Corsair TwinX PC4400 Dual Channel Memory Review

Memory by jmke @ 2004-01-29

Corsair Micro has released what is currently the fastest DDR on the market. Their Twin-X4400 (DDR550) is ideally suited for Canterwood based 875 chipset platforms. We test them on a battery of motherboards to find out how far this memory can push your system!

  • prev
  • next

Liquid3D @ 275Mhz

Intro by Liquid3D:

Corsair micro has released what is currently the fastest DDR on the market. Their Twin-X4400 (DDR550) is ideally suited for Canterwood based 875 chipset platforms, specifically the Asus P4C800E-Deluxe motherboard. Although Corsair recommends the P4C800E-Dlx, I have confidence there may be other 875/865 motherboards capable of running the memory to speed. To my knowledge Intel's Canterwood or Springdale chipset's happen to be the only chipset's capable of running such speeds. I've owned five 875/865 motherboards capable of attaining 300FSB speeds, and several easily ran 285FSB at 1:1 ratio, including the Abit IC7, and IS7. Unfortunately I'd recently donated my Abit IS7, and IC7 or I'd be comparing these as well. The motherboard Corsair tested, and guarantee’s this memory to run at DDR550, is the venerable Asus P4C800E-Deluxe. This is an Overclocker’s favourite, and my Rev. 1.0 02 has been a very stable overclocking platform. This will be the motherboard used throughout

Madshrimps (c)

The fact Hynix is producing such high quality DRAM bodes well for Overclocker's, and Hynix. In the conservative world, the Art of overclocking prompts passive-aggressive responses among chipmakers. While Intel certainly dissuades overclocking, it's ironic the fastest memories on the market, are dependent upon their chipsets. There is quite a large market for overclocking and its accoutrements, however; to manufacturers as large as Intel it's a mere pittance. I remember a chill running up my spine reading in the Inquirer not too long ago, an Intel employee had come up with a solution to deter overclocking. Similar to clock throttling, it monitored the FSB, and whenever default was exceeded it simply throttled back. We can only hope companies such as Corsair will continue to meet the needs of the growing segment of the market, known as PC-Enthusiast's or Overclocker's. This is an ideal time to overclock, and I'm reminded of Thoroughbred-B 1700 days, where the lowest model sharing that core, easily attained the highest model's default speeds. The 2.4C shares this capability, and most steppings can attain 3.2GHz speeds at default Vcore. I find it ironic, while the introduction of Barton's cache put an end to that era, so the Intel 3.2EE and Prescott, may put an end to ours. We have some time yet to enjoy ourselves, and with companies such as Abit introducing their AI7 with its "guru" overclocking hardware/software perhaps mated with Corsair Twin-4400. I certainly hope the Prescott will begin the cycle over again, not prematurely end it. Of course I'm an optimist. So long as Corsair delivers the goods we'll be able to take these processors and chipsets to their limits.

Other reviews have endeavoured running this memory at lower CAS ratings down at the default speeds of the chipset. I chose to eschew such tests, as this memory is meant for overclocking, and overclocking at seriously high FSB speeds. These are the reasons behind the higher CAS latencies, and given what Corsair has to work with, I think we should stop analysing the Horses' mouth. Especially since we already know what's in there. Those wanting 2-2-2-5 at 550MHz are going to be waiting as much as a year in my estimation, and perhaps longer. No DRAM makers are making IC's capable of running such parameters. Yes there have been a few Overclockers who have achieved this. In the process they’re pushing over 3.1V into the modules. With newer DDR only requiring 1.8V at the core, your pushing the I/O buffers into conniption fits. No memory is going to last long, or last long without many errors, and misses under such conditions. This is akin to placing a Nitrous Oxide system in an Aston Martin? You already have a vehicle built for high performance, at that point its modifying abomination. What we have in Twin-X4400 is the Aston Martin for the Canterwood autobahn. I'm choosing not to overclock, overclocking memory through tightening the latencies. I must stop the insanity. Without further adieu I introduce to you the Corsair Twin-X4400.

Madshrimps (c)


These first benchmarks are SiS Sandra memory, in Buffered mode. I ran the System at 275FSB 1:1, PAT in Standard, PAT Enabled in chipset. I used the SPD setting as I had the best results with this setting. In fact I've found the gist of high speed-DDR designed for the 875/865 chipset's runs best using the SPD setting. VDIMM was 2.85V, and CPU Vcore was default at 1.525V. Unfortunately there have been some Vcore inaccuracies, or actual over-volting issues with the Asus P4C800E-Deluxe. Many P4C800E-Deluxe owners have reported this anomaly. Fortunately, (to a certain extent) Vcore has a propensity to drop the further one overclocks:
Sisoft Sandra 2004 Memory Benchmark @ 275 MHz SPD (CL3 4-4-8) - 6280/6344Mb/s

These modules are certainly bandwidth capable, and there are few system which surpass this level. To emphasize the difference between Manual @ CAS-3-4-4-8 and SPD, the benchmarks below, were run on Manual settings CAS 3-4-4-8, PAT on Standard, and Enabled in Chipset. Once again voltage was 2.85V, and CPU remained at default, or 1.525V.
Sisoft Sandra 2004 Memory Benchmark @ 275 MHz (CL3 4-4-8) - 6231/6252Mb/s

Albeit a 49MB/s decrease, I do believe there are circumstances on Springdale/Canterwood platforms, under which it's beneficial to run under Manual (CAS 3-4-4-8). This is particularly true in 3D applications/benchmarks. For pure bandwidth, SPD seems to run more stable on Canterwood/Springdale platforms specifically when running high-latency / high-speed Dual-channel memory. The gists of benchmarks in this review were run on SPD, unless indicated. Next running in Unbuffered mode, which in SiS Sandra 2003 involves disabling 13 settings in the module. Once again I ran 275FSB, under SPD PAT enabled, and on Standard. Voltages remain the same.
Sisoft Sandra 2004 Unbuffered Memory Benchmark @ 275 MHz (CL3 4-4-8) - 3370/3384Mb/s

I then decided to drop the FSB speed to 250FSB, for those whom may not want to push their system (and processor) to such levels. Often when overclocking there is a "sweet-spot" where the system performs better than when pushed to the max. This first SiS Sandra was run at 250FSB, SPD, PAT enabled in Chipset. The CPU Vcore remained at default of 1.525V, and I left the VDIMM at 2.85V. Most likely I could have dropped the VDIMM, however for stability's sake I decided not to test this.
Sisoft Sandra 2004 Memory Benchmark @ 250 MHz SPD (CL3 4-4-8) - 5755/5765Mb/s

Sisoft Sandra 2004 Unbuffered Memory Benchmark @ 250 MHz SPD (CL3 4-4-8) - 2971/3048Mb/s

On to the next page for some 3D benchmarks ->
  • prev
  • next