NVIDIA received a lot of criticism for the arithmetic efficiency of the former
NV3x product line. In that sense NV had to catch-up in relative terms with
the competition in that department and all indications so far point to the
direction that they’ve succeeded with NV40.
SM3.0 support can so far only be tested with Far Cry’s (not yet officially
released) 1.2 patch, which for the time being is just a performance optimisation.
It’s my understanding though that those optimisations could have been
easily expanded from Cry Tek to NV3x/R4xx based accelerators too, in order
to correct the engine’s own shortcomings. Sadly enough there’s
no other SM3.0 capable application/benchmark available right now and thus I’ll
withhold for the time being any relative comment concerning it’s usefulness;
debates about inflection/deflection points when it comes to having either SM2.0+
or SM3.0 support at this point of time have been endless in public forums and
the pro/contra arguments are fairly known. What I can say though is that there’s
no such thing as supporting a feature too early for a developer or ISV for
future projects and I have severe doubts that any IHV will be able to escape
SM3.0 (+) support in the foreseeable future.
Overall my experience with the 6800 was more than just pleasant. One quite
annoying spot was in Asbestos/UT2k4 were I could notice some weird moiré patterns
on floor textures, for which I can only guess that the application might pick
up the wrong LOD value (DefaultTexMipBias=-0.500000). Setting it to 1.0 might
eliminate the moiré pattern, yet blurs the scenery way too much. It
occurred with both former and current ForceWare drivers, I reinstalled the
game just to make sure that there’s nothing awkward stuck in there, but
to no avail. It doesn’t matter whether I use the in game AF/trilinear
or the driver control panel settings, whether any kind of optimization on or
off the problem persists. I can’t say whether it’s the applications
or driver’s fault, yet it’s highly annoying and I hope it will
get fixed soon enough. A shot to illustrate what it looks like:
I couldn’t notice in any other game/application anything similar.
Other than that the driver offers a very extensive palette of various settings
where the user can enable/disable or even optimize features, as his heart desires.
I couldn’t find one setting or change in settings so far that didn’t
work, as it should; nor did I have to result to any system reboots in order
to enable a specific setting.
The list of games/demos I’ve tried so far is a bit longer than mentioned
in this review, but (with the exception of PowerVR’s SM3.0 based Cloth
demo, which refused to even initialize) I didn’t have any compatibility
problems at all, other than the ones mentioned.
I can most certainly welcome the Supersampling support via the hybrid AA modes
(despite their limited usability), as the high quality texture filtering setting
(as I’ve said before) in the driver control panel. Arithmetic efficiency
does no longer leave any headroom for complaints and the stencil OP performance
I did try to overclock the card via the driver’s self detecting frequency
feature, yet testing even with a just 355MHz clock speed in a subtropical climate
during one of the extreme heat waves and without any additional cooling aids
for the GPU, turned out to be a rather bad idea. It’s not that it wasn’t
impossible to game or benchmark with higher core/ram frequencies, it’s
just that a couple of crashes are already enough for me to not include any
results to this review. Default frequencies didn’t give me any trouble
whatsoever and that with a broken Air-Con and room temperatures up to 32C.
The GeForce 6800 is very worthy lower end representative of NVIDIA’s
new high end line of products and it’s not very hard to guess how much
more performance a user will get with a 6800GT or even 6800Ultra, especially
in very high resolutions with AA/AF enabled.
There aren’t any real comparisons in this review and thus any recommendations
on whether to upgrade or not would be out of place. For me the user should
purchase what he thinks will cover his needs most and that irrelevant of brand
or IHV. This project was merely an attempt to collect some of my impressions,
nothing more or nothing less. Corrections, remarks, question etc are of course
Last I would like to thank Loewe for hosting this review, the time and efforts
he made for the last refinements of it, Demirug’s and Zeckensack’s
precious contributions/observations and of course quite a few others that would
like to remain anonymous.