System Testing:

Testing was focused on validating ram for stability to the best of my ability and then benchmarking ram at rated timings. I would not feel comfortable publishing a review in which I felt stability was compromised, as this could mislead the reader into a purchase that was ill informed, with no fault to the reader. I will describe some initial impediments to high MHz testing that I encountered, and solutions that resolved my problems. To reasonably assure stability, I ran Memtest86+ which is a DOS based software program that attempts to find errors in memory implementation through a series of software tests. A hardware memory analyzer may be ideal, but I simply don't have access to such devices, or the resources to obtain such a product. In the end, it is left to each user on how they determine stability. Some people consider the ability to run SuperPi 1mb or 2mb benchmark and 3DMark2001 to equal stability. Some people believe that if it posts to windows it is stable; a very obtuse viewpoint in my opinion. I prefer to measure stability by what I am testing. For example, if I am testing limits of memory I find successful completion of 10+ passes of Memtest86+ (default/all tests) and 12+ hours of 3DMark2001SE in demo loop mode and I have found SuperPi 1mb benchmark to be sensitive and give me a reasonable model for predicting memory stability. I would be remiss to believe that my methodology is perfect, and as above I have stated that a hardware analyzer would be ideal, but for the purposes of this article I believe that results I will give should be reproducible, given equal quality in components.


Benchmarks:

Benchmarking focuses on applications that I feel would illustrate the performance of system memory. As such, I chose to include the following benchmarks/applications:

SiSoft Sandra 2004-Buffered memory scores

Cacheman

SuperPi 1mb

Futuremark's PCMark 2004

Futuremark's 3DMark 2001SE build 330

Doom3 Demo1 640x480 low quality


Testing Parameters:

As G.Skill PC4400-F1-3200DSU2-1GBLE memory is rated at either CAS 2 2:2:5 at PC3200 (200 MHz) specification and CAS 2.5 3:3:7 at PC4400 (275 MHz) I chose to run these benchmarks at default CPU speeds at each of these frequency and timing levels. A nice quality of the Athlon64 is the downward unlocked multiplier which allows the user to manipulate the CPU multiplier in 1/2 integer steps down from its native level. This allows us the ability to bring CPU multiplier down to 8x at 275 MHz HTT and stratify our speed to a default 2.2 GHz at both a 200 MHz HTT and at the 275 MHz HTT. Thus, any performance gains can be solely attributed to the higher HTT and memory MHz levels versus any gain in CPU speed. Finally, I chose to overclock the CPU to the maximum benchmark stable speed on air, which happened to be 2.75 GHz. This was a nice coincidence as the 10x multiplier allowed the CPU and HTT bus to fall right on 275 MHz and therefore allow testing at 275 MHz memory speed as well. As you can see, the 3500+ 0.09u Winchester core is no slouch when pushed to its air cooled limits. I hoped to soon vapor phase cool this CPU, but mounting limitations precluded me from doing so for this review.


System configuration:

JNavGT89's Test Setup
CPU	Athlon64 3500+ 0.09u Winchester core OEM ** Tested at 11x200 = 2.2 GHz(default vcore), 8x275 = 2.2 GHz, and 10x275 = 2.75 GHz(1.625v)
Mainboard	MSI K8N Neo2 Platinum Socket 939 (Beta Bios 1.3b6) ** HT set to 5X for 200HTT and 3X for 275MHz testing
Memory	G.Skill PC4400-F1-3200DSU2-1GBLE ** Timings used are CAS 2 2:2:5 for 200 MHz operation, and Cas2.5 3:3:7 for 275 MHz Operation. 1T command rate was forced and memory was run at 400 MHz setting in bios which forces memory to scale with HTT in a 1:1 ratio. Memory voltage was set at 2.65v for all testing
Other Components	Video: EVga Geforce 6800 Ultra 256mb 8X AGP-default clocks of 400/1100 Power Supply: Fortron Source 530W ATX

Onto the benchmark results ->