Galaxy 9600 GT 512MB OC Video Card Review

Videocards/VGA Reviews by geoffrey @ 2008-02-26

Galaxy is launching a very different 9600 GT, featuring a custom board design, dual slot cooler, two BIOS and windows flash tool, it is geared toward the enthusiasts out there. This sample comes factory overclocked we compare its performance to a reference 9600 GT video card, as well as a 8800 GT, 8800 GTS and AMD´s pride: HD3870. Read on to find out of this product is the best mainstream card out there!

  • prev
  • next

Test setup & Benchmark methodology

Test setup

Geoffreys' Test Setup
Madshrimps (c)
CPUIntel E6600 @ 3,6GHz
CoolingZalman 9700 LED
MainboardASUS P5E
Memory2x1Gb TEAMGROUP Xtreem 800MHz 4-4-4-10-35-4-10-10-10-2T
Graphics
  • HIS HD3870 512MB IceQ
  • Sparkle 8800GTS 320MB
  • XFX Fatal1ty 8800GTS 320MB
  • Galaxy 9600GT 512MB OC
  • Other
  • FSP Epsilon 900 PSU
  • Maxtor 80Gb PATA HDD
  • Seagate 200GB SATA HDD
  • Antec Nine Hundred housing
  • 20" Dell UltraSharp 2007FP TFT monitor


  • The Intel E6600 was being overclocked to 3,6GHz by changing the FSB from 266MHz to 333MHz and by keeping the multiplier at default (9). This way the CPU would score more or less on par with the highest clocked Wolfdale CPU's.
  • Teamgroup's pair of 1Gb DDR2 sticks were clocked to default settings via the Intel X38 chipset.
  • ForceWare 171.19 beta drivers
  • Catalyst 8.01
  • While Windows Vista is now officially launched, we decided to test with a mature Windows OS (XP SP2).

    Benchmark methodology

    The game benchmarks were completed with the Sparkle GeForce 8800 GTS 320MB, the XFX Fatal1ty Geforce 8800 GTS 320MB and the Galaxy GeForce 9600 GT 512MB OC at stock speeds, while we downclocked the HIS HD3870 512MB IceQ to ATI reference levels. We also downscaled to Galaxy 9600 GT OC to 650/1625/900 in order to simulate NVIDIA's reference 9600 GT.

    All tests were done with a 20” LCD monitor with a maximum resolution of 1600*1200, where 4xAA and 16xAF offered the best in-game graphics we found that AA disabled didn't decrease the eye-candy that much while the frame rate now actually become more acceptable. The in-game settings were always maxed out, we also made sure we were using the full games with the latest patches applied in order get the best up-to-date performance and graphics.

    With our following performance charts you will be informed at all time at what IQ settings were being used. FRAPS was used to measure the FPS during repeated manual run-troughs of a certain part of the games tested, the minimum, maximum and average values were recorded.

  • Colin McRae DIRT
  • S.T.A.L.K.E.R.
  • The Elder Scrolls Oblivion
  • Rainbow Six: Vegas
  • Bioshock
  • Futuremark 3DMark Series
    • prev
    • next
    Comment from geoffrey @ 2008/02/29
    During my test period I did notice strange behavior with the 9600GT's I was testing. Unfortunately, my available time is very limited and I did not have the time to go too much in depth with what was causing problems on my current setup.

    The problem was that one of the standard overclocked cards became instable whenever being put through heavy 3D rendering tasks, downclocking it solved the problem so I guessed we got stuck up with a bugged card (remember the news regarding transient voltage fix on 9600GT's).

    Now, nearly one week after launch date, there seems to be something sneaky going on with NV9600 series. NVIDIA has changed the ways clocks are being build up. Earlier, a 27MHz crystal was used, the clock frequency got multiplied and devided until you get the final clockspeed, 650MHz for example. With GeForce 9600GT the GPU core frequency is based on the PCIe speed, the downside here is that my system always is configured with a 110MHz PCIe clock, making the 9600GT used in our article extra overclocked by another 10%!

    So... let me warn you here: the results obtained in our article are based on a system using 110MHz PCI-Express clock instead of 100MHz, this difference stands for a 10% overclock on the 9600GT core clock and will have its impact on total system performance.

    This being said, we will fix our way of testing in upcoming articles. Those who are still puzzled, don't hesitate to ask around what I was trying to explain here above, but I insist you have a read at our source, thank you techPowerUp!

     

    reply