Above is a chart detailing all of the different
settings I used to benchmark. All of overclocked profiles I
used were tested for stability with a single pass of Memtest86.
As you can see, I got all the way up to DDR2-1100 before losing
stability without exceeding OCZ's EVP policy. DDR2-1150 would
not boot at all at that voltage, and DDR2-1125 failed in Memtest86.
Within minutes, I already had a 38% overclock. What is even
more impressive are these modules stay very cool, even at very high
speeds. This is compared to the Ballistix I have been using,
which are too hot to touch at speeds greater than DDR2-1000.
Now for the fun part. For each profile
above, I subjected the RAM to several different benchmarks,
including SiSoft Sandra, Everest, and PCMark Vantage. All
tests were run at 2.4GHz except for "OC #4", which was ran at
2.56GHz. Each benchmark was run twice with the best results
ClubOC has been using SiSoft Sandra for many years now and it's
pretty much been the main benchmarking program we have used for
testing memory. I used the most recent build available,
2008.1.13.12. Today will be no different, so let us see how
the Titanium fares.
The interesting bit to notice here, is that the
bandwidth scores don't seem to care so much about latency as they do
straight up bandwidth. Going from 5-5-5-15 in the "Auto"
profile to 3-4-4-15 in the "EPP" profile at the same clocks did not
do much to improve performance. The biggest jump came from
increasing the memory speed to DDR2-960 in OC #2. The CAS5 in
OC #3 did seem to hurt overall performance though, even with the
increase to DDR2-1030. Any losses were quickly made up when
the memory speed was maxed out at DDR2-1100.
The Sandra Latency test shows the opposite; the
biggest improvements were made when the changes in latency were the
largest. Go figure, eh?
Everest is another testing suite that looks at
memory throughput. While Sandra measures performance in terms
of Integer and Floating bit numbers, Everest looks at it in terms of
Read, Write, and Copy.
While the numbers are different, the story here
is the same. The biggest increase in performance comes from
bandwidth, not latency.
The Everest Latency test also mirrors the Sandra
test as well, with the best latency figure coming in at CAS5,
PCMark Vantage is the latest platform testing
software from FutureMark. One advantage it has over the older
PCMarks is how it separates different areas of testing into their
own testing suites. In this case, we will be using the
"Memories Test Suite". This suite tests not only RAM, but Hard
Drive performance as well, so I would expect that changes in RAM
speeds will have a smaller effect on the overall score.
The RAM actually had a larger effect than I
expected, increasing performance roughly 8%. Again, the
largest effect was increasing the bus speed to DDR2-1100.
I have been a huge fan of OCZ ever since I had
their original line of VX for my old DDR rig, and for good reason.
Not only do they produce top notch memory, they have an impeccable
reputation of standing behind their products. While running
excess of 2.0V through DDR2 memory would invalidate many warranties,
the Titanium is guaranteed up to 2.4V. And boy, did it take
that voltage and run with it. Not only did it blow through
DDR2-800 speeds, it carried those ultra-low latencies up to
DDR2-900. Relaxing the timings a bit, and it surpassed the
DDR2-1066 spec and hit DDR2-1100, and all 2.31V.
On top of it all, it runs much cooler than the Crucial Ballistix 1066 that I
paid quite a bit more for.
So is there a downside?
Not really, but I guess OCZ could include a six-pack of your
favorite poison with each set of RAM, that could sweeten the deal,
eh? Which is to say, there is no good reason not to seriously
consider these for your next DDR2 upgrade, especially considering
the price. Club Overclocker Recommended!
Note: Club Overclocker is
now using a new rating system based on a score of 1 to 5.
Please go to our rating system page for more information.
5 out of 5|
4 out of 5|
5 out of 5|
5 out of 5|
4 out of 5|
5 out of 5
4 out of 5
Project Skill Level
(5 being most difficult)
out of 5|