View Full Version : AA with NO FRAME HIT on Nv30!!!!!
sancheuz
11-27-02, 04:26 PM
Ok guys, i might be mistaking, but is it possible that what nvidia quoted in this article:
http://www.nvidia.com/view.asp?IO=feature_intellisample
What's more, the GeForce FX's innovative new architecture includes an advanced and completely transparent form of lossless depth Z-buffer and color compression technology. The result? Essentially all modes of antialiasing are available at all resolutions without any performance hit. Greatly improved image quality, with no drop in frame rate!
Does this mean that we could play games full AA and not get a performance hit?!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
If so then that alone is reason enough to shout in cheer!
PS(just in case) no ati vs nvd flame war please.
I simply don't believe in that especially at higher resolutions (>1280x960) But who knows? We shall see soon hehe
sancheuz
11-27-02, 04:45 PM
Tell you the truth, I really dont believe that either, but it sure looks nice doesnt it. There is a slight chance it could be true. And that slight chance is more then what i hoped for. I mean AA with no frame hit. Wow
If that turns out to be true...the "debate" will be over immediately....
:D
Actually, as surprising as this may seem...
I don't think that's true; but that's because it isn't phrased correctly.
What wouldn't surprise me at all would be that, once you activate basic antialiasing ( 2X probably ) , you get a small performance hit. But then, the other modes performance hit doesn't grow as linearly as on the NV20.
On the NV20/NV25, at 4X AA, cost is twice as high as with 2X AA. With 6X AA, it's 50% higher than with 4X AA. And so on. Yes, benchmarks proved this ( but this only in cases where memory bandwidth was the bottleneck, of course )
Why? Color compression. Since 90% of pixels are the same in all samples, the cost is nearly exactly the same with 2X as with 8X for those samples ( beside increased memory usage for evantual samples, of course )
Now, "no performance hit" is obviously marketing talk. But, if Color Compression delivers its promises, a 25-30% performance hit for 8xS isn't out of the question...
In case you don't know, current GF4s delivers a 70% performance hit...
But then again, that would be utopic & unlikely.
Uttar
jbirney
11-27-02, 05:04 PM
Originally posted by pelly
If that turns out to be true...the "debate" will be over immediately....
:D
Uhmmm what debate? which is better? Or which is faster? Sure if they lose 0 fps going from 1600x1200 to 1600x1200 with x8 FSAA which is impossible for a long time, then that proves they are faster. but as we seen time and time again faster != better
If it really has no hit I guess nvidia will lock 8x fsaa as default in their drivers.:rolleyes: :rolleyes:
If it really has no hit I guess nvidia will lock 8x fsaa as default in their drivers.
No, there are times when you dont want AA.
It is possible to get 8xAA without any performance hit whatsoever. But its unlikely to occur with this architecture, and this many transistors.
IT would be nice if it was like 10% hit regardless of how much AA you put on. But anyway, we shall see soon enough.
gstanford
11-27-02, 07:08 PM
If you read what David Kirk has said in the Extreme Tech & Beyond3D interviews, you'd know that on NV30 4x FSAA will be free. The NV30 was designed that way.
This doesn't mean every FSAA mode will free, just 4A FSAA.
Also, the 48gig/s effective bandwidth nVidia quotes includes 4X FSAA.
Greg
-=DVS=-
11-27-02, 07:20 PM
Would be very nice if you can have 4xAA as a defoult without Frame rate drop :eek:
Okay so 4x is free but 2x is not? :confused:
I'm assuming you meant > 4x :D
thcdru2k
11-27-02, 07:53 PM
yeah that doesn't make sense..but than, who'd want to use 2xaa if 4xaa was free :)
gstanford
11-27-02, 08:01 PM
Of course 2x FSAA will be free - after all if you know the 4X FSAA data then you also know the 2X FSAA data (you just throw some away, or do 4X FSAA anyway).
Another thing to consider: since 4X FSAA is free, 8X FSAA will have only a 2X FSAA cost and 6X would be around quincunx cost.
It also possible that nVidia will support 16X FSAA for Pro applications (movie rendering) since it will only cost what 4X FSAA did on the GF4 4600.
Greg
thcdru2k
11-27-02, 08:12 PM
2x aa is faster than quincunx though. so you mean 8x fass will have quincunx cost and 6x aa will have 2x aa cost. if this is true, thats pretty damn good.
Bigus Dickus
11-28-02, 01:40 AM
Originally posted by Nutty
It is possible to get 8xAA without any performance hit whatsoever.
Can you do that without a bandwidth penalty as well?
As far as the question of "free AA" on the NV30, at any resolution...
There ain't no such thing as a free lunch. ;)
Lezmaka
11-28-02, 01:45 AM
In my distorted mind, there is a way to get have no hit for any mode of AA the nv30 supports. It's just that it would never, ever, ever, ever, ever, happen. They intentionally slow non-AA rendering down to the point where turning on AA has no effect.
Told ya my mind was distorted.
There ain't no such thing as a free lunch.
Yup.....a fact that's as consistent as gravity....
I wish it were otherwise...but there is ALWAYS a hidden cost...
;)
Err, wondering something...
Could it be possible that the DDR2 4 prefetches be the reason for which 4X AA is (nearly) free? That they'd use those prefetches one way or another, and that if they use 6X AA for example, they can't prefetch everything so it's significantly slower?
That would be interesting - but, as always, unlikely :)
Uttar
That'd be a cool little trick......great idea man!
:D
thcdru2k
11-28-02, 02:09 AM
i don't believe it still..if they had free aa, wouldn't you think they'd be hyping it alot?
FREE 4X AA
StealthHawk
11-28-02, 02:46 AM
Originally posted by thcdru2k
i don't believe it still..if they had free aa, wouldn't you think they'd be hyping it alot?
FREE 4X AA
yeah, i think we would have been hearing such boasts up the wazoo by now. there really is no reason why they would keep such a boon a secret, except to give ATI less time to respond to the "true" power of gfFX.
But, ATI already knows all about the gfFX shaders, they now know the clocks, so they should know R9700Pro will be beaten performance-wise, and should be working on the r9700 successor with all that in mind. not to mention nvidia already gave away the color compression, and deigned to call the bandwidth saving techniques in NV30 only LMA 2+ instead of LMA 3.
Lets say for arguments sake that 4x fsaa is free wont you need to turn on aniso(which is not free)to get rid of the blur?Even if 4x fsaa if free is it the same edge quality as ATI's 4x fsaa or the same as 2x which is also free?
I'm not trying to slam nvidia just mentioning a few things I've heard on the web.
StealthHawk
11-28-02, 04:26 AM
Originally posted by Fotis
Lets say for arguments sake that 4x fsaa is free wont you need to turn on aniso(which is not free)to get rid of the blur?Even if 4x fsaa if free is it the same edge quality as ATI's 4x fsaa or the same as 2x which is also free?
I'm not trying to slam nvidia just mentioning a few things I've heard on the web.
2x? why would it be the same as 2x? nvidia cards already use RGMS for 2x FSAA, and OGMS 4x on nvidia cards looks better than that...
nutball
11-28-02, 04:27 AM
Well they don't say what the performance actually was before it wasn't hit.
"Runs UT2003 as fast with 8x FSAA turned on as it does with it turned off*"
"*on our ass-kicking Celeron 300MHz test rig"
Originally posted by StealthHawk
2x? why would it be the same as 2x? nvidia cards already use RGMS for 2x FSAA, and OGMS 4x on nvidia cards looks better than that...
Ok, but isn't Ati's 4x better than nvidia's from an IQ point of view?
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.