Ultra High Performance Memory Roundup A64 Socket 939

Memory by JNav89GT @ 2005-01-17

Socket 939 has established itself as THE gaming platform. To extract the utmost in performance from a 939 platform, some very good memory is needed. Today we are looking at 8 sets of memory modules which are based on Samsung?s TCCD chips. As many know, the TCCD chips are fast becoming legendary for their ability to clock to crazy levels with modest or even low voltages. Let's take a peak at how our provided samples behave in our test bed shall we!

  • prev
  • next

Setup, Benchmarks and Method

System setup

JNav89GT's Test Setup

Madshrimps (c)

CPU AMD Athlon FX55 S939
Mainboard MSI K8N Neo2 Platinum
Memory Corsair TwinX 1024-4400C25
Video EVga Geforce 6800 Ultra
Cooling Modified Prometeia Mach1 R402a by Chilly1

The Contestants: (All kits are 2x512Mb)

  • Centon PC3200-C2x512PC3200LL
  • Corsair TwinX 1024 PC4400C25
  • Geil Ultra X-GLX1GB3200DC
  • G.Skill PC3200/PC4400 1024-LE
  • G.Skill PC3200/PC4400 1024-LD
  • Mushkin PC3200 Level II V2
  • OCZ PC3200EL Rev2
  • PDP PC3200-PDC1G3200+XBLK
  • PQI PC3200-PQI3200-1024DPU (2.6v)


    I found in testing that completing the following benchmarks offered a reasonable prediction of system stability, plus gives us data on which to discuss potential benefits of increases in memory frequency.

  • Super Pi 1MB Test-Time in seconds(Lower is better)
  • 3DMark 2001SE
  • 3DMark 2003
  • 3DMark 2005
  • PCMark 2004
  • SiSoft Sandra Memory Bandwidth Buffered
  • Everest Home Edition--Memory Read/Latency
  • Doom3 640x480-Default Low detail
  • Far Cry 1.1 - PC Games Hardware Demo HQ: 640x480, 1024x768, 1280x1024
  • Unreal Tournament 2003--Antalus Demo HQ: 1024x768, 1280x1024, 1600x1200

    Testing/ Methodology :

    Testing was initiated by running OCZ Memtest86+ to determine max stable/passable memory frequency. Voltages were varied in an effort to find best combination of voltage for each set. Once maximum frequency was determined, full benchmarking was completed, or attempts at completing were undertaken. With most sets tested, there was often instability in attempting to complete all benchmarks listed at what passed without error in OCZ Memtest86+.

    Memory/HTT frequency was then lowered incrementally until all benchmarks could be passed in succession. Any hints of instability and frequency was again lowered until I was comfortable with results and that the ram was indeed stable at this level, as much as I can be without a hardware memory analyzer.

    Please note, an 80mm fan was installed aimed at the modules as they got very hot during testing and in the past I have noticed improvements of up to 10MHz or more in the ability to clock memory higher.

    CPU speed was unable to be stratified across different speeds for each set, therefore we choose to use an 11X multiplier through all testing. Please keep in mind that increased performance seen at higher frequencies is in part due to higher CPU speeds which ranged from 2750MHz through 3080MHz.
    • prev
    • next