If youâ€™ve been following our graphics card coverage recently, youâ€™ll notice that in both our 7600 GT and 7900 GTX reviews we used reference sample cards supplied to us by nVidia. Itâ€™s now been long enough since the launch that quite a few manufacturers are in the position to supply retail samples. We collected together five 7900 GTs all with varying clock speeds and put them to the test. As this is our first look at the 7900 GT, weâ€™ve also done SLI testing using our reference cards and included those results.
Unlike the 7800 GT which has four less pixel pipelines than the 7800 GTX, the 7900 GT is architecturally identical to the 7900 GTX, with 24 pixel pipelines, eight vertex shaders and 16 pixel output engines. The main differences are that of clock speeds and a reduction in frame buffer to only 256MB. The 7900 GTX runs with a 650MHz core and 800MHz (1,600MHz effective) memory, while the 7900 GT runs at only 450MHz core and 660MHz (1320MHz effective) memory. Much like the rest of nVidiaâ€™s cards, they donâ€™t support simultaneous HDR and FSAA. If this feature means a lot to you, then you might want to consider an X1900XTX
The up-shot of this is that the cooler is dramatically smaller - in fact itâ€™s the same as the cooler on the 7600 GT. The only noticeable difference between the 7600 GT and 7900 GT is the addition of external power and relevant circuitry. The 7900 GTX really does dwarf its siblings.
Considering the 7900 GT is around Â£150 less than the GTX and is fairly similar on paper, this looks to be a chip worth considering, so letâ€™s take a closer look at our selection.
What is interesting about all these graphics cards is that they all use the same 1.4ns Samsung memory. Doing a little maths, that equates to a frequency of 715MHz. Yet strangely the default clock speed of the 7900 GT is only 660MHz. Even the 7600 GT operates at 700MHz, using identical Samsung modules. So why have nVidia chosen to underclock the memory on these cards? To be honest, who cares â€“ the point is there is room for quite a bit of overclocking.
All the cards above use the same reference cooler and none of them got particularly hot to the touch. Weâ€™ve been told it takes around 3 months to design a new cooler and get it to market, so itâ€™ll be a while before we start to see custom installed coolers â€“ quite probably passive. We have a range in core speeds all the way from the reference 450MHz to 560MHz and as high as 590MHz thanks to some overclocking. So we can really see how this core scales.
All our testing was performed on an Asus A8N32-SLI using an Athlon 64 FX60, 2GBs of CMX1024-3500LLPRO and a Seagate Barracuda ST340083A8. All of the cards were tested using the official 84.21 ForceWare drivers, except for the SLI/Reference card results which were tested using 84.17 ForceWare, which offers almost identical performance.
Using our proprietary automated benchmarking suite, aptly dubbed â€œSpodeMark 3Dâ€, I ran Call of Duty 2, Counter Strike: Source, Quake 4, Battlefield 2 and 3DMark 06. Bar 3DMark06, these all run using our in-house pre-recorded timedemos in the most intense sections of each game I could find. Each setting is run three times and the average is taken, for reproducible and accurate results.
I ran each game test at 1,280 x 1,024, 1,600 x 1,200, 1,920 x 1,440 and 2,048 x 1,536 each at 0x FSAA with trilinear filtering, 2x FSAA with 4x AF and 4x FSAA with 8x AF.
Overclocking was performed using PowerStrip as the â€˜CoolBitsâ€™ portion of the ForceWare drivers was capped so it wouldnâ€™t let us reach the overclocking potential of each card. SpodeMark was then re-run while overclocked at 1,280 x 1,024 with 4x FSAA and 8x AF and at 2,048 x 1,536 with 4x FSAA and 8x AF.
Please bear in mind that because of the nature of overclocking, there's no guarantee you will achieve the same results as I did because there is a huge element of luck involved. However, considering these are all based on the reference design that most manufacturers use, this does give a fairly good indication of what you might expect. The lowest overclock out of these cards was 560MHz on the core and 885MHz on the memory. Both of these are still significantly over the stock speeds and demonstrate how good nVidiaâ€™s yields must be. If one was so inclined, with a little bit of extra cooling and a boost in voltage, there is no doubt in my mind that a GT could be modified to GTX speeds.