Corsair TwinX1024-3200XLPRO

Add Your Comments

“Corsair has produced an extremely excellent product” – Brian Berryman

SUMMARY: Corsair ups the ante by lowering the latency

It’s Low Latency Limbo time! How low can you go?

Label

The good folks at Corsair’s were kind enough to send their new
TwinX1024-3200XLPRO DDR kit to test out today.

When I recently reviewed Corsair’s PC3200 LLPRO memory, I’d mentioned that I had read about a number of people running
that 6-2-3-2 (default) spec DDR at 5-2-2-2. Apparently, Corsair has gotten enough of a yield at these settings to offer it as a whole new flavor.

Yes, this is the LLPRO shot from last article. I’ve reused it as they visually are identical in packaging and product to the XLPRO. It’s what’s
UNDER the heat spreaders that separates the two.

The TwinX1024-3200XLPRO is the top of the line set from Corsair’s new XL XMS Series. The XLPRO reviewed here includes the Activity Lights also found on the LLPRO
that I reviewed recently. It’s the only XL Series kit that has them, the others have the older style heat spreaders. The remaining four XL Series products are as follows:

  • TwinX1024-3200XL (2 x 512MB, black spreaders)
  • TwinX1024-3200XLPT (2 x 512MB, platinum spreaders)
  • CMX512-3200XL (single stick, black spreader)
  • CMX512-3200XLPT (single stick, platinum spreader)

All of the new XL Series run DDR400 at 5-2-2-2, default. From what I’ve read,
they’ve used Samsung TCC4 (4 ns chips) on these new products.

At default. (SiSoft Sandra 2004) I’ve “highlighted” the applicable lines.

They mention at Corsair’s website these XL Series products are capable of 500MHz (DDR250). If so, I’m gonna need a bigger boat. My P4 2.8C CPU I’ll be testing these with won’t go 250.

Included

Included with the XLPRO here is an invitation to sign up for Corsair’s monthly newsletter, and a window decal.

When I tested out Corsair’s LLPRO previously, I was amazed at the results I received. It blew my AMD XP2100+ @ 2200 system out of the water, besting it’s FSB ceiling easily. It
wasn’t until I got my Intel P4 system (used here) together shortly thereafter that I was able to tame the LLPRO.

Is Corsair poised to repeat with the XLPRO, and sink another system of mine? Let’s find out.

The Test System:

UV
  • Intel P4 2.8C Northwood core (w/ Hyper-Threading)
  • DFI LanParty 875B motherboard
  • 160GB (8 MB cache) Maxtor ATA133/7200 RPM Hard drive (IDE)
  • Sapphire (ATi) Radeon 9800 Pro (flashed w/ XT BIOS) 256MB DDRII
  • Vantec AeroFlow HSF, with Arctic Silver 5 TIM used

A pair of changes since the last review, those being getting my 9800 Pro back from RMA since then, and moving my good hard drive over into this system. These replaced the
GeForce4 Ti4200 (4X, 128MB) and Maxtor 20GB, 2 meg cache hard drive I used previously.

While the hardware changed slightly, the benchmarks I’ll use will not;

Benchmarks Used:

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

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

All runs were done at a 1:1 ratio with the CPU (synchronously). 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 with the memory installed
and running in Dual Channel mode

A slight change in the test method

Last time out, I tried to see how far I could go before raising the voltage to the memory in the BIOS. I’m departing from this and going a different route this time; From the
second set of benchmarks on here, the voltage in the BIOS will be set to 2.9v, maximum for the DFI LanParty 875P.

The reasoning here is this DDR is designed to run at very low latencies. This way we’ll find out how far the machine can go and stay at these low settings.

Those inclined to voltmodding their motherboards will likely yield slightly better results than I will.

Running the benchies!

For this first set of benchmarks, everything is set to default specs. This gives us a baseline to refer back to, as we progress along.

For reference also, I’m including a set numbers from the LLPRO tested previously. These, also at default, are taken using the same P4 system configuration the XLPRO used here.

200 FSB (DDR400)

CPU @ 14 x 200 = 2800MHz

How much of a gain at default settings does this system get with the lower latencies of the XLPRO?

LLPRO

XLPRO

Latency settings

6-2-3-2

5-2-2-2

Voltage

2.6v

2.6v

Sandra 2004 Memory Bench Integer ALU

4870 MB/s

4909 MB/s

Sandra 2004 Memory Bench Float FPU

4878 MB/s

4920 MB/s

PC Mark 2002 CPU score

6983

7010

PC Mark 2002 Memory score

9341

9411

PC Mark 2004

4444

4459

Ran stable?

Yes

Yes

As the numbers above show, there is an increase across the board with the tighter timings available with the XLPRO. What amazed me was how far I was able to get this DDR400 to
run at 5-2-2-2.

{mospagebreak}

Continuing on, the benchmarks that follow are only from the XL Series memory. I’ve raised the voltage in the BIOS to 2.9v, to see how far I could get these ultra low
settings to go.

210/215 FSB (DDR420/430)

CPU @ 14 x 210 (2940MHz) and 215 (3010MHz)

The numbers above are what I set things up for in the BIOS. The DFI LanParty 875P has a quirk where it boots a couple FSB higher. The following scores were actually taken at
212 and 217FSB respectively. Again, not quibbling, just noting this. 😉

210 FSB

215 FSB

Latency settings

5-2-2-2

5-2-2-2

Voltage

2.9v

2.9v

Sandra 2004 Memory Bench Integer ALU

5200 MB/s

5336 MB/s

Sandra 2004 Memory Bench Float FPU

5203 MB/s

5326 MB/s

PC Mark 2002 CPU score

7402

7554

PC Mark 2002 Memory score

9908

10176

PC Mark 2004

4668

4745

Ran stable?

Yes

Yes

Both of these settings ran flawlessly. At 215 (217) FSB, the 10K barrier falls in the PC Mark 2002 score.

218/220 FSB (DDR436/440)

CPU @ 14 x 218 (3052MHz) and 220 (3080MHz)

These also booted higher, at 219 and 221 FSB respectively.

Here is where I first encountered instability. I had been making jumps of 5 FSB, and going right to 220 at 5-2-2-2 resulted in the machine booting, but blue screening instantly.

I backed off to 218 at 5-2-2-2, which seemed to run fine, but too, wasn’t 100% stable. It took some time, but after a lengthy gaming session while running the Folding @ Home client
in the background, I was kicked out of the game back to the desktop.

I went in and raised the voltage to the CPU a step, and tried again at 218. I received the same results. Close, but not completely stable.

I’ve no doubt 217 FSB would be 100% stable at these settings. After this, I shoved the CPU voltage to 1.65v, where it stayed for the remainder of the benchmarking that follows. This, while
far higher than I need to usually run higher FSB than these, removed thoughts of the CPU causing instability for a while.

218 FSB

220 FSB

Latency settings

5-2-2-2

5-2-2-2

Voltage

2.9v

2.9v

Sandra 2004 Memory Bench Integer ALU

5373 MB/s

BSOD

Sandra 2004 Memory Bench Float FPU

5359 MB/s

PC Mark 2002 CPU score

7625

PC Mark 2002 Memory score

10201

PC Mark 2004

4797

Ran stable?

Almost

No

I didn’t get kicked right out of the game I was playing, it took a good half hour to forty five minutes before it happened. I happened more than once, with more than one game
(this happened in both UT2004 and EA Sports MBL03).

I then tried 220 FSB with latency settings of 6-2-3-2. This was stable. However, the numbers I got with the timings being relaxed were only slightly over the numbers at 218, with
the difference between them being less than the difference between the LLPRO and XLPRO at default, as seen on the first page.

225/230 FSB (DDR450/460)

CPU @ 14 x 225 (3150MHz) and 230 (3220MHz)

At 225 FSB, I encountered a little more instability, and had to relax the timings slightly further to 6-3-3-2.

225 FSB

230 FSB

Latency settings

6-3-3-2

6-3-3-2

Voltage

2.9v

2.9v

Sandra 2004 Memory Bench Integer ALU

5486 MB/s

5599 MB/s

Sandra 2004 Memory Bench Float FPU

5485 MB/s

5611 MB/s

PC Mark 2002 CPU score

7873

8063

PC Mark 2002 Memory score

10296

10528

PC Mark 2004

4958

5062

Ran stable?

Yes

Yes

With the latency set at 6-3-3-2, these two settings ran perfectly. As I started getting higher with the FSB setting, I started running at these settings for longer periods
of time after finishing the benchmarks.

Note the difference between 225 here, and 218 at the top of this page. With the change in latency, the memory scores (Sandra, and the memory score in PC Mark 2002) show a narrow
difference between them, despite being 12 FSB apart.

235/240 FSB (DDR470/480)

CPU @ 14 x 235 (3290MHz) and 240 (3360MHz)

Still running at 6-3-3-2 here.

235 FSB

240 FSB

Latency settings

6-3-3-2

6-3-3-2

Voltage

2.9v

2.9v

Sandra 2004 Memory Bench Integer ALU

5730 MB/s

5891 MB/s

Sandra 2004 Memory Bench Float FPU

5729 MB/s

5855 MB/s

PC Mark 2002 CPU score

8252

8421

PC Mark 2002 Memory score

10735

10974

PC Mark 2004

5185

5255

Ran stable?

Yes

Almost

Where the LLPRO had run out of gas at 233 FSB, I ran at 235 FSB with this DDR for more than a full day, to try and make it crash. It didn’t. Everything I threw at it ran perfectly.

At 240 FSB however, I ran into a similar situation as I encountered at 218. It took about an hour, but I got bounced back to the desktop in UT2004. This was the only thing that
it would happen in, everything else ran fine. I’ve since logged a few hours at lower speeds playing UT2004, and it hasn’t happened since.

Very impressive for PC3200 (DDR400)

At 240 FSB, I did achieve a personal milestone;

I finally break the 20K barrier!

I had flashed the 9800 Pro to an XT just prior to getting the XLPRO to review, and hadn’t had the chance to really overclock it and see what happens. So this is technically at
default settings for the video card, considering that for all intents and purposes, it’s now a 9800XT.

I got 6695 in 3D Mark 2003 here. Still chasing 7K in that one… 😉 I’ve yet to overclock the R360 or the DDRII much past XT speeds though (I did this for a few days, before flashing, to ensure
it would run there stable at 412/365).

{mospagebreak}

245 FSB (DDR490) and beyond

At 240, my luck had started to run out. I did get 240 stable by relaxing the timings yet further, with the expected reduction in benchmark scores over what I posted last page.

With that, I moved to 245 FSB, where something just ran out of gas. 7-3-3-2.5 would make it through POST, but not boot into Windows. Going all out to 8-4-4-3 got me into Windows XP.

The Point Of Diminshing Returns

For the brief time I was in windows at 245 FSB, things seemed shaky. I made it through the Memory Bandwidth benchmark in Sandra 2004, and was kicked right to the desktop as soon as I tried
to run PC Mark 2002.

The scores I did get in Sandra were less than what I had gotten at 240 FSB / 6-3-3-2.

This is right in the ballpark where my 2.8C P4 signs off. With the memory running at less than 1:1, I can get the 2.8C to POST, but not boot into Windows at 250.

To go from slightly unstable at 240/6-3-3-2 to very unstable at 245/8-4-4-3 sounds like it hit a wall awful fast… *Brian raises eyebrow at his P4*

Why do I have the feeling Corsair has outgunned another of my machines?

CONCLUSIONS

Conclusions

Once again, Corsair has produced an extremely excellent product. With the TwinX1024-3200XLPRO, I was able to get my system to run far faster than I had ever before.
I was able to get about a 20% overclock out of this DDR, before I really had to back off on the settings.

If the numbers carry over to my AMD machine, I would have maxed out that system with this DDR barely breaking a sweat. The LLPRO outpaced that machine at 223 FSB at very low settings.

And the XLPRO is a step beyond that.

Corsair is a brand I’d recommend to anyone looking for some new memory. This XLPRO only furthers the legend. It’s not difficult for me to see how they’ve won as many awards as they have.

I’d like to thank Corsair for sending this out to us to have a look at here today.

Cheers!

Email Brian

Leave a Reply

Your email address will not be published. Required fields are marked *