has been a supplier of memory for approximately 16-years, although it's only been the last year or so we've had any real exposure to the product here in the USA. While ethnocentric-marketing often precludes us from most of the products available, ethnocentrism isn't a negative term in and of itself, its egocentrism which concerns me.
This statement may seem prima facie
, however; many believed the Internet would allow a freedom to transcend once geographical demarcations and personal prejudices, yet such encumbrances still exist. As an independent tester, obtaining sample CPUs, graphic-cards and motherboards is about as easy as having AMD and Intel forgo a pre-nup or the likelihood they would get to that point.
It's frustrating to see other sites not much larger then [M] whom are actually courted by company Rep's and frankly such preferential treatment often leads to complacency. Sadly, as a result of having such unlimited access to hardware Reviewers will often go to print with some mediocre articles. Those of us whom are most dedicated, who will do far more research on the product, and write to captivate as well as educate the reader, are left by the wayside simply because of favoritism. Still there are some sites out there who retain a level of journalistic integrity which reflects the erudite nature of this industry. Sites such as X-Bit Labs, Xtreme Systems, Sudhian, Adrians Rojak Pot, LowYat, Lost Circuits, AMDBoard and especially VansHardware where I've found some of the most astute and infromative articles to date. Needless to say were not here to discuss the bureaucracy of the semi-conductor industry, nor the politics’ of sample distribution. We're here to discuss a product which, for all intensive purposes may have slipped under the radar.
G.Skill has labeled each 256MB stick with the following: 256MB DDR400 2-2-2-5 PC3200, DDR600 3-4-4-8 PC4800
. Simple, perhaps somewhat confusing for the inexperienced end-user, however most importantly the product lives up to it's name. As you’re about to see, these little naked PCB's pack quite a performance punch. G.Skill uses Samsung TCCD memory and seems to have their hands in some "special" bins.
Samsung has been doing some amazing things with DDRAM ICs and their TCCD may go down in history as having some of the most extensive frequency range as well as flexible timings.
It's a refreshing change not to worry about removing a heat spreader, which are in most cases superfluous and sometimes actually detrimental to performance. This is due to double-sided "thermal" tape whose name is somewhat duplicitous as more often then not this tape traps more heat then it dissipates. Were it not for cost considerations perhaps more memory suppliers could use thermal-epoxy or thermal-paste between IC and heat spreader, the solution would be much more effective then tape. Such extremes would only be necessary in the case of UTT spec'd memory or other such high voltage parts running at or above 3.0v. Voltages this high would surely benefit from such a simple solution and in all other cases eschewing heat spreaders which are widely over-used, may be more practical. G.Skill performed best at 2.8v VDIMM which did not require any "active" cooling.Testing
|CPU ||A64 3500+ Winchester .09-micron (process) core Retail|
|Mainboard ||DFI Lanparty NF4 Ultra-D (BIOS 310)|
|Memory ||G.SKILL DDR600 2x256MB|
|STORAGE - HDD ||Maxtor DiamondMax Plus 9 120GB SATA150 / Maxtor DiamondMax 10 300GB Native SATA150 |
|Graphics ||BFG 6800GT-OC (370/1000MHz GDDR3)|
|Power Supply ||Thermaltake PureSilentPower 680|
|Cooling ||PolarFLO TT (copper) Hydor L40-II (2800 LH / 230cm MAX-H) Danger Den Dbl Heater Core 2xSunon 120mm/90CFM (1/2”)|
|Operating System ||Windows XP SP2|
During testing I ran the memory at its advertised speeds and timing under SPD, which is my usual methodology prior to making any manual BIOS adjustments in the Advanced Memory setting menus. I was pleasantly surprised to find the memory ran at CL2.5-4-4-8 at its DDR600 spec, however; dropping the FSB speed to 200MHz (DDR400) also resulted in CL2.5-4-4-8 timings.
As seen below there's a slight discrepancy in the SPD. On the Left:
200FSB SPD - CL2.5-4-4-8, Center:
(manually set) 210FSB - CL2-2-2-5, on the Right:
300FSB SPD - CL2.5-4-4-8. All settings were maintained throughout at 2.8v (VDIMM) and the memory was run in Dual Channel.
To commence testing I forgo JEDEC DDR400 (as this is primarily an Enthusiast part) leaping to 210FSB (DDR420) which was the highest stable frequency at any voltage or BIOS combination allowing for CL2-2-2-5 timings. At 300FSB (DDR600) speeds I simply ran the system at SPD and found the memory ran at CL2.5-4-4-8. In fact attempting to manually adjust settings to CL3-4-4-8 resulted in instability? Once again this was the highest stable frequency obtainable at any voltage or BIOS setting (within reason).
I attempted lower latencies at DDR600 but was unsuccessful. I imagine with some time invested these modules would run beyond DDR600 speeds (check out this guide to get to DDR600+ on a DFI board
) as I did load into Windows up to DDR660, but was unable to remain stable in 3D applications. Pushing more voltage into these modules resulted in lower performance overall and the ideal VDIMM seemed to be 2.7 ~ 2.8V based on the idiosyncrasies of the DFI motherboard (BIOS version 310).
These modules performed rock solid for over two weeks in my system running 24/7 and this was the longest time I've ever run any system at 300FSB (HTT) without a random crash or hiccup. I was surprised playing Far Cry, which consumes RAM like America consumes Crude Oil, I didn't experience any crashes. This was a first for 512MB kit.
Onto our benchmark results ->