After a bit of initial experimentation using my Epox 4PDA2+ V2.0 i865PE and 2.4C, I discovered that 2.5-7-3-3 was the tightest stable timings possible, regardless of clock frequency. The ram was fine until about 480mhz, when reducing to 2.5-7-4-4 was required to retain stability. 2.5-7-4-4 and 3.0V worked up to about 540mhz, (already easily surpassing the limits of the XMS4000 we tested). The OCZ PC4000 Gold wasn’t about to quit there, however. This Epox board allows for up to 3.3V of Vdimm, so I upped it to 3.3. (Note: I would strongly urge anyone against trying this, 3.3V is enough to seriously damage your ram.)
I was expecting a CMOS reset at 3.3V, but to my surprise, the system booted into Windows without issue. After reducing timings further to 3-8-4-4, I was able to get this ram to a record shattering 297mhz FSB @ 1:1 (That’s DDR 594 folks.) At that speed my CPU was running at 3.56ghz, which is pretty much the absolute limit that my basic air cooled setup will support. It is certainly a possibility that someone with a Vapochill or some serious water cooling may be able to get more than 600mhz out of this ram. Testing was done with FSB at 297mhz, at both 5:4 (for DDR 475) and 1:1 (for DDR 594). The PC4000 Gold has no problem with async operation, but as you'll see below, you're going to want to use 1:1.
Sandra memory bandwith scores were astounding, easily surpassing the best that even the PC3700 Gold was able to achieve. Only a quad channel RDRam setup would be able to challenge these scores, and the R659 is MIA.
Going from 5:4 to 1:1 yields a healthy 600 point boost in 3dmark 2001, approching the magic 20,000 mark.
UT2003 gains were similar impressive, infact, the performance boost was more than I expected. Most impressive.
