Corsair TwinX1024-3200LLPRO

A Gig of the Good Stuff – Brian Berryman

SUMMARY: A “Gig” of the Good Stuff

Today, we’re going to take a look at some DDR400 speed memory, which the good folks over at Corsair
were kind enough to send out to us.

Corsair sent us one of their Corsair TwinX1024-3200LLPRO
high speed, low latency memory kits to take a look at.

I’ve been using Corsair memory in my main/gaming machine for quite some time now (refer back to any of my article reviews in the recent past…you’ll see it in the background
of quite a few pictures).

The ram I have been running is their CMX-512 PC3200C2 revision, using a pair of these
in Dual Channel mode on my Abit NF7-S v2.0 motherboard.

While these two older sticks of ram weren’t a matched set like the LLPRO we’ll look at today, they work perfectly in DC mode.

In the process of reviewing Corsair’s newer version of their PC3200 memory here, we’ll be taking a fairly good look at the older C2 spec ram as well, comparing the two revisions.

The lone “accessory” included with the kit is a Powered by Corsair window decal. There’s no coupons for Half Life 2, or rounded cables, but then, when you make some of the
best memory on the planet, you don’t need extras.

Of note above, in the lower right corner, is the number 2326. These are the timings (set in the motherboard’s BIOS) that this memory runs at, at default speed.

In testing these out, I found that they will actually run at tighter settings, at 6-2-2-2. I’ve read reports of some users getting these to run even lower than that, to 5-2-2-2. I did try this
setting out, and my PC wouldn’t boot. I did run them through their paces, starting at the default settings, and working my way up. This ram overclocked quite nicely in my system.

Too nicely, as we’ll see shortly.

The new PRO Series from Corsair now incorporates activity LEDs into the spine of the heat spreader. These aren’t just a gimmick, they actually work, and work well.

In the process of testing various settings (when really pushing my system hard), I did lock up the rig hard a couple of times. Looking through the window in my side panel at these
LEDs were a very quick visual cue that something was amiss, as they weren’t blinking. The page at Corsair linked to above (2nd link at top) includes a Macromedia Flash presentation of them in action.

Here is a static (still) image of them:

The effect visually is not unlike a stereo graphic equalizer. Green to red, with red indicating highest activity.

It was quite interesting to watch these lights as I ran the benchmarks I’ll get to shortly, as they visually showed how much ram was being used during testing.

• Tested (and packaged in pairs) at DDR400 at ultra low latency settings (2-3-2-6)
• SPD programmed at advanced 2-3-2-6 values
• 1024 MB total (two DIMMS of 512 MB each per kit)
• New high efficiency aluminum XMS heatsink
• 18 activity LEDs (per DIMM) show level of memory activity
• Benchmarked over multiple chipsets, processors, and motherboards
• Lifetime warranty

For these tests, I used my AMD based system. The motherboard decided to stop working recently in my Intel machine, so until I can replace the board, the AMD rig is it.

This machine has been used in most of my hardware reviews (in addition to that now defunct Intel P4 machine), and there’s only been one change since the last reviews, that
being an upgrade to the hard drive. The major parts in this machine are:

• AMD XP2100+ Thoroughbred “B” cored CPU
• Abit NF7-S v2.0 motherboard
• 160GB (8 MB cache) Maxtor ATA133/7200 RPM Hard drive (IDE)
• BFG Technologies (nVidia) Geforce 4 Ti4200 (4x AGP, 128MB ram)
• Custom made watercooling system built by me

Since the system build article I just linked to was written, I’ve changed over to a DangerDen Maze 4 waterblock. At this time, the northbridge chipset block has been taken out,
replaced by a Swiftech MCW50-T on the graphics card GPU. I also upgraded the PSU from the 420W unit mentioned, to a 550W unit.

In testing these sticks of memory out, I used three commonly used benchmarking suites:

• SiSoft Sandra 2004 v2004.2.9.104 Memory Benchmark
• Futuremark’s PC Mark 2002 **
• Futuremark’s PC Mark 2004

** Futuremark has discontinued support for PC Mark 2002. It is still available for download however, at Futuremark’s website.

{mospagebreak}

• 100% tested at 200FSB (DDR400) on high performance motherboards
• SPD programmed at JEDEC standard values
• Tested at 2.5-3-3-6 on AMD systems
• Integrated aluminum heat spreader for improved thermal performance
• Benchmarked over multiple chipsets, processors, and motherboards
• Lifetime warranty

While the latency rating for these sticks is higher, I was still able to get them to run at 6-3-2-2 at 200FSB. I did have to relax those timings much sooner than the
newer LLPRO version however, as we will see below.

The methodology I used in testing these out was, that I raised the voltage to them only as a last resort, to try to remain stable, as I increased the FSB.

All runs were done at a 1:1 ratio with the CPU (synchronously). I’ll notate as we go along anywhere I needed to change any settings. I usually run the XP2100+ at a multiplier of
11x. As the FSB went up, I had to back that down to 10.5x to keep the CPU stable. I also wound up increasing the CPU core voltage at one point, as well.

All of the tests were run a minimum of three times through, frequently more. Basically, I ran them until the numbers stopped going up, and either leveled off, or a
run produced a lower result than the previous run did.

All runs were also done in Dual Channel mode, with the sticks installed in DIMM slots #2 and 3.

Business as usual, for both sets of ram here. Note that the 3200C2 is actually running at tighter than spec latency settings of 6-3-2-2.

Pretty even, across the board, which it should be, with both sets running identical specs.

Note: While the voltage was set in the BIOS to 2.6v (the lowest setting available), both Sisoft Sandra and Motherboard Monitor 5 showed it at 2.72v. It appears the Abit
NF7-S overvolts the ram slightly.

The older C2 revision got shaky running the tight timings at 205, and I had to back the settings off a bit. After doing that, it ran 205 very stably.

The LLPRO sailed through these tests handily, with no changes.

Here we start seeing the two revisions breaking away from each other, due to the latency timings used. Note also, that the CPU score in PC Mark 2002 was also
effected by this change. Oddly, the PC Mark 2004 scores don’t reflect the different settings here at 205 FSB.

No settings changes here, other than the increase in FSB.

Noteworthy here is that the C2 scores at 207 are just about equal to the LLPRO scores at 205, meaning the difference in latency is roughly equivalent to 2 FSB. We start
seeing a difference in the PC Mark 2004 scores here, as well.

No settings changes here, other than the increase in FSB.

Here we can see the difference in latency increases the margin in the scores further with the higher FSB setting. The tighter timings are now about equal to 3 FSB (the C2 scores at 210 equal the LLPRO scores at 207).

Things did start getting a bit unsteady here at these settings, but that was due to my CPU getting close to maxed out. I know it peaks somewhere around this speed, and it showed in this round of tests.

Both sets of ram ran the tests fine, but afterwards, when just running normal applications using the C2, I got kicked out of Unreal Tournament 2004 to the desktop after about 30 minutes.
The LLPRO didn’t do this, but was acting quirky, with things loading slowly, etc. No applications crashed, but it was obvious something wasn’t “right”, either.

The LLPRO is still running at default latency settings here at 213FSB.

** The instability here wasn’t a result of the memory settings, but rather the CPU hitting it’s limit at the settings used for it.

Interestingly, the C2 scored higher in a few of the tests here, for reasons we’ll see next (below).

215 FSB required a host of changes to run, not only for the memory, but the CPU as well. The system blue screened until I raised the CPU voltage to a higher setting, so I pushed
it up to 1.9v. While this got the system to boot properly, things were still unsteady after that.

The C2:

With the bump in CPU voltage, I was able to get through the tests the first time alright, but was kicked out of UT2004 instantly, back to the desktop. At this point with the C2,
I had to drop the CPU speed via the multiplier back to 10.5x (at 215 FSB). I reran the tests, and PC Mark 2004 would not finish, kicking me out in the DiVX Rendering test.

I then raised the voltage to the DDR for the first (and only time) to 2.7v (Sandra and MBM5 reported this as 2.83v). At these settings it was extremely stable.

The LLPRO:

At 215FSB, the LLPRO finally said “No MAS!” to the tight latency settings, as the machine refused to boot up at them. Dropping the timings to 7-3-3-2.5 regained stability with the
LLPRO, at an 11x multiplier. I didn’t have to change the CPU speed with the LLPRO at 215, the CPU voltage and latency changes were enough to keep it stable.

Note the extra entries below regarding CPU speed.

With the latency settings back to equal, the scores in Sandra 2004’s Memory Bandwidth tests reflect this change, as they even out again.

The scores in both Futuremark suites reflect the differences in CPU speed above, with the C2 scores taking a plunge due to the lower CPU speed.

{mospagebreak}

At this point, I had to drop the CPU multiplier with the LLPRO, as it again would not boot. This instability was entirely a CPU thing, as going to a 10.5x multiplier
resulted in a very stable system. Note however, it’s still running the default voltage (or at least, the lowest I could set it for (2.6v)).

Once again, pretty even, across the board, which it should be, with both sets running identical specs.

This was the last fully stable set for the C2. The LLPRO was totally stable here, once the CPU speed had been decreased.

With the CPU speed again cresting 2.3GHz, things stated getting skittish again testing with the C2. It started out by loading WinXP Pro a bit rough, and wouldn’t finish the PC Mark
2004 suite, locking the machine up hard.

I then tried changing the timings for the C2 to 8-4-4-3, which wouldn’t POST. I then went back in, put the timings back, and tried 2.8v to the memory. This booted.

Once I got into Windows, I immediately ran the PC Mark 2004 suite again, which it again didn’t finish, with Internet Explorer crashing in the Web Page Rendering test.

I finally tried dropping the CPU multiplier to 10x (10 x 220 = 2200MHz), which I really didn’t want to as the scores would take a nose dive across the board, basically hitting
the wall of diminishing returns. I tried it nonetheless, and it again failed in the Web Page Rendering test in PC Mark 2004.

The partial scores shown here are from the incomplete set when it locked up hard in PC Mark 2004.

The LLPRO was completely stable at these settings.

At this point, the C2 was all in, as I couldn’t complete the benchmarks at all, despite increasing the voltage, relaxing the timings, and reducing the CPU speed yet further.

Whereas the C2 had so much difficulty with these settings, I ran the LLPRO here for a full day after running the benchmarks, just doing my usual things.
It never once showed any signs of instability at this speed.

The above screenshot was taken with Arthur Liberman’s GCPUID utility, which can be found HERE.
Look for a review of this utility, and other applications of Mr. Liberman’s in the near future.

No settings changes here for the LLPRO, other than the increase in FSB.

This set was the last fully stable set with the LLPRO. This indicates to me that more than likely the motherboard is holding me back once I pass this high a FSB setting.

At 223FSB, the machine started acting flaky when booting into Windows. It only completed part of the tests, failing in the PC Mark 2004 Web Page Rendering test again.

At this point, I tried rasing the voltage to the LLPRO (the first and only time I did this), to no avail. It again failed at the same test in PC Mark 2004.

I’m uncertain whether or not this is my CPU, or motherboard stumbling here, but I’m guessing motherboard, judging by the way things transpired with the C2 at 220FSB, where I
lowered the multiplier and it was ineffective.

The computer was stable at this speed (2340MHz) at a different multiplier and FSB combination, which really leads me to believe it to be the motherboard signing off for the
evening at about 222FSB.

While incomplete, these numbers were the highest I achieved in the Sandra 2004 Memory Bandwidth and PC Mark 2002 tests, by a very wide margin (even over 222FSB).

It gave me the impression that this ram was just getting started, as the rest of my system decided it had had enough.

225 FSB (DDR 450) x 10.5, 1.9v CPU, 2.7v DDR, 7-3-3-2.5 blue screened before ever seeing Windows.

Coupled with the fact that at DDR446, I was still running at the lowest available voltage, and fairly tight timings of 7-3-3-2.5. I feel I left a lot of this memory on the table,
unable to fully wring it out with this system.

I’d be very curious as to how it would fare in an 800FSB P4 system, which I think would truly give this memory a run for it’s money.

On the back of the TwinX1024-3200LLPRO packaging, are a number of quotes from previous reviews of Corsair products. One of them, taken from a review at
Bit-Tech.net, the reviewer states:

“The problem with this Corsair XMS Memory is not how far you can push it, but finding system components that can keep up with it!”

(Macroman, Aug 3, 2002, in a Bit-Tech.net review of Corsair XMS PC3200 memory)

This seem to be the exact issue I faced here, with these sticks of PC3200LLPRO. I feel certain they had more to give, but the balance of my system just wouldn’t carry them there.

This product impressed me, as I was able to push my system farther than it had ever gone previously, getting higher benchmark scores than I’d never seen before.

The embedded Activity LEDs, while at first glance seem kind of gimmicky, function very well. Depending on what you’re doing with the PC, you really can tell how much memory
is being accessed by way these light blink on and off. And, if you happen to push your system a bit too hard, they’re a quick indicator of a system lockup. If they’re not blinking at all,
you can be pretty sure something is amiss.

If you’re running an AMD Socket “A” based system, this would make a fine choice for your memory needs. Indeed, with any system really. With a Socket “A” system, this memory will likely
outpace the rest of your system. A Pentium 4 or AMD 64/FX based system would likely push this product further than I was able to here.

Their older CMX-512 PC3200C2 is a very good product as well, but was outperformed by the newer revision in these tests.

I’d like to thank Corsair for sending this memory kit out for us to evaluate today.

Cheers!

Email Brian