PC2-6400 DDR2 2GB Memory Kit Duel - Super Talent vs Crucial

Memory by KeithSuppe @ 2006-04-02

Today we bring two PC2-6400 heavyweights into the Sumo dohyo, Crucial Ballistix PC2-6400 and Super Talent T800UX2GC4. If you?re unfamiliar with DDR2 overclocking, DDR2 confutes the axiom smaller modules scale much higher then their larger brethren. Just how well are current BGA IC?s scaling these days?

The Wrestlers

The Contenders

Madshrimps (c)


Let’s meet tonight’s candidates:

Super Talent T800UX2GC4
(Official Website)


Madshrimps (c)


Super Talent states at their site, to have 20-years experience in the memory industry. I must confess until recently I haven't heard the name; however, this is more likely a result of my ethnocentrism the anything else. Living in the USA often precludes me from access to many brands sold on the European market and in Taiwan.

T800UX2GC4 PC2-6400 are described at Super Talent site as 800MHz DDR2 Over-Clocked Memory For Gaming Enthusiasts. 2GB (1GB x 2). IC compliment for Super Talent (2x) 128M x 64 , non-Reg., no-ECC , CL 4-3-4-8 , Dual Channel Kit of 2 , 8/16pcs.

Madshrimps (c)


Running the modules at SPD (Serial Presence Detect) resulted in CL4-6-6-16 timings which is not unusual for "overclocking" parts which do not necessarily follow JEDEC guidelines. Another factor effecting SPD is the motherboard BIOS itself, after running five different brands and models of DDR2 including three kits of DR800 I've found the Asus P5WD2-Premium misinterpreted those timings running each part at more lax settings.


Crucial Ballistix BL2KIT 12864AA804
(Official Website)


Madshrimps (c)


Crucial Technology is more familiar to the Western World (USA in particular) based in California and known as the subsidiary of Micron. Today we test their PC2-6400 part in 2048MB size. PC2-6400 is described at crucial.com as Module 2GB kit (1GBx2), 240-pin DIMM.

Madshrimps (c)


IC compliment for Crucial Ballistix 128Meg x 64, UNBUFFERED, NON-ECC, DDR2-800, 2.2V, Memory Timings: CL4-4-4-12. Running Crucial Ballistix at DDR800 under SPD (Serial Presence Detect) on the P5WD2-Premium resulted in 4-4-4-18 timings. In this case timings were much closer which most likely indicative of Asus testing and including these Micron based ICs in their BIOS settings.

Testing

Testing


Intel Test System
CPU Pentium-D 930 Retail 3.0GHz Dual Core 2MB x2 L2 (1.285V ~ 1.388Vcore) Socket-775
Mainboards Asus P5WD2-Premium (BIOS 0606)
Memory Crucial Tracer Ballistix 5300 (2x512MB DC CL4-4-4-12)
OCZ GOLD PC2-6400 (2x512MB)
Graphics Gigabyte GV-NX68T256DH
Power Supply PCPower&Cooling TurboCool 850 SSI
Cooling Alphacool 12V, Black Ice Extreme Double radiator 2x Papst 4412 FG2/GL 120mm
Operating System Windows XP


DDR2 has more then doubled in speed since its inception and with AMD announcing their transition to the standard the evolutionary path will most likely circumvent frequency, with a transmutation to tighter latencies. Corsair leads the effort in this respect with their 5400UL parts producing the tightest latencies available and at 675MHz an admirable performer. Latency as an obstacle was prompting DDR2 advocates to claim high speed would eventually compensate for loose timings. While this may be ideal based on Intel CPU architecture, that architecture is "finally" seeing the pint of diminished returns. We can only hope as AMD influences the market 2.5-2-2-6 timings on DDR2 may be possible.

DDR2 voltage, once thought to incur more errors then increase performance, is finally having the opposite effect, as with Crucial Ballistix PC2-6400, these modules requires 2.20V to run at their "default" settings. This increased requirement for voltage may be indicative of more robust BGA chips with greater tolerances. Super Talent does not specify voltage at their site and their memory ran exceptionally well with default DDR voltages, which for the Asus P5WD2-Premium is approximately 1.80V, Additionally these chips were also able to consume large amounts of voltage which they did to run at 4-4-4-15. I ran both kits using their SPD (Serial Presence Detect) timings as these are the manufacturers pre-programmed settings and are "usually" most stable. I overclocked both kits under SPD and then tightened timings as much as possible with the voltage I had at my disposal. The thumbnails below exemplify all settings where the memory remained 100% stable throughout the benchmarks and these will comprise our test results.

Crucial 800MHz-SPD | Crucial 920MHz-SPD | Crucial 900MHz CL4-5-5-18

Madshrimps (c) Madshrimps (c) Madshrimps (c)

Super Talent 800MHz-SPD | Super Talent 1000MHz-SPD | Super Talent 1000MHz 4-4-4-15

Madshrimps (c) Madshrimps (c) Madshrimps (c)


DDR voltage was a major factor in system stability, especially running games such as Far Cry and DOOM 3. There were numerous times I thought overclocking had corrupted my Config.sys (from the Error codes) yet increasing voltage allowed for stable operation without the need to re-copy or edit the Registry. In fact I may have learned more during these tests than with any other test. Several errors prevented me from booting into Windows and each time memory voltage increases mitigated the errors. I've included another series of thumbnails below exemplifying SiSoftware Sandra 2005.SR3 (Beta) results.

Sandra 2005.SR3

Crucial 800MHz-SPD | Crucial 920MHz-SPD | Crucial 900MHz CL4-5-5-18

Madshrimps (c) Madshrimps (c) Madshrimps (c)

Super Talent 800MHz-SPD | Super Talent 1000MHz-SPD | Super Talent 1000MHz 4-4-4-15

Madshrimps (c) Madshrimps (c) Madshrimps (c)


Benchmark charts >>>

Synthetic Benchmarks

Synthetic Benchmarks


Madshrimps (c)


Sandra 2005.SR3 (Beta) Although I've linked to Sandra 2005.SR2 were using version .SR3. To apply for Beta testing simply visit the SiSoftware site to apply.

Madshrimps (c)


ScienceMark 2.0 Membench is a comprehensive memory benchmark, measuring more specific aspects of system RAM performance.

Madshrimps (c)


SuperPi Mod v1.5 has now been officially taken over by XtremeSys.org. Version 1.5 eliminates the nefarious and questionable attempt to "cheat" attempting to further one's position in the "Challenge."

Madshrimps (c)


PCMark05 begins our FutureMark series of benchmarks and is an "all around" system performance benchmark.

Madshrimps (c)


3DMark2001 (SE) was one of the earliest benchmarks when FutureMark had changed its name to MadOnion. What many don't know is that MadOnion was not the original name of this benchmark factory, but in fact FutureMark is the original and now current name.

Madshrimps (c)


3DMark06 represents FutureMark's latest 3D-benchmark. This benchmark will primarily test the latest graphic card features, although does reflect system RAM performance, as is clearly evident in the overclocked / latency tests.

Madshrimps (c)


Onto Gaming Benchmarks ->

Games / Conclusion

Gaming Benchmarks.


For the average Gamer FPS (Frames Per Second) have been the Holy Grail of graphic card measurement, unfortunately this standard has been relied upon disproportionately almost to the exclusion of any other feature. From an empirical standpoint our visual cortex can "perceive" up to 60FPS (30FPS is incorrect) without digressing into a discussion of optical nerve, suffice to say there's more to graphics then frame rate. The same holds true for a desktop PC, which as cohesive system is so much more then the speed at which it can run. The exception to this rule has been Intel’s Pentium processor and the subsystems designed to support it, including North Bridge chipsets, and of course memory. Intel's now infamous NetBurst architecture took pipeline execution to new "depths." As I’ve said before you need a dual degree in EE and an experienced speleologist at your side to fully delve into the reasoning behind this.

If you’re familiar with pipelining you know this essentially divides instructions into stages, while this allows a processor to do more in a given clock cycle it also bring with it penalties. Anytime a cache miss occurs the pipeline must be flushed which consumes precious time. Fortunately cache misses only occur about 0.5 - 1.0% of the time; however, when this occurs with Prescott's 31-stage pipeline, the number of clock cycles you lose with each miss prediction can add up to 30% of the total processing power for a given clock rate. Ultimately a 3.0GHz Pentium is only as effective as a 2.1GHz Athlon 64. Perhaps this makes it easier to see why Intel would stand behind DDR2 which is surpassing the 1000MHz mark yet still suffers from loose timings. Now 2GB kits are becoming the norm with 4GB kits in the hopper, see any similarities?

DOOM 3 Timedemo accessing DOOM3' FPS benchmark requires pressing the keys Ctrl + Alt + ~ simultaneously. Simply type in >Timedemo demo1 and allow the benchmark to run. This must be run twice to obtain an accurate result.

Madshrimps (c)


HardwareOC Far-Cry Benchmark 1.4.3. offers many demos and settings to choose from. For our purposes we chose the PCGames demo in minimum detail running 800x600 and 1024x768 taking the average score form two instances of each.

Madshrimps (c)



Epilogue

DDR2 is an Intel phenomenon, it’s the ideal memory to perpetuate CPU core architecture which literally "spins it wheels" to get things done. NetBurst's notoriously low IPC cycles and pipeline depth are indicative of this. Perhaps our only reprieve is that Presler marks the end of our bondage to the NetBurst beast. Regardless of what DDR2 advocates say, latency matters although it's almost impossible to prove this given the nature of the system DDR2 is running on. What I've done during this review is to keep CPU speeds as low as possible even when overclocking the memory. It's obvious while there is a difference, it's not as prevalent had this been a DDR based system. To further exemplify how NetBurst is interdependent upon speed to "flush" out (pun intended) its full performance potential, I overclocked our Presler to 4044MHz and overclocked each kit of memory. Crucial to 900MHz and SuperTalent to 1080MHz running SiSoftware Sandra 2005.SR3 Bandwidth benchmark.

Madshrimps (c)


Super Talent 1080MHz-SPD

Madshrimps (c)


Madshrimps (c)


Crucial Ballistix 900MHz-SPD

Madshrimps (c)


Conclusion

Super Talent T800UX2GC4. This 2GB kit of PC2-6400 ran beyond PC2-8000 performance, and held 1000MHz at 4-4-4-15. Perhaps the most promising aspect of the Super Talent kit was its voltage, which isn't specified at the site so the gist of the tests were run at the "Auto" voltage setting (1.80V ~ 1.85V). Only when I tightened up latencies at 1000MHz was there a need for voltage increase although that only required between 1.95V ~ 2.0V. Unfortunately I looked everywhere (Search for Super talent) and couldn't find a single retailer selling their PC2-6400 in 2GB or any size?

Crucial Ballistix BL2KIT12864AA804 also performed admirably running as high as 920MHz at 2.20V under SPD, tightening the latencies to 4-4-4-18 dropped the frequency to 900MHz and required 2.30V for complete stability. Crucial sells their kit at their site for 360USD which is an excellent price for a 2GB kit which offers PC2-7200 performance. Crucial Ballistix is available at through their site at the link above or for UK users Crucial UK 252 incl VAT.

While both kits are exceptional and Super Talent did outperform Crucial, I would have to choose Crucial as the victor in this 2GB basho solely based on availability. While the avid Overclocker would want to find Super Talent and the company does offer a Lifetime warranty, Crucial is well established company with excellent service and their memory can be found almost anywhere. While Super Talent impressed me most, if I can't have it I must defer to Crucial. I would like to thank both companies for submitting their products, both are winners here.

Addendum
Super-Talent has informed their memory is becoming more widely available. Three links were sent; however, at the moment I could find only one store which carried the product we reviewed. The next two shops advertise the kit tested here but I couldn't find it in stock:

OC-wear
SuperTalent 2GB-kits PC2-6400/PC2-8000

Nivulex.nl

Mindfactory.de

Questions/Comments: forum thread
  翻译: