Corsair
has a reputation among enthusiasts as makers of some of the best
ram in the business. Over the years, as ram technologies change,
as well as getting faster, Corsair was always out there, putting
out products to meet the demands of power users.
While DDR-II is still trying to gain mindshare, DDR-I is still
very popular, particularly amongst AMD users and Intel Socket
478 owners. If you're lucky enough to own a setup that can hit
250FSB and beyond, you'll know that PC4400 ram can go a long way
in maximizing your overclocking potential. The problem with ram
at these frequencies however is that the memory timings needs
to be relaxed in order to reach these speeds.
Today
we'll be checking out Corsair's latest DDR550 kit, the TWINX1024-4400C25PT.
As the name implies, this is a 1GB TWINX kit rated at PC4400,
but unlike their previous kit (3-4-4-8 timings), these new modules
are rated at 2.5-4-4-8.
Specifications
|
Part
Number
|
Speed
|
Size
|
CAS
Latency
|
|
|
XMS-4400
|
1GB
|
2.5-4-4-8
|
You can view the full specifications in this PDF,
but what you need to know is the modules are validated with an
ASUS P4C800-E motherboard at 550MHz. Since the memory controller
operates at 400MHz stock, the ram has a lot of headroom if your
CPU and motherboard have high overclocking ceilings.
The
ram modules arrived in a hard plastic case, with each ram module
fitted in the clear plastic shell. The only reason I'm pointing
this out is because there should be no confusion in what you're
getting. You have a clear view of the ram, and the labels (indicating
the type of modules), and the packaging is snap sealed.
The
TWINX1024-4400C25PT
modules have platinum coloured heat spreaders. The TWINX1024-4400C25
kit feature the same modules, except they're encased in black
heat spreaders. Sounding like a broken record, heat
spreaders have yet to convince me that they effectively cool ram
to the point where it will make a difference in overclocking,
but I guess any cooling included won't hurt. We have been told
that heat spreaders will reduce EMI since it shields the ram from
noise inside the case, but I don't have any quantitative proof
of that.
A
closer look at the modules, and we can see the silver sticker
letting you know the specs of that stick of ram. We can see that
it's a 512MB stick, rated at 550MHz, with ram timings of 2.5-4-4-8.
As many enthusiasts know, tighter timings (say 2-2-2-6) result
in better performance, but as memory speeds increase, it is extremely
difficult to maintain stability with such low timings. To be honest,
timings seem to be having a lesser effect on performance as memory
speeds increase as this was something we've observed with our
past reviews, but nonetheless, any reduction in CAS latency (so
long as it's stable) can only be a good thing. One thing to note
is Corsair claims these modules are capable of 2-2-2-5 at DDR400,
so if your current setup cannot hit 275FSB, these modules will
still run very fast at stock, and have the potential to run quickly
at higher speeds. How do they manage that? It just so happens
that the actual memory modules used in this kit are the same Samsung
K4H560838F-440 TCCD ICs found in Corsair's low latency 3200XL
kits.
All
of the TWINX kits are tested in pairs and in a Dual Channel
environment, and although the ASUS P4C800-E validates the modules
for DDR550 operation, all their modules are tested on various
875P, 865PE and Socket 939 based AMD motherboards for compatibility.
Overclocking
A RAM
Stress Test Professional (R.S.T.) Pro2 card will be used
to test for reliability as well as stability during our overclocking
tests. The card is quite expensive, but if you're testing a
lot of ram, or you're a technician troubleshooting systems,
this card is well worth the money.
How it works is that the R.S.T. Pro2 goes into
a free PCI slot and after POST, it bypasses the OS completely,
loading a proprietary OS and software package (embedded on the
card itself). The card allows you to perform stress and performance
benchmarks without the Operating System's memory footprint,
memory resident programs, or drivers interfering with the memory
installed. We'll provide a full review in the future, but that
is the gist of it.
We proceeded with overclocking, and tested the
ram via the memory tests on the R.S.T. Pro2 until we got a failure.
In order to maximize our OC potential, we used a Pentium 4 2.4C
which has a low multiplier of 12 and will allow a bit of FSB
headroom on our ASUS
P4C800-E motherboard.
|
Timings
|
Max
OC
|
Pass
R.S.T. Pro2
|
2-2-2-5
|
217
|
No
|
| 2-2-2-5 |
213
|
Yes
|
| 2-3-3-6 |
242
|
Yes
|
| 2.5-3-3-6 |
271
|
No
|
| 2.5-3-3-6 |
265
|
Yes
|
| 2.5-4-4-8 |
278
|
Yes
|
| 3-4-4-8 |
282*
|
Yes
|
While we did not include every overclocking test in our results
above, we did include some that were worth noting. Two of the
above results failed our R.S.T. Pro2 tests, yet they did run
fairly stable nonetheless. Note that I said "fairly".
While we were able to perform benchmarks, occasionally we would
lock up the system, even though it succeeds in completing the
failed benchmark in later attempts.
At 282FSB, the memory passed the R.S.T. Pro2 tests, but only
when the vDIMM was pushed to 2.7v and we placed an 80mm fan
on top of the modules. Without the additional cooling, the ram
would fail the R.S.T. Pro2 tests, but the system would still
indeed POST and get into Windows. The system wasn't very stable
though, and without the extra cooling, the setup was happiest
at 280FSB.
Test
Setup
ASUS
P4C800-E: Intel P4 2.4C, 2 x 512MB Corsair TWINX1024-4400C25PT,
ATI 9600 XT, 120GB Seagate SATA 7200rpm, Windows XP SP1, ATI
Catalyst 4.10.
The testing configurations will be as follows:
DDR400 @
2-2-2-5
DDR542
@ 2.5-3-3-6
DDR550 @ 2.5-4-4-8
DDR564 @ 3-4-4-8
Testing software will consist of the following:
SiSoft Sandra 2004 Memory - Our standard synthetic test to
establish a baseline.
Aida 32 Memory Tests - New to our test suite, this is another
synthetic test to support (or discredit) our SiSoft results.
PiFast - A
good indicator of CPU/Motherboard performance is PiFast
version 4.2, by Xavier Gourdon. We used a computation of 10000000
digits of Pi, Chudnovsky method, 1024 K FFT, and no disk memory.
Note that lower scores are better, and times are in seconds.
Quake 3 - While it's old and moldy, it still has some value
as a system level benchmark.
SiSoft Sandra 2004 Memory
No surprises here as we see the performance scales
proportionately with the clock speed. We also see that memory
timings seem to have a lesser affect as FSB increases as well.
