With
the release of the nForce 2, Intel i865PE, and Intel i875P, Dual
Channel memory motherboards have finally reached the mainstream.
There have been Dual Channel mobos have existed previously, but
they haven't been as refined and readily available as they are
now. We've covered
the technology behind it before, but basically you will get
a nice performance boost with a Dual Channel setup over a Single
Channel motherboard, CPU speeds being equal.
Currently,
the fastest DDR supported by these motherboards is 400MHz, or
PC3200. With the release of the TWINX1024-4000,
one will have to ask why do we need 500MHz ram? The answer is
simple… overclocking. It's no secret that the Pentium 4 2.4C can
hit 250FSB without breaking a sweat. Up until now, we've had to
run at a 5:4 ratio to maintain stability with our PC3200 modules.
We do run our ram a little faster as we pass 250FSB, but stability
is questionable as we ramp up the clock speed. Furthermore, since
the CPU's FSB and memory speeds are not running in sync, performance
is not maximized.

With
the Corsair TWINX1024-4000 we'll be reviewing today, we now have
an opportunity to run synchronous at 250MHz. Can we go beyond
that? We're definitely going to try.
Specifications
As
the title of the review says, the memory is sold as a kit. We
received 2x512MB modules, tested as a pair to insure maximum performance.
Here's a bit from Corsair's site:
"Moreover,
the company developed a stringent battery of tests which every
XMS module built must pass at its rated speed before being shipped.
XMS4000 modules are tested at 500MHz with 3-4-4-8 latency settings
at 2.75V. All XMS4000 modules are tested in an Asus P4C800 dual
channel DDR motherboard based on the Intel 875P "Canterwood"
chipset. TwinX kits are tested as a matched pair and physically
packaged together immediately after passing test, guaranteeing
that the customer receives a product that has been verified
in the dual channel environment."
The
Corsair TWINX1024-4000 Memory Kit
Corsair's
packaging has improved since we've last looked at their TWINX512-3200
kit. Rather than two individually packaged ram modules, we now
got both of them in the same package. This of course doesn't really
mean much, but it does make it clearer that this kit is a matched
pair.

Like
past Corsair XMS modules, their ram is encased in black heatspreaders.
I'm preaching to the choir here, but heatspreaders have yet to
convince me that they effectively cool ram to the point where
it will make a difference in overclocking. They do look swank
though, and I guess any cooling included won't hurt.

It's
hard to demonstrate with a picture, but the heatspreaders are
attached to the ram via some frag tape. Not the ideal choice if
cooling really matters, but it is less costly than individually
applying thermal epoxy at the factory.

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 500MHz, with ram timings of 3-4-4-8.
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,
3-4-4-8 isn't bad at all for 500MHz. Sure, it can be better, and
we'll see how we can do with some tweaking.
Overclocking
and Stability Testing
Since
we're using a 200FSB (800MHz) ABIT IS7-G motherboard, the first
thing I wanted to try was what are the tightest timings I could
run the TWINX at at 200FSB. After a bit of tweaking, it seems
that 2-3-3-5 is about as low as I can go.
 |
 |
|
200FSB
(400MHz) @ SPD
|
200FSB
(400MHz) @ Tweaked
|
With
that out of the way, I did a double check if the ram can indeed
handle the 250FSB at 3-4-4-8. One thing I noticed that at the
SPD settings, the timings were 3-4-4-7 (as was the case at 200FSB).
This was confirmed using CPU-Z, so I guess the ram is slightly
tweaked out of the factory.
Next
thing we did was to adjust the timings and to see what the TWINX
would allow us to tweak to. At 250FSB, the best we managed was…

250FSB
(500MHz) @ Tweaked
The
ram was completely stable, and passed MemTest without any problems
at 2.5-4-3-6. Memory voltage did need a boost to 2.7 though. How
about OCing a little further? At these timings...

251FSB
(502MHz) @ Tweaked
We
only managed to get as high as 251FSB. I managed to get into Windows
at 253FSB, but the PC would immediately reboot after. Memory voltage
was at the ABIT IS7's maximum, 2.8v, and the CPU was cooking along
at 1.8v. We moved the timings back to 3-4-4-8, and gave it another
go, but the system would never pass 255FSB. After a few phone
calls, and more testing, it has been confirmed by both ABIT and
Corsair that the IC7 and IS7 series have a memory FSB issue at
255 and up. ABIT is working on it though, and a fix should
be available in the near future.
Obviously,
we weren't going to call it a day, so we swapped the ABIT IS7
out, and dropped an Asus P4C800-E in its place. Like the ABIT
board, we were able to match the above 200FSB and 250FSB timings,
but we weren't able to improve upon them. Our maximum overclock
was a different story.

272FSB
Max OC
To
be able to run your ram 1:1 at 272FSB is simply amazing. When
we pushed to 273, MemTest failed, but the system still ran. The
TWINX didn't seem to like any timing adjustments at this speed
though, as we could never stay stable once we got into Windows.
Just to clear something up, our 2.4C can do in excess
of 300FSB using a 5:4 ratio, but that isn't the point here. We
are trying to test for the maximum memory OC, and our board does
not offer options where the memory can run faster than the CPU
FSB.
Test
Setup
Asus
P4C800-E: Pentium 4 2.4C, 2 x 512MB Corsair TWINX1024-4000, ATI
AiW Radeon 9700 Pro, 80GB Western Digital, Windows XP SP1, ATI
Catalyst 3.5.
Test
Software for both platforms will be:
SiSoft
Sandra 2003 Memory
PC Mark 2002
Memory
PiFast
TMPGEnc
AVI-to-MPG Encoding
Unreal Tournament 2003
Splinter Cell
Unfortunently,
I do not have any 512MB/PC4000 sticks of competing products. We
will be presenting benchmarks at 200FSB, 250FSB and the maximum
OC on the P4C800-E, all in Dual Channel mode.
SiSoftware
Sandra 2003 Memory - Pentium 4 @ 12x200
Although
a synthetic benchmark, it's a popular one, freely available if
you wish to make comparison benchmarks. We will be testing the
memory speeds only.

PC
Mark 2002 -
Pentium 4 @ 12x200

For
the two synthetic benchmarks, the tweaked timings show a slight
improvement over the standard SPD. If the ram can handle it, there's
no reason to settle for stock timings I always say.
PiFast
- Pentium 4 @ 12x200
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.
As
expected, a small improvement with the tweaked timings.
TMPGEnc
MPEG Encoding - Pentium 4 @ 12x200
Video
editing is a taxing chore, and we'll be testing the TWINX using
TMPGEnc 2.512 to encode a 7.78MB, 1:30 movie trailer to a 23FPS
MPG file. Note that lower scores are better.
Tweaking
the ram to 2-3-3-5 knocks another second off the encoding time.
Though just one seconds may not seem like much, encoding a MPEG
file that's 20 times larger will add another 19 seconds and
so forth. Any time saved is always good to me.
Unreal
Tournament 2003 - Pentium 4 @ 12x200
UT2K3
s a real system killer, and can bring many systems to its knees.
We used the [H]ardocp
UT2003 Benchmarking utility version 2.1, which are excellent
tools in testing various resolutions and detail levels. We selected
the CPU test, which uses the dm-inferno map.

Splinter
Cell - Pentium 4 @ 12x200
New
to our test bench is UbiSoft's 3rd person action game, Splinter
Cell. We used the Beyond3D demo, and ran the demo at 640x480
at the lowest detail levels.

Both
games prefer fast ram, with Unreal Tournament 2003 showing the
largest gains. We've now seen what the ram can do at the CPU's
stock speed at a 1:1 ratio... let's checkout a 250FSB overclock,
1:1.