Overclocking and Testing:
The machine used to test and benchmark was configured with the components below:
» DFI NF4 Ultra-D motherboard (BIOS 623-2)
» A64 3200+ Venice core processor
» Thermalright XP120 heatsink with Coolermaster 120mm fan
» ATI Radeon X850XT video card
» Western Digital 36GB Raptor system drive
» 2 x 120GB Seagate drives in NV RAID 1 for data
It goes without saying that OCZ memory will operate at specified settings. The majority of people who are going to buy the OCZ Gold line are looking to get a good overclock with low latency. I dove right in and went looking for a high stable overclock at tight timings. OCZ has set this memory up with SPD timings of 2.5,3,3,5. These timings are loose enough that no one should have any issue booting up or running with this RAM, even if he or she does not know how to work around in their BIOS. First off, I loaded optimized defaults, changed timings to 2,2,2,5, set a CPU multiplier of 9x, a DIMM voltage of 2.8V, and proceeded to ramp up the speed. Immediately I jumped to the rated speed for this RAM which is 217MHz or 434MHz DDR (PC3500). As suspected, everything ran fine, Memtest showed no errors, and SuperPi finished a 32M calculation without issue.
The next step was to find the maximum stable speed this memory would run at the warranted 3.0 Volts. With a little fumbling and not just a few reboots, I finally landed at 228MHz or 456MHz DDR. 228MHz is a good speed, but I knew this RAM had more to give. With a little research and some past good luck with BH-5 based RAM, I felt moderately comfortable breaking the warranty and raising the Vdimm up just a bit. I consider "a bit" to be .6 volts so I moved the DFI memory voltage jumper to the 5v rail side and threw 3.6 Volts at the OCZ PC3500. This allowed me to get a full 260Mhz (520MHz DDR) and begin to test for stability. Stability can sometimes be a monster, and some things seemed to work fine while others did not.
The first two Futuremark offerings, 3DMark2001 and 3DMark2003, finished without issue. 3DMark05 refused, as did SuperPi even with the 1M calculations. I needed to slow things down or raise the voltage more. Being that I like this RAM, I decided to slow things down. I got down to 255MHz and (most) things returned to normal. SuperPi still refused to do a 32M calculation, but I have had issues with this before and have been stable. Wanting to be completely sure of stability, I chose 250MHz (500MHz DDR) for all benchmarks.
Now it's time to detail the other players I will test. I dug through the archives and came up with another 1GB set of OCZ BH-5 based RAM. This would be the 1024MB OCZ EL DDR Dual Channel Gold. In theory there shouldn't be any significant difference between these two sets of RAM. Just for comparison I also tested a set of Corsair TWINX1024-4400C25PT, which happens to be TCCD based.
Both of the OCZ sets were benchmarked at 2,2,2,5 1T 250MHz with a CPU multiplier of 9x. The Corsair TCCD based RAM was overclocked to 265MHz 3,4,4,8 2T and a CPU multiplier of 9x. This is the spot that best seemed to match up with the lower latency OCZ in terms of benchmark scores.
Benchmark Results:
Up first are a few looks at bandwidth. Sisoft Sandra provides both an Integer and a Floating point measurement. As you can see below, both sets of the low latency OCZ memory faired equally.
|
|
