Nvidia GeForce 8800 GTS graphics card
Cheaper than the GTX, but more bangs per buck?
Review Plonk a Sparkle card based on Nvidia's GeForce 8800 GTS graphics chip next to its bigger brother, the GeForce 8800 GTX - reviewed here  - and you can quickly spot a couple of differences. The GTS is visibly shorter than the GTX - 23cm against 24cm - and it has a single six-pin PCI Express power connector sticking off the end, instead of the pair of side-mounted connectors that you find on the GTX...
Other than that, the GTX and GTS have a great deal in common which isn't surprising as they're both based on the same 'G80' chip. The specifications tell us that the GTS board has a 500MHz core - the GTX runs at 575MHz - 640MB of 1,600MHz memory - against 768MB running at 1,800MHz. The memory controller is 320-bit job rather than 384-bit, and that cuts the memory bandwidth from 86.4GBps to 64GBps.
The 8800 GTX introduced us to unified shaders - or Stream Processors, as Nvidia likes to call them - and the GTS has also been cut back in this department. The GTX has 128 SPs running at 1,350MHz while the GTS has 96 running at 1,200MHz.
The final change comes in the Raster Operation Units (ROPs) which are cut from six in the GTX to five in the GTS. Each ROP can render four pixels so this is the equivalent of cutting back from 24 old-style ROPs to 20 ROPs, which is consistent with the other changes. The GeForce 8800 GTS hasn't been crippled but it is certainly hobbled compared to the GTX.
On the bright side, these changes mean that Nvidia recommends that you install a 450W power supply for a single GTX graphics card while a GTS only requires a 400W unit, and of course the GTS is quite a bit cheaper than the GTX. Now we're not going to sneeze at a £100 saving as it's 20 percent of the price of a GTX, but what is the impact of those changes?
We plugged the Sparkle 8800 GTS into the same quad-core Intel Core 2 Extreme test rig that we used for the 8800 GTX review  with the 96.89 drivers that came in the box. We also used the 97.02 drivers that Nvidia posted when the GeForce 8800 was launched but we were unable to see any significant difference between the two driver versions. Most of the 8800 GTS' 3DMark06 numbers with 97.02, for example, were merely a fraction of a percentage point higher or lower than the 96.89 results.
In 3DMark06, the performance gap between the GTX and GTS was 20 percent at low and medium resolution and climbed to 24 percent at high resolution. Add in 4x Anti Aliasing and 8x Anisotropic Filtering and the gap widened by another couple of percent, but all things considered the 8800 GTS performed very credibly.
In F.E.A.R. the frame rate dropped as the resolution increased, just as you would expect, but the differential between GTX and GTS was more pronounced than it had been in the synthetic 3DMark06 test. At 1,280 x 960, for example, the gap was 17 per cent, but at 1,600 x 1,200 it grew to 29 per cent, and when we cranked the resolution to the maximum 2,048 x 1,536 that our monitor could stand the difference was 35 per cent.
If you intend to use your new Nvidia board to play games at enormous resolutions you're going to need a GTX but otherwise we reckon that a GTS will suit you nicely but on this showing we feel it should be priced at £325 or possibly as low as £299 and not the £375 that's the typical asking price.
It's stretching the English language to its limits to call the £375 Nvidia GeForce 8800 GTS a budget model, however it is undeniably cheaper than the GeForce 8800 GTX and it offers stunning performance using DirectX 9.0c and Windows XP. How it will bear up under DirectX 10 remains to be seen. ®