A note on percentages for you guys here.

[Doomtrooper]

No, ATI has had decent drivers since the Radeon, not the best but have been improving greatly.
ATI's drivers in the Rage days were not very good (and thats when most people had ATI cards and base their arguement from), nor was Nvidia and their Riva 128 sets..time to stop living in the past.

90% of the people claiming 'bad drivers' are people that never had a ATI card period, a simple browse on NV fan forums shows it clearly' I'd never buy a ATI card, look at what the reviews say...look at Rage3D forums problems' which is ludicrous.
People that have problems with their card post on a forum with 30,000 members, so posts per/member will be alot higher.
People that post issues are very minor when you look at the installed base, those people are enjoying their cards.

The driver things is getting old...no IHV drivers are perfect..Nvidia hasn't had a major architectural change in two years and that certainly helps when tweaking drivers and NV is very good with developers to hook up with contracts to ensure maximum stability on their hardware..that is changing.

Example when a developer is a offical partner with a IHV like Epic:

http://www.imgmagazine.com/news/story.php?ArticleID=808

As part of their efforts to work more closely together, NVIDIA will provide Epic with early versions of new hardware and drivers, as well as extensive testing of those
drivers with existing Unreal Engine products and new unreleased versions of the engine. To facilitate this, the agreement calls for regular technical reviews and
exchanges between the two companies, so that each one has a clear understanding of the inner workings of the other's software and hardware technology.

Three games came out of that:

UT 2003
Unreal 2
Americas Army

 

[RussSchultz]

And the athlon is better than the P4 clock per clock, but that isn't the fastest processor around, now is it?

Thus Joe's comment was valid to see what part of that advatage is coming from architecture and what part from brute speed. Some of us here find that interesting....

Yes, I can read. I know what Joe was saying. It is interesting (and lookie, I even stated that), but it still doesn't change the fact that clock for clock really doesn't mean much in the end, as the P4vs.Athlon comparison has shown.

Again, not that I'm saying that the GFFX is equivalent to the P4, just that measuring clock for clock doesn't really mean much when extrapolating into the future. Who knows, maybe a 4 pipe NV3x chip can run at 500mhz with no problems with cooling(since it will have "half" the die area, hence half the heat generated) and beat the tar out of a RV350 that can only run at 350 or 400. Or maybe not, since the RV350 is supposed to be at 0.13u and will likely run faster than 350/400.

 

[Joe DeFuria]

And the athlon is better than the P4 clock per clock, but that isn't the fastest processor around, now is it?

Point?

A year or so ago, when the P4 was introduced, it was given the relative thumbs down.

Not by me incidentally.

Now? It seems like its shown its merit by heading upwards of 3ghz.

Right. And since day one, the architecture was touted as being built for faster clocks, not instructions per clock. Is there any indication that's the case for GeForceFX? I would say that most people would agree that given the actual MHz of each chip and the fab process they are on, they are both more or less geared for similar clock rates on a given process, or in fact, the advantage might go to ATI.

For that matter, is there any indication that the GeForceFX is faster than the Radeon 9700 Pro, irrespective of clock? Only in specific situations, not across the board.

Not that that's any indication at all as to what the GFFX architecture can do, but still: clock to clock comparisons are interesting, but don't mean much in the end.

My point was, it means much more to me than Chalnoths "assesment of FX architecture promise", which is to base performance numbers in "cases selectively excluding 1600x1200 with 4X FSAA".

You read way to much into my posts, Russ.

 

[Sherlock]

Not that this has anything to do with "A note on percentages for you guys here" but history and common sense have shown that tech engineers design a chip to produce as little heat as possible. The more work that is done by the processing unit the more heat is generated ala athlon and higher end P4's. Also architecture flow determines heat production ala processor architecture limitations/ overclocked/over volted processor. So which situation applies here? Is the architecture limited or is this processor just clock for clock doing more than the r300? Someone please show me how NV30 is doing more clock for clock than r300 because I can't see it based on the evidence we currently have. So that leaves us with architecture limiting the flow of either instructions (overclocked) or voltage (over-volted); hence the dustbuster solution.

 

[Althornin]

Anyway, if you are concerned about clock-for-clock performance, the best way to compare would be against a Radeon 9500 Pro (with 8 pipes and a 128-bit interface) clocked at the same speed. With this comparison, at a resolution lower than 16x12x4, I think the FX will win (most of the time). Would be nice if somebody could post some benches...

This is worthless if the core is bandwidth limited. This test would show nothing about clock for clock performance. Its like saying "give the GFFX a 256bit bus and see how they compare.
The bus width is PART of the chip - so is the relative ram speed.

 

[KimB]

So for the smarter folks out there...what is the amount of MB you need for a MSAA method at 16x12 @ 32 bit colors?

If downsampling is done at scanout:
87.9MB

If downsampling is done at buffer swap:
65.9MB

 

[KimB]

My point was, it means much more to me than Chalnoths "assesment of FX architecture promise", which is to base performance numbers in "cases selectively excluding 1600x1200 with 4X FSAA".

You read way to much into my posts, Russ.

I don't think that that resolution/FSAA setting should ever be selectively excluded. It just should never be the only benchmark run on the FX, since it obviously has problems there (that is, benches at that setting currently don't appear to translate well to benchmarks at other settings).