have a fairly capable memory controller, although its BCLK limit is fairly average at around 215MHz. Conversely, people with more capable CPUs may find this results easy to beat.
Two low-speed 80mm fans were positioned over the RAM at all times.
Absolute maximum performance limits were not sought, but rough numbers that most users should be able to acheive (subject to the points raised above.) There is scope for performance differences between kits, so I belive its the fairest way to present results.a QPI voltage was set at 1.44v for all tests. CPU voltage was adjusted as necessary for the clockspeeds but the CPU multiplier was kept fairly low to remove MHz as a possible bottleck.
RAM and Uncore are closely connected in I7 chips. Uncore speed (derived from multiples of the RAM speed) is MINIMUM 4x the RAM MHz so if the RAM is at 1GHZ, the Uncore speed is 4GHz.
Not every chip is capable of such frequencies so some might buy this RAM (or other high-performing kits) and not be able to use it at default speed due to a CPU limitation.
Windows 7 RC1 is the only 64-bit OS I have access to and due to software compatability I was not able to run as many stress tests as I would have liked. Despite being a synthetic app, Memtest should show up any cracks in RAM stability.
Despite the relatively low number of results posted, that does not mean there were not a lot of results discarded along the way as the RAM was tightened up. Memtest benching takes a long time and I do not have an unlimited time budget!!!
Going over 950MHz resulted a loss of stability or unreliable behaviour. I encountered a sitaution where one RAM channel was dropped from detection upon startup. BIOS and Windows Task Manager reported a 4096MB RAM volume and there have been other users reporting similar problems at high RAM speeds. I believe this is either the first sign of instability or a limit in the IMC. I did not carry out any testing when this occured as it would be inappropriate for the purpose of this review
I do expect readers of this review to have a basic understanding of I7 behaviour. If anyone is truely stuck, PM me and i'll do what I can do help.
As always, I will not test at Crucials rated spec (subject to being aware of the stock spec!) I trust their quality control.
From a basic spec of 800MHz, 8-8-8-24, 2T @ 1.65v I was hoping to be able to find extra running options without resorting to higher vDIMM values. I had aready "found" 7-7-7-24, 1T to be stable at 1.65 before I realised there are XMP profiles loaded. However, even the XMP profile sets a command rate of 2T, so being able to move from 2T to 1T without increasing any voltages was a plus point. Moving on from this, I was able to tighten up the tRP to 6 without any trouble, to give 7-7-6-18, 1T at 1.65v.
Despite trying with higher voltages- up to 1.80v, I was unable to get the computer to POST with the CAS or tRCD set to 6. I have a feeling this is a BIOS limitation.
Aiming now at higher MHz, I found 850MHz 7-8-7-18, 1T to be fine at 1.75v.
I was slightly disappointed that I couldnt hold tRCD or tRP to their -7 and -6 latencies respectively much over 800MHz, even with 1.80v.
Moving on again, 900MHz and on to 933MHz were no problem at 8-8-7-18, 1T at 1.65v
This did impress me as it overtook higher-binned products. Stability tailed off quickly after 933MHz and I chose not to push it took hard as for true 24/7 stability there should be some breathing space left in the RAM.