A month ago, an Asian site reported that NVIDIA was intending to launch a new card under the name GeForce 9600 GSO, based on a chip quite a bit less powerful than the original model. This was refuted the following day by NVIDIA France, but the report did indeed turn out to be accurate!
ZOTAC is the first to announce such cards, in three versions: 512 MB, 1 GB and 512 MB overclocked (AMP! Edition). The first two are clocked at 650/900 MHz, against 700/1000 MHz for the last.
Until now the GeForce 9600 GSO was based on a G92, a chip with 754 million transistors engraved at 65nm, a chip which is also used in the GeForce 8800/9800. It was of course castrated, as it was in the GeForce 8800 GS.
On the GeForce 9800 GTX, then, the G92 is clocked at 675 MHz with 128 calculation units, 64 texturing/filtering units and 16 ROPs, all accompanied by a 1.1 GHz 256 bit memory. On the 9600 GSO, the G92 is at 550 MHz, with only 3/4 of the units active, the same as for the 192 bit memory bus.
Now, the new GeForce 9600 GSO will be based on a G94, a chip with 505 million transistors engraved at 65nm and even at 55nm in a more recent revision. On the 9600 GT, this chip is clocked at 650 MHz and has 64 calculation units, 32 texturing/filtering units and 16 ROPs, with a 256 bit 900 MHz memory on top.
On the new GeForce 9600 GSO, there is still a 256 bit memory bus, a good thing in comparison with the current model then, with a bandwidth over 9%. Only 3/4 will however be activated … such that in the end the old 9600 GSO has a 69% higher
processing, filtering and texturing power!
There can be little doubt then that the new GeForce 9600 GSO performs less well than the old model. True, NVIDIA could not match the Radeon HD 4650/4670 with a chip such as the G92 and needed a less costly solution that nevertheless offered a significant gain in performance on the GeForce 9500 GT, which is not competitive. But keeping the same name as a better performing old card is simply not honest.