Performance
The voltage on the GeForce 7600 GS chip was increased to 1.72V which helped achieve an impressive frequency of 765MHz. This is 89% above the default frequency! The chip was cooled by a water cooling system that consisted of a Revoltec water-block (it is a CPU water-block but I had remade it so that it could be installed on GPUs), an external 600lph Eheim pump, and an Airplex EVO 360 radiator with three low-speed Aerocool Turbine 3000 fans. The liquid cooling system worked only for the graphics card during the test.

To generally improve the thermal conditions of the card, a 120mm fan was set to blow at the reverse side of its PCB.
The GPU worked at frequencies higher than 765MHz, but with image artifacts. It seemed that a minor voltage increase would solve the issue, but I had those artifacts even at 765MHz after I had increased the core voltage to 1.8V and then to 1.85V. So, the chip on my sample of the graphics card reached its frequency limit under the given conditions. I could also try to increase the frequency of some of the GPU subunits by reducing the frequency of the geometric subunit, but that would have been beyond the scope of this review. By the way, the maximum GPU frequency at which the card could work, although with image artifacts, was 798MHz!

The memory wasn’t near that good: a frequency growth of only 10MHz after a voltage increase from 1.8V to 1.95V. All further attempts led to visual artifacts. On the other hand, 1GHz is good enough for Infineon’s DDR2 memory which has never been really good for overclocking.
The XFX GeForce 7600 GT was benchmarked at the default frequencies only as a reference point for the modded 7600 GS.

The performance of the GeForce 7600 GS card grows up along with its frequency in this popular synthetic benchmark. It is important that the GPU and memory frequencies grow up together: a 100MHz increase in both frequencies provides a considerable performance boost whereas the tremendous GPU frequency leap, by 250MHz more, yields a smaller performance gain.

Another version of Futuremark’s benchmark produces the same picture as the previous test. The graphics card lacks memory bandwidth to transform the colossal overclocking potential of the G73 chip into an appropriate performance boost.


As you can see, the real gaming applications agree with the results of the synthetic benchmarks. I didn’t reach the level of the full-featured GeForce 7600 GT as I had anticipated before the tests. I had been even sure that a volt-modded and well-overclocked GeForce 7600 GS would deliver more performance than the senior model. But it turns out that you can’t get far with DDR2 memory – it provides too low a bandwidth for the graphics chip to show its best.



