GF FX performance?

S

sir doris

Regular
Not sure if u guys have seen this...

Looking through the inquirer they have a link to UKGamer which has slides showing fps in UT2K3 Asbestos set to HQ 1280x1024x32 4x FSAA and 8x Aniso with following scores:

GF4 Ti 4600 = 39.3 (fps?)
GF FX = 108.4

from Anands Radion 9700 Pro review:

GF4 Ti 4600 = 37.7
Radion 9700 Pro (16x Aniso) = 71.6

A fair performance boost? Any thoughts? Is this likely final silicon or a calculation?

Thx.
 
It been posted in several threads already - look around.

The numbers were taken from NVIDIA's slides.
 
Anandtech's benchmarks were run on a 2.53 GHz P4, whereas the NV30 was run on a 3 GHz P4. Taking that and the different levels of AF, then that lead doesn't really look particularly impressive...
 
Can't recall where I read it, but the GF FX is running on a P4 3.06Ghz while the Radeon 9700 is running on a P4 2.53Ghz. Not really a fair comparison, also factor in the dif between 16xAF and 8xAF.... etc

Comparisons at this moment aren't really worth talking about.... the drivers for the 9700 may improve greatly by Feb 03, and the scores for the NV30 can only get better too.

Edit: beaten to the punch!
 
If that rather vague post suggests that the AA on the nv30 is decent then I'll be able to stick it back on my possible gfx card for spring list. Pity everything points against it though... :-?

Would you care to expand on 'Really? 8)' ;)
 
Really-as in have you seen the NV30 doing AA?Have you seen it side by side with the R300?Why do people still think that samples are a lot less important when comparet to grid alignment?
 
Testiculus Giganticus said:
Why do people still think that samples are a lot less important when comparet to grid alignment?
Click to expand...

The most disturbing cases of jaggies appear on nearly horizontal and nearly vertical edges, which just happen to be the absolute worst case for ordered grid AA. In these particular cases, rotated/jittered grid with N samples give results only barely inferior to ordered grid with N^2 samples, as both give about N color gradations.
 
Testiculus Giganticus said:
In this particular case, mainly number of.
Click to expand...

I'd whisper "gamma correction" as an additional factor for people who are concerned about the quality and think about prior OG sampling quality. It remains the thing I found most exciting about the R300 AA, and with the nv30 having it I think the results will be better than many people seem to expect.

I understand they are concerned about gradient spread at certain angles, but I tend to think there will be enough in combination with gamma correction to achieve good AA. It is of course dependent on what 6XS and 8x actually are, but I don't see them not offering some improvement over prior modes.

BTW, did I mention gamma correction? :-? Am I the only one who thinks this is a very important feature for AA?
 
Maverick said:
Anandtech's benchmarks were run on a 2.53 GHz P4, whereas the NV30 was run on a 3 GHz P4. Taking that and the different levels of AF, then that lead doesn't really look particularly impressive...
Click to expand...
Yes it does, 40 frames more is pretty darn impressive. that is more then 30 percent.
 
arjan de lumens said:
Testiculus Giganticus said:
Why do people still think that samples are a lot less important when comparet to grid alignment?
Click to expand...

The most disturbing cases of jaggies appear on nearly horizontal and nearly vertical edges, which just happen to be the absolute worst case for ordered grid AA. In these particular cases, rotated/jittered grid with N samples give results only barely inferior to ordered grid with N^2 samples, as both give about N color gradations.
Click to expand...
Mathematically corect(somewhat), but in this case, the reasoning is a bit off-8x has a rotated GRID.
 
arjan de lumens said:
Testiculus Giganticus said:
Why do people still think that samples are a lot less important when comparet to grid alignment?
Click to expand...

The most disturbing cases of jaggies appear on nearly horizontal and nearly vertical edges, which just happen to be the absolute worst case for ordered grid AA. In these particular cases, rotated/jittered grid with N samples give results only barely inferior to ordered grid with N^2 samples, as both give about N color gradations.
Click to expand...

N-1 colour gradations :)
 
Yes it does, 40 frames more is pretty darn impressive. that is more then 30 percent.
Click to expand...

??

Let's run those benches on a Radeon 9700 Pro on a P4 3.5 Ghz and turn off aniso alltogether, then we'll see what's more impressive...

Read: the comparison is so apples to oranges we can't draw much of anything from it at all.
 
Yes, it's impressive if the R300 score was not CPU limited, and the Geforce FX score had the same image quality.

I can't see why the FX would not have a non-ordered-grid sampling pattern for 4x multisampling. OGSS is obviously much easier to implement than RGSS, but I don't know why OGMS would be much easier to implement than rotated or jittered grid multisampling. Are we absolutely sure that the 4x MS pattern on the FX is ordered grid? Maybe it makes Z compression easier or something...
 
Mathematically corect(somewhat), but in this case, the reasoning is a bit off-8x has a rotated GRID.
Click to expand...

Not according to the interview with nVidia. 8X contains supersampling elements, but is ordered grid. Of course, the nVidia rep could be mistaken on that...
 
You must log in or register to reply here.

Similar threads

C
Historical GPU FlOPs performance
2 3 4
Replies
66
Views
53K
Arun
A
T
Toms look at the Geforce FX
2
Replies
29
Views
5K
Derek Smart [3000AD]
D
Back
Top