Corsair XMS2 DDR2 TWIN2X1024-4300C3PRO Review

Memory by KeithSuppe @ 2005-02-20

Corsair has tightened the reigns on DDR2 latencies reducing seek times below that of many high-speed DDR kits. Able to run at CL3-3-3-6, Corsair?s 4300C3PRO has brought hope to those who believed the only way to extract performance from DDR2 was to run the memory at super high clock speeds and hope for bandwidth. Does Corsair?s 4300C3PRO perform and overclock without the need for insanely high latencies? Read on to find out.

  • next

Introduction

Introduction

Corsair continues its legacy providing the PC-world with high performance cutting edge memory designs. They are truly a company which dares to go where no others have gone before, and the announcement of their XPERT memory is evidence of this goal. Corsair has always made it a priority to appeal to that niche market segment of overclockers and enthusiasts who demand much more from their memory then JEDEC standards stipulate. Today we'll evaluate a true enthusiast part: Corsair's 4300C3PRO, it doesn't measure temp as EXPERT does, however; it features real-time LED bank activity. Most importantly this latest DDR2 part brings latencies down to CL3-3-3-6 which brand this memory with some of the tightest timings of any DDR2 to date.

Madshrimps (c)


With so many recent coverts to the overclocking community, manufactures have tried to make overclocking and BIOS tweaking as simple as possible. This manifests itself in software such as MSI's Core Cell, and Abit's µGuru the latter featuring a specialized controller chip made for just this purpose. These utilities offer a plethora of options including voltage monitoring/adjustment, fan-speed monitoring/adjustment, and of course CPU monitoring/frequency adjustment all from within the Operating System.

While these programs can do much of what was once done only in the BIOS, they do have their limitations. Some do not offer the ability to adjust memory timings, and of course timings (latencies) are an integral ingredient in the overclocking recipe. I was pleasantly surprised while testing this memory to discover that the SPD (Serial Presence Detect) maintains the CL3-3-3-6 timings even while overclocked. Someone who doesn't have a wealth of experience could simply avoid the Advanced Chipset menu option leaving timings on Auto, or SPD and still successfully overclock.

Madshrimps (c)


Test Setup & Overclocking

Test System
CPU Intel P4 550 Retail-3.4GHz LGA 775
Mainboard Abit AA8
Memory Corsair 4300C3PRO 1024MB kit (SPD)
Graphics Card ATI Radeon X800 XT
Power Supply OCZ Power Stream 420W
Operating System Windows XP SP2
H20 System Alphacool Xtreme Pro Set utilizing NexXxos BOLD for S775


I ran series of benchmarks first at default system speed (3.4GHz/200FSB) and then repeating the series at the highest stable system overclocked speed (4.0GHz/240FSB). I ran the memory at a 1:1 ratio in all tests and at 1.95V while overclocked. Benchmarks included Everest v1.51, SuperPi, Sandra 2005, ScienceMark, CPUBench2003, DOOM III, and FarCry. To exemplify the memory's ability to hold its timings regardless of FSB speed, I've provided a few screenshots. First at 3.4GHz 200FSB/1:1, and the second at 4.0GHz 240FSB/1:1:

Madshrimps (c)

Madshrimps (c)


While I was able to run most almost every benchmark to 250FSB (4.2GHz) in 1:1 ratio, Far Cry crashed half way through the demo regardless. While I did raise Vcore, VDDR, and NB voltage, there's no way to raise video card voltage, nor lock the PCIe bus successfully. To date I'm unaware of any i915X/i925X chipset based PCIe 16X motherboard which offer graphic card voltage adjustment, or supply a BIOS able to lock the PCIe bus without errors. The Abit AA8 provides PCIe adjustability; however this results in a system crash whether done in Windows through µGuru or in the BIOS itself.

I'm convinced in our case the inability to bench FarCry/Doom3/PCMark04 at 250FSB is graphics card related, as I've experienced the same problem repeatedly with higher rated memory (PC2-5400) memory installed. I've supplied a Sandra bandwidth benchmark at 250FSB, and will include other successful 250FSB benchmarks as well given the circumstances.

Madshrimps (c)


Holding their latencies at 250FSB is quite an accomplishment, and were it not for BIOS/motherboard limitations I've no doubt we could have taken this memory a notch higher.

Benchmarks

Madshrimps (c)


Next the Lavalys benchmark replacing Aida32 Everest v1.51 which includes a latency measurement:

Madshrimps (c)

Madshrimps (c)


The memory holds up well overclocked at 250FSB at 1:1 aspect ratio this is a significant increase over its default speed.
  • next