Graphics performance
We've narrowed our performance testing to 3DMark06's Shader Model 3.0 tests and F.E.A.R.'s built-in performance benchmark. Both were run at 1920x1440—the highest resolution supported by our test monitor—with 4X antialiasing and 16X anisotropic filtering.

Obviously, the GeForce 8800 GTX cards have a considerable lead over the GTS models. What's more interesting to see, however, is how much the higher clocked models manage to outpace their competition. The EVGA ACS³ makes more of its clock speed advantage over the rest of the GTX field than the Foxconn does over other GTS cards. That's notable because it's the Foxconn card that actually enjoys a greater clock speed jump, at least percentage-wise, over its stock-clocked compatriots.

Overclocking
The handy automatic overclocking utility built into Nvidia's graphics drivers and then relocated to its nTune system utility doesn't seem to be working properly with the GeForce 8800 series, so we had to kick it old-school with manual slider manipulation and loads of trial-and-error testing. Each of our overclocked configurations had to loop successfully through three iterations of 3DMark's Shader Model 3.0 tests at 1920x1440 with 4X antialiasing and 16X aniso, and then endure ten minutes of Oblivion at the same graphics settings.

We were able to hit the following core and memory clock speeds with each card (doubling the memory clock speed gives you the effective memory clock):

  • BFG GeForce 8800 GTS — 653MHz core, 962MHz memory
  • EVGA GeForce 8800 GTX ACS3 — 643MHz core, 1048MHz memory
  • Foxconn FV-N88SMBD2-ONOC — 650MHz core, 1060MHz memory
  • MSI NX8800GTX — 629MHz core, 1053MHz memory
  • OCZ GeForce 8800 GTX — 622MHz core, 1047MHz memory
  • PNY XLR8 GeForce 8800 GTS — 653MHz core, 1059MHz memory
  • XFX GeForce 8800 GTX — 659MHz core, 1060MHz memory
The XFX GeForce 8800 GTX hit managed a higher core clock speed than any other card by 9MHz, and it also shared the memory clock crown with Foxconn's FV-N88SMBD2-ONOC, although only by 1MHz over PNY's GTS. Note that with only a couple of exceptions, all the cards hit about the same core and memory clock speeds, regardless of their GTS or GTX designation.

As is always the case, overclocking success is never guaranteed and is often just the luck of the draw. Because Nvidia takes care of the manufacturing for all of these cards, there probably isn't much the add-in board partners can do to ensure greater overclocking success.

Pushing these cards to their limits changes how they stack up a little, with the factory "overclocked" cards no longer sitting in the lead. These results suggests the overclocking potential of stock-clocked cards isn't being eroded by cherry-picking for faster models.