Testbed and Methods

To check out the result of the modifications described in the previous sections, I used an open testbed configured like follows:

• Processor: AMD Sempron 3000+ (Palermo) Socket 939
• Mainboard: ASRock 939Dual-SATA2 (ULi1695+ULi1567)
• CPU cooler: Arctic Cooling Alpine 64
• Memory: 2x512MB TwinMOS (Winbond UTT CH-5, -AA4T)
• Power supply: 600W be quiet! Dark Power ATX2.2
• Hard disk drive: 120GB ATA-100 Hitachi (7200 rpm)
• Optical drive: NEC ND-3520A DVD-RW

The CPU was clocked at 2800MHz (350x8) at 1.6V voltage. The memory worked as DDR-133 at a clock rate of 233MHz with 2-2-2-5-1T timings and a voltage of 3.4V.

The testbed ran Windows XP SP1. I overclocked the graphics card by means of RivaTuner 2.0 RC16.

The graphics card driver was ForceWare 91.31. I left the default performance/quality ratio and didn’t use anisotropic filtering or FSAA.

The following benchmarks were used:

• 3DMark 2003 version 3.6.0
• 3DMark 2005 build 1.2.0
• Serious Sam 2 version 2.066 (Direct3D); Medium graphics quality settings; Greendale demo
• F.E.A.R. version 1.02 (Direct3D); Medium graphics quality settings; standard demo

Graphics cards of only one class are included into the tests, so I ran all the benchmarks at 1024x768 resolution. The settings in the synthetic benchmarks were left default.

The Gainward Bliss 7600 GS 256MB was tested in the following modes:

• Default frequencies: 450/500 (1000) MHz
• Overclocking without modifications: 513/816 (1632) MHz
• Max frequencies at overclocking with a GPU volt-mod, without firmware modification: 753/816 (1632) MHz
• Frequencies after selecting the negative delta of -85MHz for the geometric subunit: 753/838/838 (geometric/shader/ROP) /816 (1632) MHz

I also tested an XFX GeForce 7600 GT 256MB (it is pre-overclocked by the manufacturer to 570/725 (1450) MHz) and an EVGA e-GeForce 7600 GS 256MB (400/400 (800) MHz, DDR2 memory) for the comparison’s sake. This will give us a clear understanding of what can be achieved with the Gainward card.